# How Anthropic’s product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)

## Metadata
- Channel: Lenny's Podcast
- Duration: 86 min
- YouTube: https://www.youtube.com/watch?v=PplmzlgE0kg

## Transcript

**[00:00] Speaker A:** I think it is very hard to be the right amount of AGI pilled. It's very easy to build the product for the super AGI strong model.  
我觉得要对 AGI 保持恰到好处的信念程度是很难的。为超强 AGI 模型打造产品其实很容易。  
**[00:07] Speaker A:** The hard thing is figuring out for the current model, how do you elicit the maximum capability?  
难的是针对当前的模型,如何最大限度地激发它的能力。  
**[00:13] Speaker B:** I've never seen anything like the pace you folks at Anthropic are shipping at.  
我从没见过像你们 Anthropic 这样的发布速度。  
**[00:17] Speaker A:** We want to remove every single barrier to shipping things. The timelines for a lot of our product features have gone down from 6 months to 1 month and sometimes to even one day.  
我们希望移除所有阻碍产品发布的障碍。很多产品功能的开发周期已经从 6 个月缩短到 1 个月,有时甚至只需要一天。  
**[00:27] Speaker B:** You're interviewing hundreds of PMs and you just keep feeling like they're approaching it very incorrectly.  
你面试了几百个 PM,总觉得他们的思路很不对。  
**[00:32] Speaker A:** The PM role is changing a lot. It's changing really quickly. The thing that is extremely important for building AI native products is iterating so quickly, figuring out a way for you to actually launch features every single week.  
PM 这个角色正在发生很大的变化,而且变化非常快。打造 AI 原生产品最重要的是快速迭代,想办法做到每周都能发布新功能。  
**[00:44] Speaker B:** What do you think are the emerging skills PMs need to develop?  
你认为 PM 需要培养哪些新兴技能?  
**[00:48] Speaker A:** It comes back to product taste.  
归根结底还是产品品味。  
**[00:50] Speaker A:** Becomes much cheaper to write, the thing that becomes more valuable is deciding what to write.  
当写代码变得更便宜时,更有价值的就是决定写什么。  
**[00:56] Speaker A:** Today my guest is Kat Woo, head of product for Claude Code and Co-work at Anthropic.  
今天的嘉宾是 Kat Woo,她是 Anthropic 公司 Claude Code 和 Co-work 的产品负责人。  
**[01:00] Speaker A:** Kat is at the center of everything that is changing in AI and product and building.  
Kat 正处于 AI、产品和开发领域所有变革的中心。  
**[01:05] Speaker A:** And she and her team are building the product that is most changing the way that we all build our products.  
她和她的团队正在打造一款产品,这款产品正在深刻改变我们所有人构建产品的方式。  
**[01:11] Speaker A:** She is so full of insights and wisdom and lessons.  
她有太多的洞察、智慧和经验可以分享。  
**[01:14] Speaker A:** This is an episode you cannot miss.  
这期节目你绝对不能错过。  
**[01:17] Speaker A:** Before we get into it, don't forget to check out lennyspodcast.com for an insane set of deals available exclusively to Lenny's newsletter subscribers.  
在正式开始之前,别忘了访问 lennyspodcast.com,那里有专门为 Lenny 新闻订阅用户提供的超值优惠。  
**[01:26] Speaker A:** With that, I bring you Kat Woo.  
接下来,让我们欢迎 Kat Woo。  
**[01:33] Speaker A:** Kat, welcome to the podcast.  
Kat,欢迎来到播客。  
**[01:35] Speaker B:** Thanks for having me.  
谢谢邀请。  
**[01:37] Speaker A:** I have so many questions. I'm so excited to have you on this podcast.  
我有好多问题想问。很高兴你能来做客。  
**[01:41] Speaker A:** I want to start with giving people an understanding of your role alongside  
我想先让大家了解一下你和 Boris 的分工。  
**[01:45] Speaker A:** Boris, everybody knows Boris. His episode is the number one most popular episode on this podcast. No pressure.  
Boris 大家都知道。他那期节目是这个播客最受欢迎的一期。没有压力哈。  
**[01:51] Speaker A:** He created Claude Code. He leads the team, ships a bazillion PRs a day from his phone.  
他创建了 Claude Code,领导团队,每天用手机就能提交无数个 PR。  
**[01:57] Speaker A:** I don't even know what the number is anymore. I think people don't give you enough credit for the success that Claude Code has had and co-work and all the things you all are building.  
我都不知道现在具体数字是多少了。我觉得大家对你在 Claude Code、Co-work 以及你们团队所有成就中的贡献认可得还不够。  
**[02:06] Speaker A:** Help us understand your role on the team, how you work with Boris, how you split responsibilities, just like what does the PM role look like on the Claude Code team?  
能跟我们讲讲你在团队中的角色吗?你和 Boris 如何协作,如何分工,PM 在 Claude Code 团队是什么样的?  
**[02:16] Speaker B:** I feel very lucky to work with Boris. He's been an amazing thought partner. He's our tech lead.  
能和 Boris 共事我感到很幸运。他是个很棒的思想伙伴,也是我们的技术负责人。  
**[02:19] Speaker B:** He's very much the product visionary and he is great at setting like this is what the product needs to be in three months, six months from now. This is what the AGI pill version of the product is.  
他非常擅长产品愿景规划,能够清晰地设定产品在三个月、六个月后应该是什么样子,以及最终 AGI 版本的产品形态。  
**[02:34] Speaker B:** And a lot of my role is figuring out okay what is the path from where we are today  
而我的主要职责是弄清楚如何从当前状态  
**[02:37] Speaker A:** To like that vision 3 to 6 months from now. And I spend more of my time on the cross functional, so making sure that our marketing team, sales team, finance, capacity, etc. are like bought in on the plan and that we're all rowing in the same direction and that once the feature is ready that there aren't any blockers to shipping it.  
一步步实现三到六个月后的愿景。我把更多时间花在跨职能协调上,确保市场、销售、财务、产能等团队都认同这个计划,大家朝着同一个方向努力,并且在功能准备好后不会有任何阻碍发布的因素。  
**[02:58] Speaker A:** I think in many ways it works well because we kind of like mind meld, but it is actually like remarkably blurry of a line. Like I think we're like 80% mind melded and then there's like this 20% of things that like maybe I care a lot more about than him. So like I'll drive those and then like 20% where he cares a lot more than me and he just like drives those.  
我觉得这种合作方式很有效,因为我们在很多方面想法一致,但实际上分工界限确实很模糊。大概 80% 的事情我们想法一致,剩下 20% 可能是我比他更在意的事情,那我就来推动;另外 20% 是他比我更在意的,他就自己去推动。  
**[03:20] Speaker B:** This episode is brought to you by our season's presenting sponsor WorkOS. What do OpenAI, Anthropic, Cursor, Vercel, Replit, Sierra, Clay and hundreds of other winning companies all have in common? They are all powered by WorkOS. If you're building a product  
本期节目由本季冠名赞助商 WorkOS 为您呈现。OpenAI、Anthropic、Cursor、Vercel、Replit、Sierra、Clay 以及数百家成功公司有什么共同点?它们都在使用 WorkOS。如果你在开发  
**[03:36] Speaker A:** For the enterprise, you've felt the pain of integrating single sign-on, SCIM, RBAC, audit logs, and other features required by large companies.  
面向企业的产品,你一定体会过集成单点登录、SCIM、RBAC、审计日志等大公司要求的功能有多痛苦。  
**[03:42] Speaker A:** WorkOS turns those deal blockers into drop-in APIs with a modern developer platform built specifically for B2B SaaS.  
WorkOS 把这些交易障碍变成了即插即用的 API,提供专为 B2B SaaS 打造的现代开发者平台。  
**[03:52] Speaker A:** Literally every startup that I'm an investor in that starts to expand upmarket ends up working with WorkOS.  
我投资的每一家开始向上拓展市场的初创公司,最后都会选择 WorkOS。  
**[03:58] Speaker A:** And that's because they are the best.  
因为他们就是最好的。  
**[04:00] Speaker A:** Whether you are a seed-stage startup trying to land your first enterprise customer or a unicorn expanding globally, WorkOS is the fastest path to becoming enterprise-ready and unblocking growth.  
无论你是种子轮创业公司想要拿下第一个企业客户,还是独角兽公司在全球扩张,WorkOS 都是实现企业级就绪、解锁增长的最快路径。  
**[04:10] Speaker A:** It's essentially Stripe for enterprise features.  
它本质上就是企业功能领域的 Stripe。  
**[04:12] Speaker A:** Visit workos.com to get started or just hit up their Slack where they have actual engineers waiting to answer your questions.  
访问 workos.com 开始使用,或者直接去他们的 Slack,那里有真正的工程师随时解答你的问题。  
**[04:18] Speaker A:** WorkOS allows you to build faster with delightful APIs, comprehensive docs, and a smooth developer experience.  
WorkOS 让你通过友好的 API、全面的文档和流畅的开发者体验更快地构建产品。  
**[04:24] Speaker A:** Go to workos.com to make your app enterprise-ready today.  
访问 workos.com,让你的应用今天就具备企业级能力。  
**[04:30] Speaker A:** Something that you shared actually before we started recording is the fact that you're interviewing hundreds of PMs all the time.  
你在录制前分享的一个情况是,你一直在面试大量的产品经理。  
**[04:36] Speaker A:** Like if I had a nickel every time someone asked me for an intro to someone at Anthropic to go work at Anthropic as a PM, I'd have 30 billion in ARR.  
如果每次有人让我介绍 Anthropic 的人帮他们去那里做产品经理,我能拿到五美分的话,我现在就有 300 亿年收入了。  
**[04:45] Speaker A:** It's just like the number one place people want to go work at. So I can only imagine how many PMs you're interviewing.  
这就是大家最想去工作的地方。所以我能想象你要面试多少产品经理。  
**[04:49] Speaker A:** You told me that you're just seeing people doing it wrong, the way they're approaching what they think it takes to be a successful AI PM.  
你告诉我,你看到很多人做错了,他们对成为成功的 AI 产品经理需要什么的理解有偏差。  
**[04:57] Speaker A:** Talk about what you're seeing and what people need to understand about what it takes to be successful these days.  
说说你观察到了什么,以及大家需要理解当下成功需要具备什么。  
**[05:03] Speaker B:** I think before AI, technology shifts were a lot slower. So you could plan on the 6 to 12 month time horizons.  
我觉得在 AI 之前,技术变革的速度要慢得多。所以你可以按照 6 到 12 个月的时间跨度来规划。  
**[05:12] Speaker B:** And because you were shipping features at a bit of a slower rate, there was a lot more emphasis on coordinating with all  
而且因为功能发布的节奏比较慢,需要花很多精力去协调所有  
**[05:18] Speaker A:** The other partner teams to make sure that they're shipping features that unblock your features because code at that time was very expensive to make.  
合作团队,确保他们发布的功能能为你的功能扫清障碍,因为那时候写代码的成本非常高。  
**[05:26] Speaker A:** I think now with AI and with how much that has accelerated engineering and with how quickly the model capabilities are improving, the timelines for a lot of our product features have gone down from 6 months to one month and sometimes to one week or even one day.  
但现在有了 AI,工程开发速度大大加快,模型能力提升也很快,很多产品功能的开发周期已经从 6 个月缩短到一个月,有时候是一周甚至一天。  
**[05:42] Speaker A:** And with that, we actually need to make sure that products ship quite quickly.  
在这种情况下,我们实际上需要确保产品能够非常快速地发布。  
**[05:48] Speaker A:** And what that means is as a PM, there should be less emphasis on making sure that you're aligning your multi-quarter roadmaps with your partner teams and more emphasis on okay, how can we figure out the fastest way to get something out the door?  
这意味着作为产品经理,应该少花精力去对齐多季度路线图,而是更多地思考:我们怎样才能最快地把东西推出去?  
**[06:05] Speaker A:** How can we figure out how to make a concept corner of our product suite where an engineer has an idea or a PM has an idea and by the end of the week we are able to get it into our users' hands.  
我们怎样才能在产品套件中建立一个概念试验区,让工程师或产品经理有了想法后,在一周内就能把它交到用户手中?  
**[06:19] Speaker A:** I think the PMs who do the best on AI native products are the ones who can figure out how can I shorten the time from having this idea to actually getting the product in the hands of users and help define what are the most important tasks that need to work out of the box for my product.  
我认为在 AI 原生产品上做得最好的产品经理,是那些能够缩短从产生想法到真正把产品交付用户手中这段时间的人,并且能帮助明确哪些是产品必须开箱即用的最重要任务。  
**[06:36] Speaker B:** So what I love about this is what you're saying is just like people haven't grasped how fast they need to move and how much of the job now is just moving, is helping the team move fast.  
我特别喜欢你说的这点——人们还没有意识到他们需要以多快的速度行动,以及现在这份工作有多大程度上就是关于推进,帮助团队快速前进。  
**[06:47] Speaker B:** What helps do that? What do you do? What does your PM team do to help them move this fast other than have access to the most advanced models?  
那么如何做到这一点呢?你们具体怎么做?除了使用最先进的模型之外,你的产品经理团队还做了什么来帮助他们如此快速地推进?  
**[06:56] Speaker A:** I think the first thing is to set clear goals because LLMs are so general that actually creates a lot of ambiguity in who we're building for, what problems we're trying to solve, what the top use cases are.  
我认为首先要设定清晰的目标,因为大语言模型太通用了,这实际上会在我们为谁构建、要解决什么问题、最重要的用例是什么这些方面造成很多模糊性。  
**[07:09] Speaker A:** And so I think a great PM is able to say, okay, our key user is professional developers, the main  
所以我认为优秀的产品经理能够明确说出,好的,我们的核心用户是专业开发者,主要的——  
**[07:16] Speaker A:** The problem that we want to solve for this feature is maybe there's like too many permission prompts and people are feeling fatigue.  
我们想通过这个功能解决的问题可能是权限提示太多,用户感到疲劳。  
**[07:22] Speaker A:** And like the use case is we want professional developers at enterprises to safely get to zero permission prompts.  
而用例就是我们希望企业中的专业开发者能够安全地实现零权限提示。  
**[07:32] Speaker A:** And that actually sets a pretty clear goal because it rules out a lot of potential approaches for reducing permission prompts so that people can get a lot more done with one prompt.  
这实际上设定了一个相当清晰的目标,因为它排除了很多减少权限提示的潜在方法,这样人们就能用一个提示完成更多工作。  
**[07:42] Speaker A:** And then I think the second thing that's very important is figuring out some repeatable process for getting these features shipped.  
然后我认为第二件非常重要的事情是找到一个可重复的流程来发布这些功能。  
**[07:49] Speaker A:** So for Cloudcode what we do is we actually ship almost all of our features in research preview.  
对于 Cloudcode,我们的做法是几乎所有功能都以研究预览版的形式发布。  
**[07:56] Speaker A:** We clearly brand this when we ship something so that users know that this is an early product.  
我们在发布时会清楚地标注这一点,让用户知道这是一个早期产品。  
**[08:01] Speaker A:** This is just an idea, this is just something that we're trying to get feedback on and iterating on, and that this might not be supported forever.  
这只是一个想法,只是我们想要获取反馈并迭代的东西,而且这可能不会永久支持。  
**[08:08] Speaker A:** And what this does is it  
这样做的效果是——  
**[08:10] Speaker A:** It reduces our commitment for shipping something. We can just get something out in a week or two.  
它降低了我们发布某个功能的承诺。我们可以在一两周内就把东西推出去。  
**[08:17] Speaker A:** And then the third thing that a PM should do is help create the framework for the team so that they know when to pull in cross-functional partners and what those cross-functional partners' expectations are.  
产品经理应该做的第三件事是帮助团队建立框架,让他们知道何时引入跨职能合作伙伴,以及这些跨职能合作伙伴的期望是什么。  
**[08:26] Speaker A:** So for example, we have a really tight process between engineering, marketing, and docs.  
比如,我们在工程、市场营销和文档之间有一个非常紧密的流程。  
**[08:31] Speaker A:** So when engineers have a feature that they feel is ready and that we've dogfooded internally, they post it in our Evergreen launch room.  
当工程师有一个他们认为已经准备好并且我们内部已经试用过的功能时,他们会把它发布在我们的 Evergreen 发布频道里。  
**[08:40] Speaker A:** And then Sarah, who leads our docs, and Alex, who leads PMM, and Tar and Lydia on DevRel, just jump in and can turn around the marketing announcement for it the very next day.  
然后负责文档的 Sarah、负责产品市场营销的 Alex,以及开发者关系团队的 Tar 和 Lydia 就会介入,第二天就能完成营销公告。  
**[08:51] Speaker A:** And because we have this really tight process, it lowers the friction for any engineer to ship something, and PM is the role that should be setting this up.  
正因为我们有这个非常紧密的流程,它降低了任何工程师发布功能的摩擦,而产品经理就是应该建立这个流程的角色。  
**[08:59] Speaker B:** How do PRDs fit into this? The fact that you said that goals are a really—  
产品需求文档在这里面是怎样的定位?你刚才说目标是非常——  
**[09:03] Speaker A:** Important part, just like being aligned on what does success look like? Who is this for? Who's this not for? Are you writing PRDs? Is it just like a couple bullet points? How has that evolved in the world of ABM?  
重要的部分,就像在成功是什么样子、这是为谁做的、不是为谁做的这些问题上达成一致?你们会写产品需求文档吗?还是只是几个要点?在 AI 原生产品的世界里这是如何演变的?  
**[09:12] Speaker B:** So there's two things that we do. One is we have very rigorous metrics and we do metrics readouts with the entire team every week. The goal of this is to make sure that everyone deeply understands all the facets of our business, what our key goals are, how they're trending, and what drives them.  
我们做两件事。一是我们有非常严格的指标体系,每周与整个团队进行指标解读。目的是确保每个人都深入理解我们业务的各个方面、我们的关键目标是什么、它们的趋势如何,以及是什么在驱动它们。  
**[09:29] Speaker B:** The second thing that we do is we have this list of team principles. And this includes who our key users are, why those are our key users. And the reason that we articulate all of this is so that everybody on the team feels like they understand how our business works. They understand what's important to us and what we're willing to trade off. And it lets people make decisions by themselves without feeling like they're  
我们做的第二件事是有一份团队原则清单。这包括我们的核心用户是谁、为什么他们是我们的核心用户。我们阐明所有这些的原因是让团队中的每个人都觉得他们理解我们的业务如何运作、什么对我们重要、我们愿意做出什么权衡。这让人们可以自己做决定,而不会觉得被——  
**[09:52] Speaker A:** Blocked on PM or any other stakeholder.  
产品经理或其他利益相关者阻碍了。  
**[09:55] Speaker B:** I love how so much of this is like, okay, we still need PMs in the future. There's so much talk of like why do we need PMs? We're just going to ship and build. We need engineers.  
我很喜欢这个讨论的一点是,未来我们仍然需要产品经理。现在有很多声音在说,我们为什么还需要 PM?我们直接发布和构建就行了,我们需要的是工程师。  
**[10:03] Speaker A:** Oh, we actually do PRDs sometimes. So I think for features that are particularly ambiguous, it does help to write out just a one-pager on what the goals are, what the delightful use cases are, what the failure modes currently are that we need to fix. And there are occasionally some projects, especially things that require heavy infrastructure that do take many months. And for those situations, we do write PRDs still.  
其实我们有时候还是会写 PRD 的。我觉得对于那些特别模糊不清的功能,写一份一页纸的文档还是有帮助的,列出目标是什么、理想的使用场景是什么、当前需要修复的失败模式是什么。偶尔也会有一些项目,尤其是那些需要大量基础设施建设、要花好几个月的项目,在这些情况下我们还是会写 PRD。  
**[10:29] Speaker B:** I want to drill a little bit further into just how you're able to move so fast. I've never seen anything like the pace folks at Anthropic are shipping at. Like someone made this calendar of launches across Anthropic and it was literally every day there was like a  
我想再深入了解一下你们是如何做到这么快的速度的。我从来没见过像 Anthropic 这样的发布节奏。有人做了一个 Anthropic 发布日历,上面显示几乎每天都有一个  
**[10:45] Speaker A:** major feature or product. So, one question people had online is you guys just launched this—not launched but built this incredible model Mythos that is still in preview because it's so powerful people are a little afraid of what it can do.  
重大功能或产品发布。所以,网上有人问,你们刚刚推出了——不是推出而是构建了这个令人惊叹的模型 Mythos,它现在还处于预览阶段,因为它太强大了,人们对它能做什么有点担心。  
**[10:58] Speaker A:** Have you guys been using this? Is this part of the reason you've been able to move so fast?  
你们一直在使用这个模型吗?这是你们能够快速推进的部分原因吗?  
**[11:03] Speaker B:** We've been moving pretty fast for several quarters now. So, I think it's not fully Mythos. Mythos is an incredibly powerful model.  
我们已经保持了好几个季度的快速节奏了。所以我认为这不完全是 Mythos 的功劳。Mythos 确实是一个非常强大的模型。  
**[11:12] Speaker B:** But we do use the models internally and I think this has increased our rate of shipping a little bit, but I don't think it explains the bulk of the increase.  
但我们确实在内部使用这些模型,我认为这在一定程度上提高了我们的发布速度,但我不认为这能解释大部分的提升。  
**[11:20] Speaker B:** I think a lot of it is the process and the expectation on the team. So we're very low on process. We want to remove every single barrier to shipping things. We want to make sure every single person on the team feels empowered to take their idea from just an idea to out in the world in less than a week, sometimes.  
我认为很大程度上是流程和团队的期望。我们的流程非常精简。我们想要移除每一个阻碍发布的障碍。我们希望确保团队中的每个人都有能力把自己的想法从构思阶段推进到实际发布,有时候不到一周就能完成。  
**[11:40] Speaker A:** Even in a day.  
甚至一天就能完成。  
**[11:41] Speaker B:** Cool. Oh man, what an advantage to have the best model and also be building product. That's so cool.  
太酷了。天哪,拥有最好的模型同时又在做产品开发,这是多大的优势啊。太棒了。  
**[11:46] Speaker A:** We are very lucky to be able to work with the Frontier models.  
我们非常幸运能够使用这些前沿模型。  
**[11:49] Speaker B:** Oh my god, what an awesome advantage. Just like build a thing and then use it and then accelerate faster. It's so interesting. There's a couple like these other side things I want to just kind of go on these like side quests on this conversation. There's so much happening with Anthropic and I just I'm so curious to get your insight. One is a week ago or so the whole source code of Claude code leaked. Somebody got it out there. I think it was a mistake someone made. Is there anything you can comment there just like what happened? What went wrong? What should people know?  
天哪,这真是个了不起的优势。就是构建一个东西,然后使用它,然后加速得更快。太有意思了。这次对话中还有几个我想探讨的话题,就像是一些支线任务。Anthropic 发生了太多事情,我真的很想听听你的见解。一个是大约一周前,Claude 的整个源代码泄露了。有人把它弄出去了。我想这是有人犯的一个错误。关于这件事你能说点什么吗?发生了什么?哪里出了问题?人们应该知道些什么?  
**[12:15] Speaker A:** So we immediately looked into this when we saw it. We realized that this was the result of human error. There was...  
所以我们看到这个情况后立即进行了调查。我们发现这是人为失误造成的。有一个  
**[12:23] Speaker A:** A human working with Claude to write a PR. This was just an update to how we release our packages and it actually went through two layers of human review.  
工作人员在使用 Claude 帮助编写 PR。这只是对我们发布包的方式的一次更新,而且它实际上经过了两层人工审核。  
**[12:33] Speaker A:** And so this was a result of human error and we've hardened our processes to make sure that it doesn't happen in the future.  
所以这是人为失误的结果,我们已经加强了流程以确保将来不会再发生这种情况。  
**[12:39] Speaker B:** Is this person still at Anthropic? Are they doing alright?  
这个人还在 Anthropic 吗?他们还好吗?  
**[12:42] Speaker A:** Yes, yes. It's a process failure and the most important thing is to just learn from it and to add more safeguards so that doesn't happen again.  
是的,还在。这是一个流程失败,最重要的是从中吸取教训,并增加更多的保障措施,这样就不会再发生了。  
**[12:50] Speaker A:** And so that's what we've been focused on and most of those have shipped.  
这就是我们一直在关注的,而且大部分措施已经实施了。  
**[12:54] Speaker B:** Okay. Another question I had is OpenClaude. So recently there's been this move to keep people from using Claude subscription with their OpenClaude.  
好的。我还有一个问题是关于 OpenClaude 的。最近有一个动作是阻止人们在 OpenClaude 中使用 Claude 订阅。  
**[13:06] Speaker B:** People got really upset that they're confused why this is happening. It feels like there's harm caused to the open source community.  
人们对此非常不满,他们不明白为什么会这样。感觉这对开源社区造成了伤害。  
**[13:13] Speaker B:** What do people—what do  
人们需要——人们需要  
**[13:15] Speaker A:** People need to understand about kind of what went into this decision?  
了解这个决定背后的考虑是什么?  
**[13:18] Speaker B:** So, we've been seeing a lot of demand for Claude and we've been working very hard to both scale our infrastructure and also to make our models more token efficient so that you can get more usage out of it.  
我们看到对 Claude 的需求非常大,我们一直在努力扩展基础设施,同时也在让我们的模型更加 token 高效,这样你就能获得更多的使用量。  
**[13:28] Speaker B:** It wasn't designed for third party products which have different usage patterns than our first party ones.  
它不是为第三方产品设计的,而第三方产品的使用模式与我们自己的产品不同。  
**[13:36] Speaker B:** We spent a bunch of time trying to figure out what is the most seamless transition that we can offer.  
我们花了很多时间试图找出我们能提供的最无缝的过渡方案。  
**[13:44] Speaker B:** And so I was very happy to be able to say that everyone gets some credits alongside their subscription.  
所以我很高兴能够宣布,每个订阅用户都会获得一些额度。  
**[13:49] Speaker B:** But yeah, we did have to make the hard decision that we needed to prioritize our first party products and our API.  
但确实,我们不得不做出一个艰难的决定,就是需要优先保障我们自己的第一方产品和 API。  
**[13:55] Speaker B:** And so this is a decision that resulted from that.  
所以这个决定就是基于那个考虑做出的。  
**[14:00] Speaker A:** Yeah, this like to me it makes so much sense.  
是的,对我来说这完全说得通。  
**[14:02] Speaker A:** Like you guys are subsidizing this usage at like 200 bucks a month and there's like it's like basically unlimited use of this and like...  
你们每月 200 美元的订阅其实是在补贴这些使用成本,而且基本上是无限使用,然后……  
**[14:11] Speaker A:** I think people don't understand businesses are trying to make money. We're trying to be profitable here. We can't just give away compute when it's so in demand.  
我觉得人们不理解企业是要赚钱的。我们要实现盈利。我们不能在算力需求这么大的时候白白送出去。  
**[14:17] Speaker A:** So I get it. Coming back to the PM team, what does the PM team look like at Anthropic? How many PMs are there? How are they kind of organized?  
所以我理解。回到 PM 团队的话题,Anthropic 的 PM 团队是什么样的?有多少 PM?他们是怎么组织的?  
**[14:26] Speaker B:** Yeah, so we have a few PM teams. I think we're maybe around 30 or 40 PMs right now.  
是这样,我们有几个 PM 团队。我想现在大概有 30 到 40 个 PM 左右。  
**[14:31] Speaker B:** So we have the research PM team who Diane leads, and this team is responsible for understanding all of the feedback from our customers for our models and then feeding that to the research team to act on it, and they also shepherd the model launch.  
我们有研究 PM 团队,由 Diane 领导,这个团队负责了解客户对我们模型的所有反馈,然后把这些反馈传递给研究团队去处理,他们还负责推进模型发布。  
**[14:49] Speaker B:** There's the Cloud developer platform team that maintains the APIs that Claude is built on top of, and they also release things like managed agents, which is a way for you to build your agents and we can host it on your behalf.  
还有 Cloud 开发者平台团队,维护 Claude 所基于的 API,他们还发布像托管代理这样的功能,让你可以构建自己的代理,我们代为托管。  
**[15:01] Speaker B:** And then there's Claude that works on both  
然后是 Claude 团队,负责  
**[15:06] Speaker A:** Cloud Code and the Cowork core products. There's Enterprise that helps make Cloud Code and Cowork easier to adopt for all of our enterprise customers.  
Cloud Code 和 Cowork 核心产品。还有企业团队,帮助我们所有企业客户更容易地采用 Cloud Code 和 Cowork。  
**[15:14] Speaker A:** And so this is everything from like cost controls, RBAC, security controls, and just making sure that these enterprises feel very confident and comfortable using our tools.  
这包括成本控制、基于角色的访问控制、安全控制等所有方面,确保这些企业对使用我们的工具感到非常有信心和放心。  
**[15:25] Speaker A:** And then we also have our growth team that is responsible for growing across our entire product suite.  
然后我们还有增长团队,负责整个产品套件的增长。  
**[15:32] Speaker A:** So we work very closely with them on Cloud Code and Cowork growth, and I know they also work with our other teams on CDP growth—so growth of people who use the Cloud API.  
所以我们在 Cloud Code 和 Cowork 的增长上与他们密切合作,我知道他们也和其他团队合作推动 CDP 的增长——也就是使用 Cloud API 的用户增长。  
**[15:42] Speaker B:** So speaking of growth, Amol was just on the podcast. He had this really interesting insight that most people haven't been sharing.  
说到增长,Amol 刚上过播客。他有一个很有意思的见解,大多数人都没有分享过。  
**[15:48] Speaker B:** There's always this sense that we need fewer PMs in the future. What's the—why do we need PMs? Engineers can just ship.  
总有一种感觉是未来我们需要更少的 PM。为什么需要 PM 呢?工程师可以直接发布产品。  
**[15:54] Speaker B:** His take is that because engineers are moving so fast, PMs and—  
他的观点是,因为工程师推进得太快了,PM 和——  
**[15:59] Speaker A:** Designers are squeezed. There's less time to stay on top of everything that is happening. There's a feature shipping every day.  
设计师被挤压了。没有足够的时间跟上正在发生的一切。每天都有功能在发布。  
**[16:04] Speaker A:** So his take is he needs more PMs because it's hard to keep up.  
所以他的观点是他需要更多 PM,因为很难跟上节奏。  
**[16:08] Speaker A:** What's your take there? Do you feel like there will be an increase in hiring of PMs? What do you think is going on with the PM profession long term?  
你怎么看?你觉得 PM 的招聘会增加吗?你认为 PM 这个职业长期来看会怎样?  
**[16:15] Speaker B:** I think all of the roles are merging. PMs are doing some engineering work, engineers are doing PM work, designers are PMing and also landing code.  
我觉得所有角色都在融合。PM 在做一些工程工作,工程师在做 PM 的工作,设计师在做 PM 的事也在提交代码。  
**[16:26] Speaker B:** You can either hire a lot more engineers who have great product taste or you can keep your engineering hiring the same and hire a lot more PMs to help guide some of their work.  
你可以选择招聘更多有出色产品品味的工程师,或者保持工程师招聘规模不变,招聘更多 PM 来帮助指导他们的一部分工作。  
**[16:36] Speaker B:** On our team we're pretty focused on hiring engineers with great product taste. This way we can reduce the amount of overhead for shipping any product.  
在我们团队,我们更专注于招聘有出色产品品味的工程师。这样我们可以减少发布任何产品的管理开销。  
**[16:47] Speaker B:** There are many engineers on our team who are fully able to end-to-end go from seeing user feedback on Twitter through to shipping.  
我们团队有很多工程师完全能够端到端地从 Twitter 上看到用户反馈一直做到发布产品。  
**[16:57] Speaker A:** A product at the end of the week with almost no product involvement. And this I think is actually like the most efficient way to ship something.  
到周末就能交付一个产品,而且几乎不需要产品经理介入。我认为这其实是最高效的交付方式。  
**[17:06] Speaker A:** So I think like engineer and PM are kind of overlapping and you will get a lot of benefit from having more of either.  
所以我觉得工程师和产品经理的职能是有重叠的,无论增加哪一方的人手都能获得很多好处。  
**[17:14] Speaker A:** I think product taste is still a very rare skill to have and we'll pretty much hire anyone who we feel has demonstrated this strongly.  
我认为产品品味仍然是一项非常稀缺的技能,基本上只要有人能强有力地展现出这种能力,我们就会录用。  
**[17:25] Speaker B:** And your background was in engineering, right?  
你的背景是工程师出身,对吧?  
**[17:27] Speaker A:** Yeah, I was an engineer for many years. I was then a VC very briefly before joining Anthropic and actually almost all the PMs on our team have either been engineers or ship code here on Claude code and so that's one of the things that I think helps build trust with the team and also just enables us to move a lot faster and then actually our designers also have been front-end engineers before.  
对,我做了很多年工程师。之后短暂做过风投,然后加入了 Anthropic。实际上我们团队几乎所有产品经理要么曾是工程师,要么在这里为 Claude 写代码。我认为这是帮助我们与团队建立信任的因素之一,也让我们能够行动得更快。而且我们的设计师之前也都做过前端工程师。  
**[17:54] Speaker B:** Wow because that's the big  
哇,因为这是个大问题——  
**[17:56] Speaker A:** Question like there's definitely this merging that's happening, the Venn diagrams you're combining.  
确实正在发生这种融合,各个职能的维恩图正在合并。  
**[17:59] Speaker A:** I think the big question for a lot of people is if you're coming from engineering or product or design, which of those core skills is going to be most valuable?  
我认为很多人面临的大问题是:如果你来自工程、产品或设计背景,这些核心技能中哪一个会最有价值?  
**[18:07] Speaker A:** I could see at Anthropic and on Claude, code engineering is very valuable.  
我能看到在 Anthropic 和 Claude 团队,编程工程能力是非常有价值的。  
**[18:10] Speaker A:** I'm curious if at other companies, if you have a design background, becoming a PM is more valuable or just a PM in general.  
我很好奇在其他公司,如果你有设计背景,成为产品经理是否更有价值,或者说单纯的产品经理角色是否更有价值。  
**[18:16] Speaker B:** I still think it comes back to product taste.  
我仍然认为归根结底还是产品品味。  
**[18:18] Speaker B:** Like as code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write.  
随着写代码的成本大幅降低,更有价值的是决定写什么。  
**[18:23] Speaker B:** Like what is the right UX for this feature?  
比如这个功能的正确用户体验是什么?  
**[18:25] Speaker B:** What is the most delightful way that a user can experience it?  
用户体验它的最令人愉悦的方式是什么?  
**[18:28] Speaker B:** We get tens of thousands of GitHub issues asking for every single thing under the sun, and it takes a lot of care and taste to figure out, okay, which of these is worth building and what is the right way to build it, and I think that that skill is what matters.  
我们收到成千上万个 GitHub issue,要求各种各样的功能,需要投入大量精力和品味来判断:好,这些里面哪些值得做,以及正确的做法是什么。我认为这种能力才是最重要的。  
**[18:50] Speaker A:** Certainly can come from any background, but I think that's the most important thing.  
这种能力当然可以来自任何背景,但我认为这是最重要的。  
**[18:53] Speaker A:** I think the reason why an engineering background is particularly useful, at least for the next few months, is if you have an engineering background, you have a better sense for how hard something should be.  
我认为工程背景特别有用的原因,至少在接下来几个月内,是如果你有工程背景,你会更清楚某件事应该有多难。  
**[19:05] Speaker A:** And that's often a factor in what you choose to build. So like if something is very easy to build, then maybe instead of debating it, you just spend an hour doing it.  
而这往往是你选择做什么的一个因素。比如如果某件事很容易做,那与其争论不如直接花一小时把它做出来。  
**[19:13] Speaker A:** But if something is harder to build and you know that upfront, you know that, okay, this will just cost a lot more for our team to get this out the door.  
但如果某件事更难做,而你提前知道这一点,你就知道:好,这会让我们团队付出更大的代价才能完成。  
**[19:23] Speaker A:** So it helps a bit with the prioritization.  
所以这对优先级排序有一定帮助。  
**[19:27] Speaker B:** You said in the next, for the next few months—is that just like because the models will get so good potentially in the next few months?  
你说「在接下来几个月内」——是因为模型可能在接下来几个月会变得非常好?  
**[19:34] Speaker B:** You may not even need to know that as much?  
你可能甚至不太需要了解这些了?  
**[19:35] Speaker A:** I think the valued skill sets do change quite frequently, and so it's really hard to predict more than a few months out.  
我认为有价值的技能组合确实变化得相当频繁,所以很难预测几个月以后的情况。  
**[19:45] Speaker A:** It's less a commentary on what shift I think will happen and more of a commentary that I think large shifts will happen.  
这与其说是在评论我认为会发生什么转变,不如说是在评论我认为会发生重大转变。  
**[19:53] Speaker B:** So you're not saying that's when Mythos comes out and we'll change everything and that we don't need to know anything about engineering.  
所以你不是说那时候 Mythos 会发布,然后改变一切,我们就不需要懂任何工程知识了。  
**[19:58] Speaker A:** No, I'm just saying that every few months it seems like there's a—  
不是,我只是说每隔几个月似乎就会有一次——  
**[20:02] Speaker B:** Yeah.  
是的。  
**[20:03] Speaker A:** There's a large increase in coding capability which then changes what other roles are valuable.  
编程能力会有一次大幅提升,然后就会改变其他哪些角色是有价值的。  
**[20:08] Speaker A:** I think the most important thing is to be able to have this first principles thinking where you can figure out how the tech landscape is changing, what the team really needs from you, and to jump in and fix that hole, because I think the work is becoming more amorphous, which means that a great PM is able to understand what all the gaps are, to figure out what the highest priority—  
我认为最重要的是要有第一性原理思维,能够弄清楚技术格局是如何变化的,团队真正需要你做什么,然后跳进去填补那个空缺,因为我觉得工作正在变得更加不定形,这意味着一个优秀的产品经理能够理解所有的缺口在哪里,找出最高优先级的——  
**[20:43] Speaker A:** ones are and then to just like figure out okay how do I learn that skill set or what is like the skill set that I have that I can like apply to this challenge.  
是哪些,然后就想办法,好吧我怎么学习那个技能,或者我有什么技能可以应用到这个挑战上。  
**[20:51] Speaker A:** So I think the current environment values people who are able to wear a lot of hats, are able to swap them, and are like very low ego about what work they do to help the team move faster.  
所以我认为当前的环境看重那些能够身兼多职、能够灵活切换角色、并且对自己做什么工作来帮助团队加速没有自我执念的人。  
**[21:06] Speaker B:** I love this answer. There's this question I've been asking people in your shoes, folks that are kind of at the bleeding edge of what AI is capable of and building with the latest tools, which is just like where will human brains continue to be useful and necessary for a while until we get to super intelligence.  
我很喜欢这个回答。有个问题我一直在问处于你这个位置的人,就是那些站在AI能力最前沿、用最新工具构建产品的人,就是人类大脑在哪些方面会继续有用、继续必要,至少在我们达到超级智能之前的一段时间里。  
**[21:21] Speaker B:** What I'm hearing here is essentially picking the things to work on, knowing where the market's going and figuring out what to prioritize essentially.  
我从你这里听到的本质上是选择要做什么事情,知道市场走向,以及弄清楚优先级。  
**[21:30] Speaker B:** And then it's knowing if the thing you've built is good and right.  
然后就是知道你构建的东西是否好、是否正确。  
**[21:34] Speaker A:** And getting it out there in some early version at least. Does that sound right? Is there anything else of just like where human brains will continue to be useful for at least the next few months?  
还有至少把它以某个早期版本的形式发布出去。这样理解对吗?还有其他关于人类大脑至少在接下来几个月里会继续有用的地方吗?  
**[21:43] Speaker B:** I think humans still provide a level of common sense that the models don't.  
我认为人类仍然提供了模型所不具备的常识水平。  
**[21:49] Speaker B:** And there's like a thousand moving pieces to any product launch. Some of them are very small, but there's always a lot that could potentially go wrong.  
任何产品发布都有成千上万个活动部件。有些很小,但总有很多地方可能出错。  
**[21:59] Speaker B:** I think the model doesn't always have a great sense of who all the stakeholders are, how they relate to each other, what their preferences are, what are the right venues to communicate with them to keep them on board.  
我认为模型并不总是很好地了解所有利益相关者是谁,他们之间的关系如何,他们的偏好是什么,与他们沟通以保持他们支持的正确渠道是什么。  
**[22:09] Speaker B:** I think a lot of this more tacit common sense, like EQ kind of knowledge, is still very valuable.  
我认为很多这种更隐性的常识,比如情商类的知识,仍然非常有价值。  
**[22:17] Speaker B:** Of course, we want the models to get better at this and I think they will be, but right now I think there's still gaps.  
当然,我们希望模型在这方面变得更好,我认为它们会的,但现在我觉得还是有差距的。  
**[22:24] Speaker A:** How do you just kind of deal as a human going through so much constant change, just like just being on—  
作为人类,你如何应对这么多持续不断的变化,就像身处——  
**[22:30] Speaker A:** The inside of the tornado? Maybe it's calm there, but just like how do you stay on top of what's going on? How do you stay sane through all this craziness that we're moving through?  
龙卷风内部?也许那里很平静,但就是你怎么跟上正在发生的事情?你怎么在我们正在经历的所有这些疯狂中保持理智?  
**[22:39] Speaker B:** I think our team is full of people who lean into the chaos. So we try to face every challenge with a smile because there's always so much going on. There's always so many risks and tricky situations that, you know, if you get too stressed about anything you'll burn out.  
我认为我们团队里都是拥抱混乱的人。所以我们试图微笑着面对每一个挑战,因为总是有太多事情在发生。总是有那么多风险和棘手的情况,你知道,如果你对任何事情都太紧张,你就会精疲力竭。  
**[22:54] Speaker B:** And so we really look for people who can kind of look at a challenge and be like, that's going to be hard, but I'm excited to tackle it and I'm going to do the best that I possibly can. And I know I won't be perfect, but I'll be able to sleep at night knowing that I did my best.  
所以我们真的在寻找那些能够看着一个挑战然后说,这会很难,但我很兴奋去解决它,我会尽我所能做到最好。我知道我不会完美,但我能够安心入睡,因为我知道我已经尽力了。  
**[23:11] Speaker A:** That's an interesting answer to just like what skills will be important in this future, because I forget who said this, maybe Ben—man, that this is the most normal the world will—  
这是一个有趣的答案,关于在这个未来什么技能会重要,因为我忘了是谁说的,可能是Ben——天哪,说这是世界将会——  
**[23:21] Speaker A:** ever be.  
最正常的时候。  
**[23:23] Speaker B:** Yeah, it definitely gets harder. Like I feel like there are a lot of weeks where maybe Sunday night there's some like P0 and then by Monday there's like a P0 and by Monday afternoon there's a P0000 and you're like wow, I can't believe I was so worried about that P0 from Sunday.  
是的,确实会越来越难。我感觉有很多周,可能周日晚上有个P0级问题,然后到周一有个P0级问题,到周一下午有个P0000级问题,你就会想,哇,真不敢相信我周日还那么担心那个P0。  
**[23:40] Speaker B:** But I think you just have to acknowledge that there's only so much that you can do, that you need to sleep well so that you can make good decisions the next day and just like brutally prioritize where you spend your time. What's the most important thing to get right? And be okay letting things go. Like there's products that we ship that aren't as polished as I wish they were. But you know, our top goal is to help empower professional developers. And if a product isn't successful, as long as it's not blocking the core use case, it's okay because we'll hear the feedback and we'll fix it in the next  
但我认为你必须承认你能做的只有这么多,你需要睡好觉才能在第二天做出好的决策,并且要残酷地优先考虑你把时间花在哪里。什么是最重要的要做对的事情?并且要能接受放手一些事情。比如我们发布的产品有些并不像我希望的那样精致。但你知道,我们的首要目标是帮助赋能专业开发者。如果一个产品不成功,只要它不阻碍核心用例,就没关系,因为我们会听到反馈,我们会在下一个版本中修复它。  
**[24:16] Speaker A:** Release. Launching a feature that is buggy is the kind of thing that would have kept me up at night. But it is something that I am now able to live with, knowing that okay, we're going to get that quick feedback and we're going to fix it in the next release.  
发布一个有bug的功能是那种会让我彻夜难眠的事情。但这是我现在能够接受的事情,知道好吧,我们会得到那个快速反馈,我们会在下一个版本中修复它。  
**[24:31] Speaker B:** What I'm imagining is there's that gif, I think it's maybe from Pirates of the Caribbean, where it's this guy walking down a pair of stairs on a ship and the whole ship is just being demolished around him and he's so chill, just strolling down the staircase as everything's falling apart. And that's interesting because everyone I've met from Anthropic is just so chill and just so optimistic.  
我想象的是有那个gif动图,我想可能是来自《加勒比海盗》,就是这个人在船上走下一对楼梯,整艘船在他周围被摧毁,而他非常淡定,就这样漫步走下楼梯,周围一切都在崩塌。这很有意思,因为我遇到的每个Anthropic的人都非常淡定,非常乐观。  
**[24:51] Speaker A:** Yeah, that's I think that's a really interesting insight, just like having this calmness and optimism versus just like, oh my god, everything's crazy and going nuts. Yeah, I think if you don't have it, you'll get pretty burnt out. I think we also tend to hire  
是的,我觉得这是一个很有意思的洞察,就是保持这种冷静和乐观,而不是那种「天哪,一切都乱套了」的状态。我觉得如果你没有这种心态,很容易就会精疲力竭。我们在招人的时候也倾向于  
**[25:05] Speaker A:** People who have been in the industry for a while and have experienced lots of ups and downs and have a good sense for what gives them energy and how to maintain their energy over time, and I think that's helped us a lot.  
招那些在行业里待了一段时间、经历过各种起起落落的人,他们对什么能给自己带来能量、如何长期保持精力有很好的感觉,这对我们帮助很大。  
**[25:20] Speaker B:** So interesting. Something that I wanted to ask about is, so there's these roles blurring. Engineers are becoming PMs, everyone's dogs are cats, everyone's everyone. What do we lose in that world? Do we lose like career ladders and clear career paths? Do we lose design consistency, code quality? You know, there's probably some downsides. What are some things you find are just like, okay, that's something we're sacrificing for the greater good?  
很有意思。我想问的是,现在这些角色界限在模糊,工程师在做 PM 的事,所有人都在做所有事。在这种情况下我们会失去什么?我们会失去职业阶梯和清晰的职业路径吗?会失去设计一致性、代码质量吗?可能会有一些负面影响。你觉得有哪些是我们为了更大的利益而不得不牺牲的?  
**[25:42] Speaker A:** We're sacrificing product consistency. Historically, when code was expensive to write, you would carefully plan out everything in your product suite, how every product relates to each other, what the use case for every single one is, how they integrate, and you would...  
我们牺牲的是产品一致性。以前,当写代码成本很高的时候,你会仔细规划产品套件中的所有内容,每个产品之间的关系、每个产品的使用场景、它们如何集成,然后你会  
**[25:59] Speaker A:** Pretty much have one product for each use case. And now with AI moving so quickly and with so many ideas that we need to test out, we do sometimes have features that overlap with each other.  
基本上为每个使用场景做一个产品。但现在 AI 发展得太快了,我们需要测试的想法太多了,所以有时候确实会有功能之间相互重叠。  
**[26:11] Speaker A:** A lot of the times it's because there's two form factors that we love internally and we want the external audience to tell us which one is better.  
很多时候是因为我们内部有两种都很喜欢的形式,我们想让外部用户告诉我们哪一种更好。  
**[26:22] Speaker A:** What that means for someone who's a new user though is a new user might not know, okay, what is the best path to accomplish X.  
但这对新用户来说意味着,新用户可能不知道,好吧,完成某件事的最佳路径是什么。  
**[26:32] Speaker A:** There is more education we need to do to help people understand what the core features are and what the best practices are for using them.  
我们需要做更多的用户教育,帮助人们理解核心功能是什么,以及使用它们的最佳实践是什么。  
**[26:40] Speaker A:** I think this is the cost of launching a lot of features.  
我认为这是快速发布大量功能的代价。  
**[26:46] Speaker A:** I think users also feel like it's hard to keep up with the latest.  
我觉得用户也会感到很难跟上最新的东西。  
**[26:53] Speaker A:** Usually in traditional PM you ship a feature every month or quarter. And so it's really easy for a user to understand, okay, I just need to check in on this once a month and I'll learn some  
通常在传统的产品管理中,你每个月或每个季度发布一个功能。所以用户很容易理解,好的,我只需要每个月查看一次,就能学到一些  
**[27:04] Speaker A:** New things, and if I ignore it for six months, it's fine. I don't feel like I'm missing out.  
新东西,如果我六个月不管它,也没关系,我不会觉得错过了什么。  
**[27:09] Speaker A:** I think with these agentic tools, not just Coda Code and Cowork, but like across the whole ecosystem, people feel this need to like check Twitter every single day to see what the absolute latest thing is.  
我觉得对于这些智能体工具,不仅仅是 Coda Code 和 Cowork,而是整个生态系统,人们感到需要每天都查看 Twitter 来看最新的东西是什么。  
**[27:24] Speaker A:** And I think there's more we can do to help people feel less like they're on this ever-increasingly fast treadmill, and that they feel like—I would love people to feel like they can just open these tools.  
我认为我们可以做更多事情,让人们不那么觉得自己在一个越来越快的跑步机上,让他们感到——我希望人们能感到他们可以直接打开这些工具。  
**[27:38] Speaker A:** The tools will educate them or like teach them what they want to know, and that they can just feel more brought along.  
工具会教育他们或者教他们想知道的东西,让他们感到更被带着走。  
**[27:48] Speaker B:** Yeah, I saw you launch this really interesting feature the other day. I think it's /powerup where it basically walks you through all the cool ways and all basically all the best practices to use Claude Code. Is that kind of along these lines?  
是的,我看到你前几天发布了一个很有意思的功能。我想是 /powerup,它基本上会引导你了解所有很酷的方式,基本上就是使用 Claude Code 的所有最佳实践。这是不是沿着这个思路?  
**[27:57] Speaker A:** Yeah, exactly. So, in the past, we  
是的,没错。所以,过去我们  
**[27:59] Speaker A:** We didn't actually want to do something like PowerUp because we felt like the product should be intuitive enough that you don't actually need to go through any tutorial.  
其实不想做像 PowerUp 这样的东西,因为我们觉得产品应该足够直观,你实际上不需要经过任何教程。  
**[28:08] Speaker A:** And over time, we've just realized that there's just so many features and there's so much demand for a built-in onboarding experience that we diverged a bit from our original principle saying no onboarding flow and added this because there's just so many users who wanted to know there's 100 features.  
但随着时间推移,我们意识到功能实在太多了,对内置引导体验的需求也太大了,所以我们稍微偏离了最初「不做引导流程」的原则,加入了这个功能,因为有太多用户想知道有 100 个功能。  
**[28:27] Speaker A:** What are the 10 that I absolutely need to use? And so we put that together.  
哪 10 个是我绝对需要使用的?所以我们把它做出来了。  
**[28:32] Speaker B:** Yeah, it's such a bizarre world. So Anthropic has been really successful with B2B enterprises where traditionally you don't launch a bunch of stuff, you just kind of have a quarterly release maybe, and it's like the opposite of every day we got something new.  
是的,这真是个奇怪的世界。Anthropic 在 B2B 企业市场非常成功,而传统上你不会发布一堆东西,可能就是每季度发布一次,但现在完全相反,每天都有新东西。  
**[28:43] Speaker B:** So just maybe following that thread, the run Anthropic has been on is just otherworldly. Anthropic was way behind  
顺着这个话题,Anthropic 的发展势头简直不可思议。Anthropic 一开始远远落后  
**[28:51] Speaker A:** When it started, it was just like one of the least funded companies, didn't have distribution, wasn't the first to go, OpenAI was way ahead. It was just like no way Anthropic has any chance to compete significantly long term.  
刚开始的时候,它是资金最少的公司之一,没有分发渠道,也不是第一个出发的,OpenAI 遥遥领先。当时感觉 Anthropic 根本没有机会长期显著竞争。  
**[29:03] Speaker A:** Now it's just killing it, just beating the biggest companies' teams with so much—the growth is just like 11 billion dollars in ARR in one month percent growth. By the time this comes out, it'll probably be even higher.  
现在它简直势不可挡,击败了最大公司的团队,增长速度惊人——一个月就达到 110 亿美元的 ARR,百分比增长也很高。等这期节目出来的时候,可能会更高。  
**[29:20] Speaker B:** Just being on the inside, what are some ingredients that have allowed Anthropic to be this successful and kind of come from behind and do this well?  
作为内部人员,有哪些因素让 Anthropic 能够如此成功,从后面追上来并做得这么好?  
**[29:29] Speaker A:** The two most important things are, one, this unifying mission. It's hard to state how important this is. We hire people who care most about bringing safe AGI to all of humanity, and this is actually something that we reference frequently in our decisions about what our entire product org should focus on shipping. And because we put this mission above—  
最重要的两点是:第一,这个统一的使命。很难说清楚这有多重要。我们招聘的都是最关心为全人类带来安全 AGI 的人,而且我们在决定整个产品组织应该专注交付什么时,会经常参考这个使命。正因为我们把这个使命放在——  
**[29:57] Speaker A:** Any individual product line, we're able to make very fast decisions that cut across the entire org and execute on them in a unified way.  
任何单个产品线之上,我们才能够非常快速地做出跨整个组织的决策,并以统一的方式执行。  
**[30:07] Speaker A:** So I think this is something that I've never seen at a company of our scale.  
所以我认为这是我在我们这个规模的公司从未见过的。  
**[30:12] Speaker B:** And so just to make sure that's clear, so essentially having the number one mission is safety alignment, making sure AI is good for the world.  
为了确保理解清楚,本质上就是把第一使命定为安全对齐,确保 AI 对世界有益。  
**[30:18] Speaker B:** And you're saying just having that as a clear mission makes decisions a lot easier to make.  
你是说仅仅把这个作为明确的使命,就让决策变得容易得多。  
**[30:24] Speaker A:** If there's two competing priorities, we'll talk about which one is more important for Anthropic's mission.  
如果有两个相互竞争的优先事项,我们会讨论哪一个对 Anthropic 的使命更重要。  
**[30:28] Speaker A:** And it makes it a lot easier to decide which of the two we prioritize.  
这让我们更容易决定优先处理哪一个。  
**[30:32] Speaker A:** And then everyone will stand behind the one that we decide.  
然后每个人都会支持我们决定的那个。  
**[30:38] Speaker A:** And so sometimes that means that, hey, we want to ship something on Claude code, but this other thing is more important.  
所以有时候这意味着,嘿,我们想在 Claude Code 上发布某个功能,但另一件事更重要。  
**[30:43] Speaker A:** And so we deprioritize shipping this and we just wait until later.  
于是我们就降低这个发布的优先级,等到以后再说。  
**[30:46] Speaker A:** What's really  
真正  
**[30:47] Speaker A:** What's interesting about that is that it explains, I think, versus another company maybe that rhymes with OpenAI, they did a lot of different things. And what I'm hearing here essentially is like, okay, we're not going to launch a social network, we're not going to launch a feed of interesting information because it's not aligned to this mission, and that has kept Anthropic focused, which seems to be a core ingredient to the success.  
真正有意思的是,我觉得这解释了,相比另一家名字听起来像 OpenAI 的公司,他们做了很多不同的事情。而我在这里听到的本质上是,好的,我们不会推出社交网络,不会推出有趣信息流,因为这些不符合我们的使命,这让 Anthropic 保持了专注,而这似乎是成功的核心要素。  
**[31:10] Speaker B:** Well, when I think about mission, I think about putting Anthropic's goals ahead of any individual or any individual product. And so for me, I think the second thing that we're very good at is focus. I think mission to me is slightly different. Mission means that teams are willing to make sacrifices that hurt their own goals and their own KPIs in service of Anthropic's goals and Anthropic's KPIs. And people are very happy to make those trade-offs. So like an extreme example...  
嗯,当我想到使命时,我想到的是把 Anthropic 的目标放在任何个人或任何单个产品之前。所以对我来说,我认为我们非常擅长的第二件事是专注。我认为使命对我来说略有不同。使命意味着团队愿意做出牺牲,损害自己的目标和自己的 KPI,来服务于 Anthropic 的目标和 Anthropic 的 KPI。而且人们非常乐意做出这些权衡。比如一个极端的例子...  
**[31:46] Speaker A:** If Claude failed but Anthropic succeeded, I would be extremely happy, and the whole team is very willing to make decisions that follow that chain of thought.  
如果 Claude 失败了但 Anthropic 成功了,我会非常高兴,整个团队都非常愿意按照这个思路做决策。  
**[31:58] Speaker B:** I don't know if you can talk about this in depth, but do you feel like the open source decision is a part of this? Just like, okay, this is not furthering the mission of Anthropic, we need to stop this because it's not working in the way we want it to work?  
不知道你能不能深入谈谈这个,但你觉得开源决策是这个的一部分吗?就像,好吧,这没有推进 Anthropic 的使命,我们需要停止这个,因为它没有按我们想要的方式运作?  
**[32:09] Speaker A:** I think one of the most important things for Anthropic is to grow the number of users that we're able to reach. One of the ways that we're able to do this is with the Claude subscriptions with our first-party products, and so we just very much want to double down on that, but that does come at the expense of third-party products sometimes.  
我认为对 Anthropic 来说最重要的事情之一是增加我们能够触达的用户数量。我们能做到这一点的方式之一是通过 Claude 订阅和我们的第一方产品,所以我们非常想在这方面加倍投入,但这有时确实会以牺牲第三方产品为代价。  
**[32:28] Speaker B:** So we've been talking about Claude, agentic work, all these things. Something that I want to make sure people get, and I'm curious just how you use these  
我们一直在谈论 Claude、智能体工作,所有这些东西。有件事我想确保大家理解,我也很好奇你是如何使用这些  
**[32:34] Speaker A:** Tools. So there's Cloud Code, there's Cloud Desktop, there's Cowork. What's the best way to understand when to use which? When do you use each of these three?  
工具的。有 Claude Code,有 Claude Desktop,还有 Cowork。理解何时使用哪个的最佳方式是什么?你什么时候使用这三个中的每一个?  
**[32:44] Speaker B:** So I tend to use Cloud Code in the terminal when I'm just kicking off like a one-off coding task and I want all of the latest features.  
我倾向于在终端中使用 Claude Code,当我只是启动一个一次性的编码任务,并且想要所有最新功能时。  
**[32:53] Speaker B:** The CLI is our initial product surface and it's also the one where our features often land first, and so it's the most powerful of all the tools.  
CLI 是我们最初的产品界面,也是我们的功能经常首先落地的地方,所以它是所有工具中最强大的。  
**[33:03] Speaker B:** So that's what I tend to use when I'm just like trying to kick off one or like maybe like a handful of tasks at a time.  
所以当我只是想启动一个或者可能几个任务时,我倾向于使用它。  
**[33:12] Speaker B:** I think Desktop really shines when you're doing something that requires front-end work.  
我认为 Desktop 在你做需要前端工作的事情时真正发光。  
**[33:16] Speaker B:** And so one thing that I love to do is to use our preview feature. So if I'm building a web app, I'll often use Cloud Code and Desktop. I'll have the preview pane open on the right-hand side so that I can actually see the web app that I'm making in real  
所以我喜欢做的一件事是使用我们的预览功能。如果我在构建一个 web 应用,我经常会同时使用 Claude Code 和 Desktop。我会在右侧打开预览面板,这样我就能实际看到我正在制作的 web 应用  
**[33:31] Speaker A:** Time as I'm chatting with Claude, it's also really great for people who want something a bit more graphical. A terminal can feel very unfamiliar to someone who's nontechnical.  
在我与 Claude 聊天时实时显示,这对想要更图形化界面的人来说也非常好。终端对非技术人员来说可能感觉很陌生。  
**[33:42] Speaker A:** You get a bunch of these scary popups on your machine and you can't click around the way that you're used to in pretty much every other product that you use.  
你的机器上会出现一堆这些吓人的弹窗,而且你不能像在几乎所有其他你使用的产品中那样点击操作。  
**[33:49] Speaker A:** So there's a lot of people who just don't feel comfortable in terminal, and if that's you, I would highly recommend checking out Claude Code on desktop.  
有很多人在终端里操作会感到不舒服,如果你也是这样,我强烈推荐试试桌面版的 Claude Code。  
**[33:57] Speaker A:** Desktop is also great for getting an at-a-glance view of everything that's happening.  
桌面版还有个好处,就是能让你一眼看清所有正在进行的事情。  
**[34:01] Speaker A:** So you can see your CLI terminal sessions in desktop, you can see your other desktop sessions, you can see your sessions that you kicked off on web and mobile.  
你可以在桌面版里看到你的 CLI 终端会话,看到其他桌面会话,还能看到你在网页端和移动端启动的会话。  
**[34:12] Speaker A:** So it's a one-stop control plane where you can see all of your tasks.  
它就像一个一站式的控制面板,让你能看到所有任务。  
**[34:14] Speaker A:** I think the benefit of web and mobile is that it's really great for kicking things off on the go. So CLI and desktop both require  
我觉得网页版和移动版的优势在于,它们特别适合在外出时启动任务。因为 CLI 和桌面版都需要  
**[34:24] Speaker A:** You to be on your local laptop. And this is constraining because sometimes you're out and about, you're like touching grass, you're going on a walk and you don't have your laptop open and you don't—I can't count the number of people who I've seen like holding their laptop open like tethered to their phone while they're outside.  
你在本地笔记本电脑上操作。这就有局限性了,因为有时候你在外面,比如出去走走、散散步,没有打开笔记本电脑,而且——我见过太多人在外面举着打开的笔记本电脑,还要连着手机热点。  
**[34:39] Speaker A:** And this just means that we're missing a product that solves that need.  
这说明我们缺少一个能解决这种需求的产品。  
**[34:44] Speaker A:** And so for me, what mobile lets you do is kick off these tasks on the go so that you don't need to bring your laptop everywhere and make sure that your laptop's open wherever you are.  
所以对我来说,移动版的作用就是让你能在外出时启动这些任务,这样你就不需要到哪儿都带着笔记本电脑,也不用确保笔记本一直开着。  
**[34:57] Speaker B:** I love that. I've seen people on plane—like it's just such a meme now. Just I need to finish, let this agent finish. I can't shut this down. I need Wi-Fi.  
我太喜欢这个了。我见过有人在飞机上——这都成梗了。就是「我得让这个 agent 完成任务,不能关机,我需要 Wi-Fi」。  
**[35:04] Speaker A:** And then I think for co-work, the role that this fills is there's a lot of work that everyone does where the output isn't code. So whether that's like  
然后我觉得 Cowork 的定位是,大家有很多工作的产出物并不是代码。比如  
**[35:13] Speaker A:** Getting to Slack zero or inbox zero, or whether that's creating a slide deck for some customer meeting that's coming up, or whether that's writing a quick doc on what the goals of a feature are or what the launch plan for a feature is.  
清空 Slack 消息或者清空收件箱,或者为即将到来的客户会议制作幻灯片,或者快速写一份文档说明某个功能的目标是什么,或者这个功能的发布计划是什么。  
**[35:25] Speaker A:** All these tasks produce outputs that are non-code, and Cowork is best positioned for that.  
所有这些任务产出的都不是代码,而 Cowork 最适合处理这类工作。  
**[35:32] Speaker A:** So the way that I split the products in my mind is if I'm building something where the output is code, I'll use Claude Code or Desktop or Claude Code on mobile.  
所以我在心里是这样区分这些产品的:如果我要构建的东西产出是代码,我会用 Claude Code、桌面版或者移动端的 Claude Code。  
**[35:43] Speaker A:** And if the output is anything that's not code, I'll use Cowork for it.  
如果产出不是代码,我就会用 Cowork。  
**[35:48] Speaker B:** People are just like sleeping on the success that Cowork—it's just like growing incredibly fast, and I think people still don't understand maybe what it's for.  
大家真的是低估了 Cowork 的成功——它增长得非常快,但我觉得人们可能还不太理解它是用来做什么的。  
**[35:58] Speaker B:** And so what if you give us a couple use cases just in your work as a PM? What are some like really interesting, maybe unexpected ways you use Cowork to save you time, get more work done?  
能不能给我们讲几个使用场景?作为产品经理,你有哪些特别有意思的、也许出人意料的方式在用 Cowork 来节省时间、完成更多工作?  
**[36:08] Speaker A:** If you're getting started on Cowork, the first thing that you really need to do is connect all the data sources that are relevant to your role, because Cowork can only do a great job if it has access to all the context that it needs to be able to curate the output for you.  
如果你刚开始用 Cowork,首先真正需要做的是连接所有与你工作相关的数据源,因为只有 Cowork 能访问到它需要的所有上下文,它才能为你精心定制出色的产出。  
**[36:23] Speaker A:** So what that means for me is I connect it to my Google Calendar, I connect it to my Slack, to my Gmail, to my Google Drive, so that it just knows it has the flexibility to find relevant context, to ask questions, to pull in threads, and this substantially improves the quality of the result.  
对我来说就是,我会连接 Google Calendar、Slack、Gmail、Google Drive,这样它就能灵活地找到相关上下文、提出问题、提取对话线索,这会大幅提升结果的质量。  
**[36:42] Speaker A:** The kinds of things I use it for are, like last night I was working where we have this Cowork Cloud conference coming up and there's a few talks that I'm giving there, and one of the talks that we're doing talks about the transition of Cowork from an assistant to like a full-on agent, and one of the things that I wanted to do in this talk was to showcase all of the products that we've  
我用它做的事情比如,昨晚我在准备即将到来的 Cowork Cloud 大会,我要在那里做几场演讲,其中一场演讲要讲 Cowork 从助手到完整 agent 的转变,我想在这场演讲中展示我们为了实现这个转变  
**[37:09] Speaker A:** We've been shipping that to enable this transition and also to figure out, okay, what are the success stories that people have had internally that we can use as demos.  
一直在发布的所有产品,还要找出内部有哪些成功案例可以用作演示。  
**[37:17] Speaker A:** And so I have my Google Drive connected, I have Slack connected. Alex, who's our product marketer, put together like a draft of what the points that he thinks we should cover are.  
我连接了 Google Drive 和 Slack。我们的产品营销 Alex 整理了一份草稿,列出了他认为我们应该涵盖的要点。  
**[37:32] Speaker A:** And so I just like fed this all into Cowork. I told Cowork the narrative that I want to tell.  
所以我就把这些都输入给 Cowork,告诉它我想讲述的叙事线。  
**[37:37] Speaker A:** And it actually just worked for an hour. It walked through Twitter to see what we launched.  
然后它真的就工作了一个小时。它浏览了 Twitter 看我们发布了什么。  
**[37:42] Speaker A:** It looked through our evergreen launch room. It looked in our Codebase announce channel, which is where our team posts demos of how they've been getting the most value out of Codebase.  
它查看了我们的常青发布室,还看了我们的 Codebase announce 频道,那是我们团队发布如何从 Codebase 获得最大价值的演示的地方。  
**[37:53] Speaker A:** And it synthesized all this together to this 20-page deck that I woke up to this morning, and I read through it and it was like pretty good.  
它把所有这些综合成了一份 20 页的幻灯片,我今早醒来看到的,读了一遍,质量还挺不错的。  
**[38:00] Speaker A:** There were a few tweaks, so I did have to give it a round of feedback.  
有一些地方需要调整,所以我确实给了一轮反馈。  
**[38:04] Speaker A:** Of feedback. I like my slides to have extremely minimal words and it was a little too wordy, but you know, it was far faster than what I would be able to produce.  
反馈意见。我喜欢幻灯片上的文字极简,而它生成的文字稍微有点多,但你知道,这比我自己做要快得多。  
**[38:15] Speaker A:** And because Cowork has access to our whole design system, it actually looks like an Anthropic designer put it together.  
而且因为 Cowork 可以访问我们整个设计系统,它实际上看起来就像是 Anthropic 的设计师做出来的。  
**[38:22] Speaker A:** When you visually see it, you're like, "Oh, this is incredibly polished."  
当你视觉上看到它时,你会觉得「哦,这真的非常精致」。  
**[38:29] Speaker A:** So these are the kinds of things that are so much faster. Making this slide deck would have taken me hours, but instead it turns out a draft that is actually quite good so I could focus on making sure that the demos are amazing that we plug into it.  
所以这些事情快了太多。制作这个幻灯片本来要花我好几个小时,但现在它输出了一个其实相当不错的草稿,这样我就可以专注于确保我们插入的演示效果惊艳。  
**[38:45] Speaker B:** This sounds like a dream come true to PMs. Putting decks together is so annoying.  
这对产品经理来说简直是梦想成真。做幻灯片真的太烦人了。  
**[38:49] Speaker A:** It's so slow.  
太慢了。  
**[38:51] Speaker B:** And I love that people will see this deck whenever you present this. This will be out in the world. Obviously it's not the one-shotted version, but...  
我很喜欢的一点是,当你展示这个幻灯片时,人们会看到它。它会出现在公众面前。显然这不是一次生成的版本,但是……  
**[38:58] Speaker A:** You've iterated on it, so just to help people try this for themselves. So step one is connect their—what did you say? Slack. What else do you suggest they connect?  
你已经迭代过了,所以为了帮助大家自己尝试这个。第一步是连接他们的——你刚才说什么来着?Slack。你还建议他们连接什么?  
**[39:07] Speaker B:** Slack, Google Calendar, Gmail, Google Drive. You should connect your communications tools and where you store your source of truth data for what your team cares about, what you care about, and what you're working on.  
Slack、Google Calendar、Gmail、Google Drive。你应该连接你的沟通工具,以及你存储团队关心的内容、你关心的内容和你正在做的事情的真实数据源的地方。  
**[39:21] Speaker A:** Okay. And then what was the prompt roughly that you put in there to generate this deck?  
好的。然后你输入的提示词大概是什么来生成这个幻灯片的?  
**[39:26] Speaker B:** So I just wrote, 'Make me a slide deck for the Code with Claude conference. This is what our PMM suggested it should cover. This is the current draft that I made that I don't like. This is one that I made manually that I don't like, but I linked it. Can you start by creating a proposed outline with details? Also, make sure it doesn't overlap too much with a keynote talk, which is more important.' And then Claude read a bunch—  
我就写了,「给我做一个 Code with Claude 大会的幻灯片。这是我们的产品营销经理建议应该涵盖的内容。这是我做的当前草稿,我不喜欢。这是我手动做的一个,我也不喜欢,但我附上了链接。你能先创建一个带详细信息的提议大纲吗?另外,确保它不要和主题演讲重叠太多,主题演讲更重要。」然后 Claude 读了很多——  
**[39:47] Speaker A:** Of the links that I sent to it and created a proposed outline. So then I read through its proposal and all the different ideas that it had generated for what we could cover and I just made a decision on what I wanted to actually be in the final deck.  
我发给它的链接,并创建了一个提议大纲。然后我通读了它的提议以及它为我们可以涵盖的内容生成的所有不同想法,我就决定了我实际想要放在最终幻灯片中的内容。  
**[40:02] Speaker A:** And I think this is like an example of what the role of the PM still is today. It's like Claude is a great brainstorming partner. It's able to synthesize a massive amount of information really quickly and present all of the possibilities to you.  
我认为这就是产品经理今天仍然扮演的角色的一个例子。Claude 是一个很好的头脑风暴伙伴。它能够非常快速地综合大量信息并向你展示所有的可能性。  
**[40:17] Speaker A:** But the role of the PM is still to make the end decision of okay what should belong in the final product.  
但产品经理的角色仍然是做最终决定,决定什么应该属于最终产品。  
**[40:24] Speaker A:** So for this what I ended up deciding was that I wanted the talk to cover the progression from making local tasks successful to making every PR green to like helping engineers land more PRs and for each of these which demo would be the most compelling and then after this decision about the outline Codeium just like went off for a few hours and built  
所以对于这个,我最终决定的是,我希望演讲涵盖从让本地任务成功,到让每个 PR 都通过,再到帮助工程师提交更多 PR 的进展过程,并且对于每一个阶段,哪个演示会最有说服力,然后在做出关于大纲的决定之后,Codeium 就自己跑去花了几个小时构建——  
**[40:49] Speaker A:** The whole slide deck.  
整个幻灯片。  
**[40:50] Speaker B:** This is so awesome. What an awesome part of the job to not have to do anymore. And it feels like you're talking to essentially a deck designer that also has like actual knowledge about what you've worked on and can make it actually the content what you want it to be, not just make it look really nice.  
这太棒了。工作中不用再做这部分真是太好了。感觉你基本上是在和一个幻灯片设计师对话,而且它实际上了解你做过的工作,可以让内容真正成为你想要的样子,而不仅仅是让它看起来很漂亮。  
**[41:08] Speaker B:** How did you do the design system piece? How does that work? How does it know the design system of Anthropic?  
你是怎么做设计系统这部分的?这是怎么工作的?它怎么知道 Anthropic 的设计系统?  
**[41:14] Speaker A:** So what I did for this is we actually already have like a standardized deck that we use across all of our external engagements. And so I just gave Claude access to that. And so it's able to see like what colors we use, what fonts we use, the different kinds of—  
我做的是,我们实际上已经有一个标准化的幻灯片模板,用于我们所有的对外活动。所以我就给了 Claude 访问权限。这样它就能看到我们使用什么颜色、什么字体、不同种类的——  
**[41:31] Speaker B:** What's it called? Like slide formats that are possible. And so it has like 20 of these example slides.  
叫什么来着?就是可能的幻灯片格式。所以它有大约 20 个这样的示例幻灯片。  
**[41:37] Speaker A:** Give an example. Got it. So you like upload here's our template work from—  
举个例子。明白了。所以你就像上传「这是我们的模板,按照这个来做」——  
**[41:40] Speaker A:** This.  
对。  
**[41:40] Speaker B:** Yeah, you can also connect to like your Figma MCP if you have your slide format saved there and it can pull that in.  
是的,你也可以连接到你的 Figma MCP,如果你把幻灯片格式保存在那里,它可以把那个拉进来。  
**[41:48] Speaker A:** Along those lines, something I'm always curious about is what's kind of in your stack of tools as a PM at Anthropic—obviously Claude, Code, and co-work and all the Anthropic tools. What else are you using? Slack you mentioned? Is there anything else?  
沿着这个思路,我一直很好奇的是,作为 Anthropic 的产品经理,你的工具栈里有什么——显然有 Claude、Code 和 Cowork 以及所有 Anthropic 的工具。你还在用什么?你提到了 Slack?还有别的吗?  
**[42:02] Speaker B:** So my stack is pretty heavily Claude, Code, co-work. Anthropic largely runs on Slack. I feel like it's like the core OS of our company and day-to-day. A lot of—I would say maybe 30% of my time is pushing the boundaries of what co-work can do so that I have a very strong sense of what we're not good at. And I spend a lot of time talking with the model to understand why it makes  
我的工具栈主要是 Claude、Code 和 co-work。Anthropic 公司基本上是靠 Slack 运转的,感觉它就像是我们公司的核心操作系统,贯穿日常工作。我大概有 30% 的时间在探索 co-work 的边界能力,这样我就能清楚地知道我们哪些方面还不够好。我花很多时间和模型对话,想理解它为什么会犯  
**[42:38] Speaker A:** Mistakes that it does. We actually have a lot of internal tools that we make.  
那些错误。其实我们内部做了很多工具。  
**[42:42] Speaker A:** Like I think one of the things that Claude has really unlocked for our entire company is it really lowers the barrier to making any custom app that you want.  
我觉得 Claude 给整个公司带来的一个重要改变是,它真正降低了制作任何定制应用的门槛。  
**[42:52] Speaker A:** And so we've seen this surge in personalized work software that people are building for custom use cases instead of using tools that don't perfectly fit the use case.  
所以我们看到公司里涌现出大量个性化的工作软件,大家针对特定场景开发工具,而不是凑合使用那些不太合适的现成工具。  
**[43:06] Speaker B:** I got to hear more. What are some examples? What are things you've built or other people built that are really popular and useful?  
我得多听听。有什么具体例子吗?你或者其他人做了哪些特别受欢迎、特别实用的东西?  
**[43:12] Speaker A:** One of the sales folks on Claude, he realized he was making these repetitive decks over and over and over again.  
Claude 团队有个做销售的同事,他发现自己在反复制作类似的演示文稿。  
**[43:20] Speaker A:** And so he actually has this web app that he built with the examples of the core Claude decks that we know work well.  
于是他做了一个网页应用,里面有我们知道效果很好的核心 Claude 演示文稿模板。  
**[43:28] Speaker A:** So like a 101, 201, and mastering Claude.  
比如入门版、进阶版和精通 Claude 版。  
**[43:30] Speaker A:** And then he has a way to input specific customer context that pulls from Salesforce.  
然后他设计了一个输入界面,可以从 Salesforce 拉取特定客户的背景信息。  
**[43:37] Speaker A:** From Gong that pulls from other notes so that we can customize the decks for specific customers.  
从 Gong 以及其他笔记里提取信息,这样我们就能为特定客户定制演示文稿。  
**[43:41] Speaker A:** And so it'll pull out things like, okay, this customer is using like Bedrock or Claude for Enterprise or Console, which affects what features are available to them.  
系统会提取出这样的信息:这个客户用的是 Bedrock 还是 Claude for Enterprise 还是 Console,这会影响他们能用哪些功能。  
**[43:51] Speaker A:** It will pull out things like, okay, this customer is concerned about like the code review stage of the SDLC.  
系统还会提取出:这个客户关注软件开发生命周期中的代码审查阶段。  
**[43:57] Speaker A:** And so we'll add a slide about our code review features there.  
那我们就会加一页关于代码审查功能的幻灯片。  
**[44:02] Speaker A:** It'll pull out things like, okay, this customer needs to be like HIPAA compliant or needs XYZ security controls.  
系统还会识别出:这个客户需要符合 HIPAA 合规要求,或者需要某些特定的安全控制。  
**[44:09] Speaker A:** And so we'll make sure to add a slide or two in their deck about that.  
那我们就会确保在他们的演示文稿里加上一两页关于这方面的内容。  
**[44:11] Speaker A:** And then for example, if this is a customer that's on Vertex or Bedrock and doesn't want to use Claude for Enterprise, then we'll just take out some of the slides that are Claude for Enterprise only features.  
再比如,如果客户用的是 Vertex 或 Bedrock,不想用 Claude for Enterprise,那我们就会把那些只有 Claude for Enterprise 才有的功能页面删掉。  
**[44:27] Speaker A:** And so normally this is like manual work that could take 20 to 30 minutes, and so people either like spend that  
通常这种手工操作要花 20 到 30 分钟,所以大家要么花时间去做,  
**[44:34] Speaker A:** Time doing it or they'll just decide not to do it and use the general deck. With this it takes like a few seconds and you get a tailored deck.  
要么就干脆不做,直接用通用版本。有了这个工具,几秒钟就能生成一份定制化的演示文稿。  
**[44:42] Speaker B:** What's interesting about it is like Slack is like the tool that nobody's—it's just like nobody's trying to create their own. Slack just continues to win and it's just like the way you describe it is kind of the OS of so many companies.  
有意思的是,Slack 这个工具没人想去——就是没人试图自己做一个。Slack 就这样一直赢下去,就像你说的,它已经成为很多公司的操作系统。  
**[44:53] Speaker B:** It's so interesting like people talk about Salesforce as just like SaaS. We don't need SaaS software anymore. We're going to build our own. It's like Slack is a durable tool that nobody wants to try to compete with and build a better version.  
特别有意思的是,大家都在说 Salesforce 只是 SaaS 软件,我们不再需要 SaaS 了,我们要自己做。但 Slack 是个持久耐用的工具,没人想去竞争或者做个更好的版本。  
**[45:04] Speaker A:** I think it's pretty important communications infrastructure and I think they do the core task of helping everyone get real-time updates incredibly well.  
我觉得它是很重要的通信基础设施,而且他们在帮助所有人获取实时更新这个核心任务上做得非常出色。  
**[45:13] Speaker B:** Yeah. Like people hate on Slack, but it's really great at what it's trying to do and like the most cutting edge teams are hooked on it. So interesting.  
是的。虽然有人吐槽 Slack,但它在自己要做的事情上真的很优秀,而且最前沿的团队都离不开它。很有意思。  
**[45:21] Speaker A:** Yeah, and I also love how easy they've made it to customize it. And so we love making Slack bots, and this kind of hackability means that we're able to integrate with Slack the way that we want to. So really appreciate Slack's work on that.  
对,我也很喜欢他们把定制化做得这么简单。所以我们很喜欢做 Slack 机器人,这种可定制性意味着我们能按自己想要的方式与 Slack 集成。真的很感谢 Slack 在这方面的工作。  
**[45:37] Speaker B:** Time to buy some CRM stock. I am so excited to tell you about this season's supporting sponsor, Vanta. Vanta helps over 15,000 companies like Cursor, Ramp, Duolingo, Snowflake, and Atlassian earn and prove trust with their customers. Teams are building and shipping products faster than ever thanks to AI. But as a result, the amount of risk being introduced into your product and your business is higher than it's ever been. Every security leader that I talk to is feeling the increasing weight of protecting their organization, their business, and not to mention their customer data. Because things are moving so fast, they are constantly reacting, having to guess at priorities, and  
该买点 CRM 股票了。我很高兴向大家介绍本季的赞助商 Vanta。Vanta 帮助超过 15,000 家公司,比如 Cursor、Ramp、Duolingo、Snowflake 和 Atlassian,赢得并证明客户的信任。得益于 AI,团队构建和发布产品的速度比以往任何时候都快。但结果是,引入到产品和业务中的风险也比以往任何时候都高。我接触的每个安全负责人都感受到保护组织、业务以及客户数据的压力越来越大。因为事情发展太快,他们总是在被动应对,不得不猜测优先级,  
**[46:19] Speaker A:** Having to make do with outdated solutions. Vanta automates compliance and risk management with over 35 security and privacy frameworks including SOC 2, ISO 27001 and HIPAA.  
不得不凑合使用过时的解决方案。Vanta 自动化了合规和风险管理,支持超过 35 个安全和隐私框架,包括 SOC 2、ISO 27001 和 HIPAA。  
**[46:31] Speaker A:** This helps companies get compliant fast and stay compliant more than ever before. Trust has the power to make or break your business.  
这帮助公司快速达到合规要求,并且比以往任何时候都更能保持合规状态。信任有能力成就或毁掉你的业务。  
**[46:38] Speaker A:** Learn more at vanta.com/lenny.  
访问 vanta.com/lenny 了解更多信息。  
**[46:42] Speaker A:** And as a listener of this podcast, you get $1,000 off Vanta. That's vanta.com/lenny.  
作为本播客的听众,你可以获得 1000 美元的 Vanta 优惠。网址是 vanta.com/lenny。  
**[46:48] Speaker B:** Okay. So you talked about all these different teams and how they use Claude Code and Copilot to operate. Which teams do you find other than engineering? I imagine engineering is the biggest token spender, but if not that'd be really interesting. What's kind of like the second place function right now for tokens?  
好的。你提到了所有这些不同的团队以及他们如何使用 Claude Code 和 Copilot 来工作。除了工程团队之外,你发现哪些团队用得多?我想工程团队应该是最大的 token 消耗方,但如果不是的话那就很有意思了。目前 token 使用量排第二的职能部门是什么?  
**[47:04] Speaker A:** Oh, applied AI is amazing at pushing the boundaries of what Claude Code and Copilot can do. A lot of our applied AI team  
哦,应用 AI 团队在推动 Claude Code 和 Copilot 能力边界方面做得非常出色。我们应用 AI 团队的很多成员  
**[47:14] Speaker A:** spends time with our customers helping them adopt our API. And so sometimes our applied team will for example make prototypes on behalf of these customers which Claude Code makes so much faster than it used to be.  
会花时间帮助客户采用我们的 API。所以有时候我们的应用团队会为这些客户制作原型,而 Claude Code 让这个过程比以前快得多。  
**[47:26] Speaker A:** They also have the dual goal of needing to manage a lot of customer comms, a lot of like customer inbound and historical context call notes.  
他们还有一个双重目标,就是需要管理大量的客户沟通、客户咨询以及历史背景和通话记录。  
**[47:38] Speaker A:** And so they're both extremely heavy on co-work and on Claude Code.  
所以他们在 Cowork 和 Claude Code 上的使用都非常频繁。  
**[47:42] Speaker B:** And just to understand applied AI, is that like is that like forward deployed engineering sort of role? Like what do they how would you how would most people describe what the applied AI team is doing?  
我想理解一下应用 AI 团队,这是不是类似前置部署工程师那种角色?大多数人会怎么描述应用 AI 团队在做什么?  
**[47:51] Speaker A:** Yeah, it's helping our customers adopt the latest API and model features across their company both for powering their company's products and also for internal acceleration.  
是的,就是帮助我们的客户在整个公司范围内采用最新的 API 和模型功能,既用于驱动他们公司的产品,也用于内部提效。  
**[48:05] Speaker B:** Got it. So it's like customer success go-to-market-y kind of like forward deployed engineering sort of.  
明白了。所以有点像客户成功、市场拓展类的前置部署工程师那种感觉。  
**[48:10] Speaker A:** Exactly. It's like a very technical go-to-market person.  
没错。就像是技术能力很强的市场拓展人员。  
**[48:13] Speaker B:** Got it. Okay. Awesome. So you're saying that might be the second org that uses the most tokens.  
明白了。好的,很棒。所以你是说这可能是使用 token 第二多的部门。  
**[48:19] Speaker A:** Yeah. And then we also see them pushing the boundaries of what Cowork can do. So for example, a lot of these folks cover multiple customers and in any given day can have like five to ten customer engagements on a high day.  
对。然后我们还看到他们在推动 Cowork 能做什么的边界。比如说,这些人中很多都负责多个客户,在繁忙的日子里一天可能会有五到十次客户对接。  
**[48:36] Speaker A:** And so what they often use Cowork to do is the night before they'll ask it to summarize, okay, what are all my customer meetings that are coming up the next day?  
所以他们经常用 Cowork 做的事情是,前一天晚上让它总结一下,好,我第二天有哪些客户会议要开?  
**[48:45] Speaker A:** What are all the things that this customer has asked me for? What's top of mind for them? What are the action items from the past meetings? And Cowork will just put together this dossier, this brief of what they should be aware of going into the next meeting. And Cowork can also research answers, so if a  
这个客户向我提出了哪些要求?他们最关心什么?之前会议中有哪些待办事项?然后 Cowork 就会整理出这份档案,这份简报,告诉他们进入下次会议前应该了解什么。Cowork 还可以研究答案,所以如果  
**[49:04] Speaker A:** Customer asked, okay, when is feature X going to launch? CoWork can help the PM person research through Slack to get the latest ETA, add that to the notes so that during the customer call the PM person has the absolute latest. And these are just workflows that people are building for themselves and sharing with other people on their team.  
客户问,好,功能 X 什么时候上线?Cowork 可以帮助产品经理通过 Slack 搜索获取最新的预计时间,把它加到笔记里,这样在客户电话会议期间,产品经理就能掌握最新信息。这些都是人们为自己构建的工作流程,然后分享给团队里的其他人。  
**[49:25] Speaker B:** So cool. Something that kind of, this question, this trend, I don't know, question topic comes up a lot recently, which is token spend exceeding people's salary, where people just use AI and it costs more than how much they're making. Are there any numbers floating around Anthropic of just like how much token spend, say engineers spend, I don't know, a month, a day, PMs, anything like that?  
太酷了。有个话题,这个问题,这个趋势,我不知道该怎么说,最近经常被提起,就是 token 支出超过了员工工资,人们使用 AI 的成本比他们的收入还高。Anthropic 内部有没有一些数据,比如工程师一个月、一天花多少 token,产品经理呢,有这类数据吗?  
**[49:50] Speaker A:** It is clear to us that as the models get better, people delegate far more tasks to it and they spend a lot more hours in tools like Claude Code and CoWork. And so we do see the token cost per engineer or...  
我们很清楚的是,随着模型变得更好,人们会把更多任务委托给它,他们在 Claude Code 和 Cowork 这类工具上花的时间也多得多。所以我们确实看到每个工程师或者  
**[50:04] Speaker A:** Like per any knowledge worker increase, every time that there's a model jump or like a substantial product improvement.  
每个知识工作者的 token 成本在增加,每次有模型升级或者产品有重大改进的时候都会这样。  
**[50:12] Speaker A:** I think it's still much lower than what the average engineer salary is, but we see the percentage increasing over time.  
我觉得这个成本仍然远低于工程师的平均工资,但我们看到这个百分比随着时间在增长。  
**[50:21] Speaker B:** It's such an interesting—like we talked about how you have access to the most cutting edge models and other advantages of working at Anthropic. I believe you guys have basically unlimited tokens. You can use as much as you want. Is that right?  
这真的很有意思——就像我们讨论过的,你们可以使用最前沿的模型,还有在 Anthropic 工作的其他优势。我相信你们基本上有无限的 token。你们想用多少就能用多少。对吗?  
**[50:33] Speaker A:** We can use a lot of tokens. Some people do run into limits, so...  
我们可以使用很多 token。有些人确实会遇到限制,所以...  
**[50:36] Speaker B:** Okay, there's a limit. Okay, Baris, shut it down. Okay.  
好吧,还是有限制的。好的,Baris,别用了。好吧。  
**[50:42] Speaker B:** Like, it's so interesting how many advantages come from having the most advanced model. It's such an interesting flywheel that starts to kick in.  
真的很有意思,拥有最先进的模型能带来多少优势。这是一个非常有意思的飞轮效应,开始发挥作用了。  
**[50:50] Speaker A:** I think we also believe a lot in empowering our internal teams to build as fast as possible. And we also trust that everyone understands how much capacity that serving these...  
我认为我们非常相信赋能内部团队,让他们能够尽可能快地构建产品。同时我们也信任每个人都理解为这些模型提供服务的成本有多高...  
**[51:02] Speaker A:** Models truly cost, and we trust our team to use the tokens responsibly. So it's very frowned upon to waste tokens, but we do trust individuals to make that judgment call.  
真正的成本是多少,我们相信团队会负责任地使用 token。所以浪费 token 是非常不被认可的行为,但我们确实信任个人能做出正确的判断。  
**[51:14] Speaker B:** Awesome. Coming back to the PM role, we talked a little bit about this, but I think this will be really interesting for people to hear. Just what I want to understand is what do you think are the kind of emerging skills that PMs need to develop, or that AI companies most look for when they're hiring PMs these days?  
很好。回到 PM 这个角色,我们之前聊过一点,但我觉得这个话题对大家会很有意思。我想了解的是,你认为 PM 需要培养哪些新兴技能,或者说 AI 公司在招聘 PM 时最看重什么?  
**[51:35] Speaker A:** I think the hardest skill is being able to define what the product should look like a month from now. I think there's a lot of ambiguity in what models are capable of in that timeline and how user behavior will change.  
我认为最难的技能是能够定义一个月后产品应该是什么样子。因为在这个时间线内,模型的能力会如何发展、用户行为会如何变化,都存在很大的不确定性。  
**[51:52] Speaker A:** But I think there are patterns that the best PMs can see based on how users are pushing the limits of the existing product, and the best PMs can sense that.  
但我认为最优秀的 PM 能够看到一些模式,基于用户如何突破现有产品的极限,最好的 PM 能够感知到这一点。  
**[52:04] Speaker A:** Can set a direction and can steadily execute towards it and change the path if the model capabilities are much better than or worse than what they had originally expected.  
能够设定方向并稳步执行,如果模型能力比最初预期的好得多或差得多,也能够调整路径。  
**[52:14] Speaker A:** I think it is very hard to be the right amount of AGI-pilled because I think everyone can see this future where the models are extremely smart and can do almost everything, in which case you actually don't need that complicated a product.  
我认为很难把握对 AGI 的信念程度,因为每个人都能看到这样一个未来:模型会变得极其聪明,几乎什么都能做,在那种情况下,你其实不需要那么复杂的产品。  
**[52:31] Speaker A:** You can actually just have a text box again where you tell the model what you want, and it's so smart that it can add any tool or add any integration that it needs to get the job done.  
你实际上只需要一个文本框,告诉模型你想要什么,它足够聪明,可以添加任何需要的工具或集成来完成任务。  
**[52:43] Speaker A:** It knows when it's uncertain and can ask clarifying questions. It's kind of very easy to build the product for the super AGI strong model.  
它知道自己什么时候不确定,可以提出澄清性的问题。为超级 AGI 强模型构建产品其实是很容易的。  
**[52:53] Speaker A:** I think the hard thing is figuring out for the current model: How do you elicit the maximum capability? How do you help users get onto the  
我认为困难的是弄清楚对于当前的模型:如何激发出最大的能力?如何帮助用户进入  
**[53:07] Speaker A:** The golden path? How do you guide users to interact with the model's strengths and patch its weaknesses? This skill is pretty rare.  
黄金路径?如何引导用户与模型的优势互动,并弥补它的弱点?这种技能是相当罕见的。  
**[53:19] Speaker B:** And how do you build that skill? Is it just using each, basically understanding the limits of each model, having—you talked about taste—understanding, having taste into what the model maybe is capable of, what it's great and not great at, where it's changed?  
那你如何培养这种技能呢?是不是就是使用每个模型,基本上就是理解每个模型的局限性,拥有——你提到的品味——理解、拥有对模型可能能做什么、擅长什么不擅长什么、哪里发生了变化的品味?  
**[53:32] Speaker A:** I think it's spending a ton of time talking and using the model. One of the things I really like to do is to ask the model to introspect on its own behaviors. So sometimes when I notice that the model does something unexpected, like for example, there are situations where the model will make a front-end change and run tests but not actually use the UI. It's actually pretty useful to ask the model to reflect on why it did this. And sometimes they'll say that, hey, there was  
我认为是花大量时间与模型对话和使用模型。我真正喜欢做的一件事是让模型对自己的行为进行反思。所以有时当我注意到模型做了一些意外的事情,比如说,有些情况下模型会做前端更改并运行测试,但实际上不使用 UI。这时让模型反思它为什么这样做是非常有用的。有时它们会说,嘿,有  
**[54:04] Speaker A:** Like something confusing in the system prompt, or I didn't realize that the front-end verification was like part of this task, or hey, I delegated the verification to this sub-agent and the sub-agent didn't do the test and I didn't check its work.  
系统提示中有些令人困惑的地方,或者我没有意识到前端验证是这个任务的一部分,或者嘿,我把验证委托给了这个子代理,子代理没有做测试,而我没有检查它的工作。  
**[54:17] Speaker A:** A lot of times, just like being very curious about why the model made the decision that it did will show you what misled it so that you can fix the harness in order to close this gap.  
很多时候,对模型为什么做出这个决定保持好奇心,会让你看到是什么误导了它,这样你就可以修复框架来缩小这个差距。  
**[54:31] Speaker A:** The other thing that helps is to figure out who are the users who you trust the most to give you accurate feedback about the model.  
另一件有帮助的事情是找出你最信任的用户,他们能给你关于模型的准确反馈。  
**[54:42] Speaker A:** Usually there's like a handful of people who are much better than others at articulating what makes a specific model or model harness combination good.  
通常有少数几个人在阐述是什么让特定的模型或模型框架组合变得优秀方面,比其他人强得多。  
**[54:53] Speaker A:** And there's a lot of people who will give you feedback, but not everyone's feedback is as qualified.  
会有很多人给你反馈,但不是每个人的反馈都同样有价值。  
**[54:56] Speaker A:** And so finding a group of those like five people you trust is really important for getting very fast  
所以找到你信任的那五个人组成的小组,对于获得非常快速的  
**[55:03] Speaker A:** Feedback. I think the third thing that is useful but not everyone loves doing is building evals. You don't need to build hundreds of evals for them to be useful. Just building 10 great evals is important for helping the team quantify what the goal is and what their progress towards it is and what they're missing. And so I think evals is this underappreciated thing that more PMs, more engineers should be working on.  
反馈非常重要。我认为第三件有用但不是每个人都喜欢做的事情是构建评估。你不需要构建数百个评估才能让它们有用。只需构建 10 个优秀的评估,就能帮助团队量化目标是什么、他们朝着目标的进展如何、以及他们缺少什么。所以我认为评估是一个被低估的东西,更多的 PM、更多的工程师应该在这上面投入精力。  
**[55:33] Speaker B:** We've covered evals a bunch. There's this trend of just like that is the future of product management is writing evals because essentially it's what does success look like? Okay, cool. Let me actually concretely define it and then we'll know. How much of your time are you spending writing evals would you say?  
我们已经讨论了很多关于评估的内容。有一个趋势就是,这就是产品管理的未来——编写评估,因为本质上就是成功是什么样子?好的,让我实际具体地定义它,然后我们就会知道。你会说你有多少时间花在编写评估上?  
**[55:46] Speaker A:** I think the importance of evals varies a bit based on the feature that you're working on and or like what the problem you're trying to solve is. So there are a lot of folks on our team who do spend  
我认为评估的重要性会根据你正在开发的功能或者你试图解决的问题而有所不同。所以我们团队有很多人确实花费  
**[55:58] Speaker A:** A lot of time working on eval. Have a small pod of folks who collaborate very closely with research to more precisely understand our code behaviors and what the largest areas of improvement are and trying to measure those pretty concretely.  
大量时间在评估上。有一小组人与研究团队非常紧密地合作,更精确地理解我们的代码行为以及最大的改进领域在哪里,并试图相当具体地衡量这些。  
**[56:15] Speaker A:** I personally jump into evals when there's a feature that I think needs a bit more product definition and often the output of this is okay here are like five evals that I made, this is how you run them, these are the ones that succeed and these are the ones that don't, and this is like the prompt that I've used to increase the success rate.  
我个人在有一个我认为需要更多产品定义的功能时会投入到评估中,通常输出是这样的:好的,这是我做的五个评估,这是如何运行它们的,这些是成功的,这些是不成功的,这是我用来提高成功率的提示。  
**[56:39] Speaker A:** It varies a lot though based on the exact feature. Not every feature needs it but I think features such as memory benefit a lot from this.  
但这在很大程度上取决于具体的功能。不是每个功能都需要评估,但我认为像记忆这样的功能从中受益很多。  
**[56:45] Speaker B:** Point you made about people being very good at evaluating models so interesting. It's almost like a human eval of just like okay they understand where it's spiking or it's maybe lacking. Is there  
你提到的那个观点很有意思——人们非常擅长评估模型。这几乎就像是一种人工评估,他们能理解模型在哪些方面表现突出,哪些方面可能有所欠缺。是否有  
**[56:57] Speaker A:** Anyone specific that you want to shout out that's very good at this?  
在这方面有没有特别想表扬的人?  
**[57:00] Speaker B:** Uh, two people who I think are incredible at this are, um, one Amanda who definitely molds Claude's character.  
嗯,我觉得有两个人在这方面特别出色,一位是 Amanda,她确实塑造了 Claude 的性格特质。  
**[57:11] Speaker B:** It's just like such a hard role because the task is so ambiguous.  
这真的是一个很难的角色,因为任务本身就很模糊。  
**[57:15] Speaker B:** Even coding is easier because you can verify the success, whereas crafting the character requires a very strong sense of conviction in who Claude should be.  
甚至编程都更容易一些,因为你可以验证成功与否,而塑造性格则需要对 Claude 应该是什么样子有非常强烈的信念。  
**[57:26] Speaker B:** And I think she has like an incredible ability to not only mold the character, but also to like articulate what the goals are, what the character, what's successful and what's not.  
我觉得她不仅有出色的能力去塑造这个性格,还能清晰地阐述目标是什么,这个性格是什么样的,什么算成功什么算不成功。  
**[57:39] Speaker B:** The other group of people who I really trust is just like the Claude Code team.  
另一群我非常信任的人就是 Claude Code 团队。  
**[57:44] Speaker B:** Um, so we often have team lunches and whenever there's a new model we're testing.  
我们经常有团队午餐,每当我们在测试新模型的时候。  
**[57:49] Speaker B:** One of the fastest ways for us to get feedback is to just like at these team lunches just like go to every single person and just be like, "Hey, what is your vibe on the model?"  
我们获取反馈最快的方式之一就是在这些团队午餐上,逐个问每个人:「嘿,你对这个模型的感觉怎么样?」  
**[57:59] Speaker B:** And oftentimes  
而且很多时候  
**[58:01] Speaker A:** We'll get feedback like, "Okay, this model is not fully explaining its thinking. It's too abrupt," or like, "Hey, this model just loves writing a ton of memories, but we're not sure if the memories are high quality or not," or like some people will notice that, "Okay, this model loves to test itself," which is great, or like, "This model isn't testing itself enough."  
我们会得到这样的反馈:「好吧,这个模型没有完全解释它的思考过程,太突兀了」,或者「嘿,这个模型就是喜欢写一大堆记忆,但我们不确定这些记忆质量高不高」,或者有些人会注意到「好的,这个模型喜欢自我测试」,这很好,或者「这个模型自我测试得不够」。  
**[58:24] Speaker A:** So that informs what data we look at to verify, okay, is this a larger pattern. So we have a ton of data, but it is very hard to extract insights, and so the feedback from this group helps us inform, okay, what are the hypotheses we want to test, and then we're able to extract data to test that.  
所以这会告诉我们应该看哪些数据来验证,好吧,这是不是一个更大的模式。我们有大量数据,但很难提取洞察,所以这个团队的反馈帮助我们明确,好的,我们想测试哪些假设,然后我们就能提取数据来测试。  
**[58:45] Speaker B:** This point you made about the character of Claude—I had Ben Mann on the podcast, co-founder, and he talked about this, just like the character, the constitution of Claude is such an important part of Claude, and I didn't realize until afterwards just like, like, people like—  
你提到的关于 Claude 性格的这一点——我采访过联合创始人 Ben Mann,他也谈到了这个,就是性格、Claude 的宪法是 Claude 非常重要的一部分,我直到后来才意识到,就像,人们——  
**[59:01] Speaker A:** With OpenClaude actually, one of the reasons people are sad is like the personality of Claude is, because Claude's personality is so good and fun and interesting unlike other models, and the way he put it is the personality is what makes Claude so good at so many things.  
对于 OpenClaude,人们感到难过的原因之一就是 Claude 的个性,因为 Claude 的个性非常好、有趣且引人入胜,不像其他模型,他的说法是,个性正是让 Claude 在很多方面如此出色的原因。  
**[59:18] Speaker A:** It feels like this trivial side thing. Okay, it's going to be funny and interesting and talk in a fun way, but it's like so core to the success of Claude.  
这感觉像是一个微不足道的附带事项。好吧,它会很有趣、很吸引人、用有趣的方式说话,但这其实是 Claude 成功的核心。  
**[59:26] Speaker A:** Is there anything you can share about just like what people may not understand about why the character, as you described, and the personality is so key?  
关于为什么性格,如你所说的,和个性如此关键,有什么可以分享的吗?人们可能不理解的?  
**[59:34] Speaker B:** When you reflect on everyone you've worked with, there's just some people where you're like, I really like their energy. Like, I really like their vibe.  
当你回想与你共事过的每个人,总有一些人让你觉得,我真的很喜欢他们的能量,很喜欢他们的氛围。  
**[59:42] Speaker B:** And when people think about Claude and Claude Code, this is one of the things that people bring up the most where they just really love that Claude is like it's  
当人们想到 Claude 和 Claude Code 时,这是人们最常提到的事情之一,他们真的很喜欢 Claude 的那种  
**[59:53] Speaker A:** It's like lighthearted and fun, but it also is extremely competent at your task.  
轻松有趣,但同时在你的任务上又极其胜任。  
**[59:59] Speaker A:** People really like that Claude's low ego.  
人们真的很喜欢 Claude 的低姿态。  
**[01:00:02] Speaker A:** And so if you tell it, hey, you did this thing wrong, it's like truly sorry. It's like, oh shoot, like, thanks for telling me. Like, let me fix it. Let's work together.  
所以如果你告诉它,嘿,你这件事做错了,它会真诚地道歉。它会说,哦糟糕,谢谢你告诉我,让我修正一下,我们一起来解决。  
**[01:00:09] Speaker A:** It's also very positive.  
它也非常积极正面。  
**[01:00:11] Speaker A:** So if you're feeling like, oh, this is like an insurmountable task, I don't know how to get started, Claude is like, okay, it's okay. These are like the steps that I think we should take. Like, do you want me to get started on it for you?  
所以如果你觉得,哦,这是一个难以逾越的任务,我不知道如何开始,Claude 会说,好的,没关系,这些是我认为我们应该采取的步骤,你想让我先开始吗?  
**[01:00:25] Speaker A:** I think part of what makes a great coworker is this positivity, this like bias towards action, this ability to give you like earnest feedback, not just agreeing with every single thing that you say.  
我认为成为一个优秀同事的部分原因就是这种积极性,这种行动导向的倾向,这种给你真诚反馈的能力,而不是对你说的每一件事都表示同意。  
**[01:00:40] Speaker A:** And so we try to imbue this into Claude because we think it makes it a lot more enjoyable to work with.  
所以我们试图将这些注入到 Claude 中,因为我们认为这会让与它一起工作变得更加愉快。  
**[01:00:45] Speaker B:** There's something I want to come back to. You talked about how when new models  
有件事我想回过头来聊聊。你提到每当新模型  
**[01:00:48] Speaker A:** Come out, you often have to kind of revisit things you've built. That's so interesting and so like frustrating, maybe just like, oh god damn it, we shipped this thing, now we have to rethink it.  
发布时,你们经常需要重新审视已经构建的功能。这真的很有意思,但可能也挺让人沮丧的,就像「天哪,我们刚发布了这个功能,现在又得重新思考了」。  
**[01:00:56] Speaker A:** Talk about just like how often you have to come back with a new model and you're like, okay, we have to redo this product that we launched a few months ago.  
聊聊你们多久需要因为新模型而回过头来,然后想「好吧,我们得重做几个月前发布的这个产品了」。  
**[01:01:03] Speaker B:** A lot of the changes that we make with a new model is removing features that are no longer needed.  
我们在新模型发布后做的很多改动,其实是移除那些不再需要的功能。  
**[01:01:10] Speaker B:** So a lot of times we add features to the product as a crutch for the model because it's not naturally doing it itself.  
很多时候我们给产品添加功能,是作为模型的「拐杖」,因为模型本身还做不到这些事。  
**[01:01:18] Speaker B:** So the classic example for this is a to-do list.  
这方面的经典例子就是待办清单功能。  
**[01:01:20] Speaker B:** When we first launched Claude Code, people would ask it to do these large refactors and Claude Code would say, "Okay, cool. I need to change these like 20 call sites," and it would go and change five of them and then stop.  
我们刚推出 Claude Code 时,人们会让它做大规模重构,Claude Code 会说「好的,我需要修改这 20 个调用点」,然后它会改完 5 个就停下来了。  
**[01:01:31] Speaker B:** And then we were like, "Okay, how do we like force it to remember to get every single one of them?"  
然后我们就想「好吧,怎么才能强制它记住要把每一个都改完呢?」  
**[01:01:37] Speaker A:** these 20?" And so Sid on our team was like, "Okay, what if we just think about what a human would do?  
这 20 个调用点?」于是我们团队的 Sid 说「要不我们想想人类会怎么做?  
**[01:01:42] Speaker A:** So a human would make a list of everything that they need to change.  
人类会列一个清单,把所有需要修改的地方都列出来。  
**[01:01:45] Speaker A:** Similar to how in VS Code you would look up all the call sites and it would be a list on the left side and you would go through them one by one and replace all.  
就像在 VS Code 里,你会查找所有调用点,它们会显示在左侧列表里,然后你逐个检查并全部替换。  
**[01:01:51] Speaker A:** How do we give this kind of tool to Claude?  
我们怎么给 Claude 提供这样的工具呢?  
**[01:01:54] Speaker A:** And so he added a to-do list and we found that with that Claude was actually able to fix all these 20 call sites.  
于是他添加了一个待办清单功能,我们发现有了它,Claude 确实能够修复所有这 20 个调用点。  
**[01:02:01] Speaker A:** But then with Opus 4 and later models we realized that we didn't need to force it to use this to-do list.  
但后来到了 Opus 4 和更新的模型,我们发现不需要强制它使用这个待办清单了。  
**[01:02:10] Speaker A:** It would naturally use it itself.  
它会自然而然地自己使用待办清单。  
**[01:02:12] Speaker A:** For the earlier models, we had to keep reminding it, hey, did you finish everything on the to-do list?  
对于早期的模型,我们必须不断提醒它「嘿,你完成待办清单上的所有事项了吗?」  
**[01:02:15] Speaker A:** You can't finish until you're done with everything on the to-do list.  
「在完成清单上的所有事项之前,你不能结束任务。」  
**[01:02:17] Speaker A:** And for the later models, without prompting, it just naturally thinks to do everything on the to-do list.  
而对于后来的模型,不用提示,它就会自然地想到要完成待办清单上的所有事项。  
**[01:02:22] Speaker A:** These days, the to-do  
现在,待办清单  
**[01:02:26] Speaker A:** List is still nice to have as like a user, because then you can more clearly see what Claude is working on.  
作为用户体验来说还是不错的,因为你可以更清楚地看到 Claude 正在做什么。  
**[01:02:33] Speaker A:** But honestly, it's such a deemphasized part of the product right now that the model may use it, the model may not use it. It's like really not necessary for it to make thorough changes anymore.  
但说实话,它现在在产品中的重要性已经大大降低了,模型可能会用它,也可能不用。它对于模型做出全面的修改已经不是必需的了。  
**[01:02:44] Speaker B:** I forget who said this on the podcast that the model will eat your harness for breakfast. And what I'm hearing here is essentially you remove things over time that you've had to add on top of the model where it was not operating the way you wanted. And essentially as the models get smarter, it just becomes simpler and simpler for it to do the thing you want it to do.  
我忘了是谁在播客里说过「模型会把你的脚手架当早餐吃掉」。我现在听到的本质上就是,你们会逐渐移除那些为了让模型按预期工作而不得不添加的东西。随着模型变得越来越智能,让它做你想做的事情就变得越来越简单。  
**[01:03:04] Speaker A:** Yeah. We can remove a lot of prompting interventions every time the model gets smarter. And we actually do this every time we launch a model. We read through the entire system prompt and we reflect on, okay, for each of these sections, does the model really  
是的。每次模型变得更智能,我们都能移除很多提示干预。实际上每次发布新模型时我们都会这样做。我们会通读整个系统提示词,然后反思「好,对于每一个部分,模型真的  
**[01:03:18] Speaker A:** Need this reminder anymore? And if not, we'll remove it.  
还需要这个提醒吗?」如果不需要,我们就会移除它。  
**[01:03:20] Speaker A:** The most exciting thing that new models unlock though is just like entirely new features.  
不过新模型解锁的最令人兴奋的事情,还是那些全新的功能。  
**[01:03:27] Speaker A:** So there's a lot of features that we've been testing out with prior models and the accuracy wasn't high enough for us to want to launch them.  
我们一直在用之前的模型测试很多功能,但准确率不够高,所以没有推出。  
**[01:03:33] Speaker A:** And so one example of this is code review.  
代码审查就是一个例子。  
**[01:03:36] Speaker A:** We tried to build a code review product a few times and we've launched like simpler versions of code review which is the slash code review command in the past and it was only with the most recent models that we felt like okay this code review is so good that our engineering team relies on this code review to pass before we merge PRs and we found that this was we've always dreamed of Quad being able to be a reliable code reviewer that can actually that we can like confidently feel catches the majority of bugs.  
我们尝试过好几次构建代码审查产品,之前推出过一些简化版本,比如 slash code review 命令。直到最新的模型出现,我们才觉得这个代码审查足够好,好到我们工程团队在合并 PR 之前必须要通过这个审查。我们一直梦想 Quad 能成为可靠的代码审查员,能真正捕捉到大部分 bug,让我们有信心依赖它。  
**[01:04:12] Speaker A:** And it was only with like Opus 4.5 and 4.6 and Sonnet 4.6 that we felt like okay we are now able to like run multiple  
直到 Opus 4.5、4.6 和 Sonnet 4.6 出现,我们才觉得现在可以同时运行多个  
**[01:04:21] Speaker A:** Code review agents simultaneously traverse the entirety of the codebase and synthesize a set of real issues that an engineer needs to address before merge.  
代码审查 agent,让它们遍历整个代码库,然后综合出工程师在合并前需要解决的真实问题。  
**[01:04:33] Speaker A:** And so this is a new capability that the newest models have unlocked.  
这是最新模型解锁的新能力。  
**[01:04:39] Speaker B:** This is another trend that is very common on this podcast: build something that will possibly be possible in the next six months, be kind of at the edge of what's working, and then it'll catch up and then it'll be an amazing product and you'll be ahead of everyone.  
这是这个播客里很常见的一个趋势:构建一个未来六个月可能实现的东西,处在刚刚能用的边缘,然后等技术跟上来,它就会变成一个很棒的产品,而你已经领先所有人了。  
**[01:04:52] Speaker A:** Yeah, exactly. It's pretty important to build products that don't necessarily work yet so that you know what is missing for this product to work, and then with the newest model you can just swap it in to the prototype you've already made and see, okay, does this new model close that gap.  
没错。构建那些暂时还不能用的产品很重要,这样你就知道这个产品要能用还缺什么,然后等最新模型出来,你就可以直接把它换到你已经做好的原型里,看看这个新模型能不能填补那个差距。  
**[01:05:12] Speaker B:** How much are you able to speak to just kind of where things are going with  
你能透露多少关于  
**[01:05:15] Speaker A:** Claude and Cowork as kind of the vision of it? I imagine you don't want to give away too much about the goal, but it feels like there's all these awesome features being added on top—dispatch control from phone and all these mobile app, all these things. What's kind of just like a way to understand the vision for all these things long term?  
Claude 和 Cowork 未来发展方向的愿景?我猜你不想透露太多关于目标的信息,但感觉有很多很棒的功能在不断加入——手机调度控制、移动应用等等。怎么理解这些东西的长期愿景?  
**[01:05:32] Speaker B:** We think about this in terms of building blocks. So for both Claude Code and Cowork, the core building block is making individual tasks successful. So you want to produce some output, you give it a clear prompt description—is it able to consistently produce acceptable output that you're able to either merge or share with your colleagues or external audience?  
我们从构建模块的角度来思考这个问题。对于 Claude Code 和 Cowork 来说,核心构建模块是让单个任务成功。你想产出某个结果,给它一个清晰的提示描述——它能不能持续产出可接受的输出,让你能够合并或者分享给同事或外部受众?  
**[01:05:54] Speaker B:** So the task is the core building block. As the models get smarter, the task success rate gets a lot higher. And then we see people moving towards doing multiple tasks at the same time. So multi-coding was this  
所以任务是核心构建模块。随着模型变得更智能,任务成功率会大幅提高。然后我们看到人们开始同时做多个任务。所以 multi-coding 在  
**[01:06:06] Speaker A:** Big thing in towards the end of 2024 and it's only increased since then.  
2024 年底成为一个大趋势,之后还在持续增长。  
**[01:06:11] Speaker A:** And so we see this as okay, great, one task works and now you can do like six tasks at a time.  
我们的理解是,好,一个任务能成功了,现在你可以同时做六个任务。  
**[01:06:17] Speaker A:** As the models get even smarter, the way that we are extrapolating this is okay, next maybe you're going to run like 50 Claudes at a time or hundreds of Claudes at a time.  
随着模型变得更智能,我们推断的方式是,接下来你可能会同时运行 50 个 Claude,或者几百个 Claude。  
**[01:06:28] Speaker A:** And so what is the infrastructure we need to build to enable that?  
那我们需要构建什么样的基础设施来实现这一点?  
**[01:06:32] Speaker A:** At that point you're probably not going to run everything locally on your machine anymore.  
到那时候你可能不会再在本地机器上运行所有东西了。  
**[01:06:35] Speaker A:** There's just like not enough RAM to do it.  
根本没有足够的内存来做这件事。  
**[01:06:39] Speaker A:** And so we're thinking about how do we make it easier for you to manage all these?  
所以我们在思考如何让你更容易管理所有这些?  
**[01:06:44] Speaker A:** These will probably run remotely.  
这些可能会在远程运行。  
**[01:06:46] Speaker A:** How do we build the interface so that you as a human know which tasks you need to look into?  
我们如何构建界面,让你作为人类知道哪些任务需要关注?  
**[01:06:51] Speaker A:** How do we make sure that the agent is fully verifying work so that when you look at a task and it says it's done, you can very quickly verify and fully trust that it is done to your spec?  
我们如何确保 agent 完全验证工作,这样当你看到一个任务显示完成时,你可以很快验证并完全相信它确实按照你的要求完成了?  
**[01:07:03] Speaker A:** And how do we make  
我们如何确保  
**[01:07:05] Speaker A:** Sure that this process is self-improving so that when you do see a task that isn't done to your liking, you can give it feedback and the model will know for every future run to incorporate that feedback so it never makes that mistake again.  
这个过程是自我改进的,这样当你看到一个任务没有按你的要求完成时,你可以给它反馈,模型会在以后的每次运行中都融入这个反馈,这样它就不会再犯同样的错误。  
**[01:07:18] Speaker A:** So this is the progression that we're bringing our users along for.  
这就是我们带领用户经历的进程。  
**[01:07:23] Speaker B:** There's a lot of people listening, a lot of product managers, a lot of maybe founders, a lot of other cross-functional folks listening. There's a lot of worry about just their role, just the future of their careers.  
有很多人在听这期节目,其中有产品经理、创业者,还有其他跨职能岗位的从业者。大家都很担心自己的职位和职业前景。  
**[01:07:36] Speaker B:** What advice would you have for just people to not just survive this transition to this very AI-driven world, but to be really successful, to essentially just to thrive in this future?  
对于如何在这个AI驱动的世界中不仅生存下来,还能真正取得成功、蓬勃发展,你有什么建议?  
**[01:07:45] Speaker B:** What are just like things people need to hear, need to be doing?  
有哪些是大家需要听到的、需要去做的事情?  
**[01:07:49] Speaker A:** I think AI gives everybody a ton more leverage than they used to. And so I would push you towards anytime you  
我认为AI给每个人带来了比以前多得多的杠杆效应。所以我建议你,每当意识到  
**[01:07:57] Speaker A:** Realize that you're doing some manual task multiple times, think about how you can use Claude Code, Copilot or other AI tools to automate that for you.  
自己在重复做某个手动任务时,就想想如何用Claude Code、Copilot或其他AI工具来自动化完成。  
**[01:08:04] Speaker A:** Most people have like creative parts of their job that they absolutely love and then like tedious parts of their job that they really hate doing.  
大多数人的工作中都有他们特别喜欢的创造性部分,也有他们真的很讨厌做的繁琐部分。  
**[01:08:14] Speaker A:** I think the beauty of AI is that it can do those tedious parts for you. It can learn from every time that you've done that manual task and generalize and then run it automatically so that you can focus on the creative parts and that means you can do a lot more than you used to be able to do.  
AI的美妙之处在于它可以替你完成那些繁琐的部分。它能从你每次手动完成任务中学习,然后归纳总结并自动运行,这样你就能专注于创造性的工作,意味着你能完成比以前多得多的事情。  
**[01:08:31] Speaker A:** So I think my like immediate push for people is figure out the repetitive parts that you can pass to Claude. Iterate on those automations until the success rate is very high and then focus on okay what more can you be doing for your team for your product for your company that like people haven't had the bandwidth to pick up so far or like what is that like pet project that you always  
所以我的直接建议是:找出那些可以交给Claude的重复性工作,不断迭代这些自动化流程直到成功率非常高,然后思考你还能为团队、产品或公司做些什么——那些之前大家一直没有精力去做的事,或者你一直想做的那个  
**[01:08:55] Speaker A:** thought the company should do that like you've never had bandwidth to do. If AI can take care of the grunt work, then you have this extra 20% time now that you might not have had before.  
宠物项目,那些你一直觉得公司应该做但从来没有时间去做的事。如果AI能处理那些苦力活,你现在就有了以前没有的额外20%的时间。  
**[01:09:08] Speaker A:** So my push is to lean into these tools, hand off the work that you're not excited to do, figure out how it can accelerate you, and then as a result, you'll be able to do so much more.  
所以我的建议是:积极使用这些工具,把你不想做的工作交出去,搞清楚它如何能加速你的工作,这样你就能完成多得多的事情。  
**[01:09:19] Speaker B:** Something core to what you just shared, which I fully agree with, is find problems to solve with AI. There's all this potential with all these tools can do. Some of the hard—like for a lot of people the hardest part is just like what should I actually do, and what you're saying here is just pay attention to things that you are doing constantly you can automate, pay attention to just like ideas that have been floating around that you haven't had time to do. It's basically like solve a problem for yourself is kind of the core advice there.  
你刚才分享的核心观点我完全同意,就是:用AI找问题来解决。这些工具有很多潜力可以做很多事。对很多人来说最难的部分就是——我到底应该做什么?而你这里说的就是:留意那些你经常在做的、可以自动化的事情,留意那些一直在脑海中浮现但没时间做的想法。核心建议其实就是为自己解决问题。  
**[01:09:45] Speaker A:** Exactly. I would also push listeners towards focusing on bringing your automations from "okay, this is a cool concept" to like "hey, this actually works 100% of the time." Like sometimes I see users trying to automate something, getting it to like 90-95% accuracy and then giving up on it.  
完全正确。我还建议听众们专注于把自动化从「好的,这个概念不错」提升到「嘿,这个真的能100%正常工作」。有时我看到用户尝试自动化某件事,做到90-95%的准确率就放弃了。  
**[01:10:05] Speaker A:** And if an automation doesn't work 100% of the time, it's not really an automation.  
如果一个自动化不能100%正常工作,它就不算真正的自动化。  
**[01:10:09] Speaker A:** And that last 5 to 10% does take more time. Also, building the automation is often a lot slower than you doing it yourself.  
而最后那5%到10%确实需要更多时间。而且,构建自动化往往比你自己做要慢得多。  
**[01:10:18] Speaker A:** I would encourage listeners to put in that time to scope some automation that you really want to get to 100%. Put in the elbow grease to teach it your preferences, to like give it feedback so that it can improve its skill so that it can get to that 100%.  
我鼓励听众投入时间去规划一个你真正想做到100%的自动化。下功夫去教它你的偏好,给它反馈让它能提升技能,达到那个100%。  
**[01:10:36] Speaker A:** And then like really then you'll be able to rely on it. There's just not much value in a 95% automation.  
然后你就真的能依赖它了。95%的自动化其实没什么价值。  
**[01:10:44] Speaker B:** I am super guilty of that.  
我在这方面特别有罪。  
**[01:10:46] Speaker A:** Really good advice for me.  
对我来说这是个很好的建议。  
**[01:10:48] Speaker B:** I am guilty of this too. I've been teaching it, I've been teaching Coworker to try to get me to inbox zero for Gmail, and it has not been—it has been very time consuming and it is definitely not there, as you probably realized.  
我也有这个问题。我一直在教Coworker帮我把Gmail清理到收件箱零邮件,但一直没有——这个过程非常耗时,而且肯定还没达到目标,你可能也意识到了。  
**[01:11:02] Speaker A:** Yeah, funny enough that's exactly where my mind goes. I have this workflow I set up where every email I get, it looks for things that are spammy, which is just like all these like, "Hey, can I come on your podcast?" or "What about this one?" Like all these things I'm just like, I don't have time for these sorts of things. And I have it categorized into a folder called spammy. And it's just like it's 95% great, but then there's like, oh wow, I missed an email because it went in there. So this is a good push for me to like, I'm going to work on this. I'm going to get it to perfect.  
是的,有意思的是我脑海中想到的正是这个。我设置了一个工作流,每封收到的邮件它都会查找垃圾内容,就是那些「嘿,我能上你的播客吗?」或者「这个怎么样?」之类的,我根本没时间处理这些事。我让它分类到一个叫spammy的文件夹里。它95%的时候都很好,但有时会出现,哇,我错过了一封邮件因为它被放进去了。所以这对我是个很好的推动,我要继续优化这个,把它做到完美。  
**[01:11:29] Speaker B:** Yeah. We also are working on making the flow for customizing these commands a...  
是的。我们也在努力让自定义这些命令的流程变得  
**[01:11:34] Speaker A:** A lot easier because right now I think you have to know too many concepts. You have to know to define a skill. You have to know to use this skill and give it feedback. And then you have to know to tell Cowork to update the skill based on all the feedback that you gave. And then you also have to know where to read the skill to make sure that the feedback was incorporated the way that you want.  
容易得多,因为现在我觉得你需要了解太多概念。你得知道如何定义一个技能,知道如何使用这个技能并给它反馈,然后你还得知道告诉Cowork根据你给的所有反馈来更新技能。然后你还得知道去哪里查看技能,确保反馈按你想要的方式被整合进去了。  
**[01:11:51] Speaker A:** It's also our job to make this flow really seamless so that it doesn't feel painful to do.  
让这个流程真正无缝衔接、不让人感到痛苦,这也是我们的工作。  
**[01:11:57] Speaker B:** Amazing. Is there anything else, Cat, you wanted to share? Anything else you wanted to leave listeners with? Anything you wanted to double down on that we haven't already touched on before we get to our very exciting lightning round?  
太好了。Cat,还有什么想分享的吗?在我们进入非常激动人心的快问快答环节之前,有什么想留给听众的?有什么我们还没谈到、你想强调的吗?  
**[01:12:08] Speaker A:** I see a lot of people playing around with AI and building prototype apps and tinkering with building workflows. I would really push people towards building apps that you're actually using every single day because I think only  
我看到很多人在玩AI、构建原型应用、尝试构建工作流。我真的建议大家去构建那些你每天实际在用的应用,因为我认为只有  
**[01:12:23] Speaker A:** Through that usage, are you actually getting the value? Like if you build a prototype app that isn't helping you get more done, then the AI isn't really adding value to your—  
通过这种使用方式,你真的获得价值了吗?比如你做了一个原型应用,但它并没有帮你提高效率,那 AI 其实并没有给你带来真正的价值——  
**[01:12:37] Speaker B:** To your day.  
对你的日常工作来说。  
**[01:12:38] Speaker A:** And there's only so much you learn from that when it's like, okay, I just one-shotted something. Oh, that's cool. And then you never come back to it. Like you're not learning a lot—  
而且从这种体验中你能学到的也很有限,就是那种「好,我一次性搞定了某个东西,哦挺酷的」,然后就再也不碰了。这样你其实学不到什么——  
**[01:12:45] Speaker B:** And you're not getting much leverage from it—  
你也没有从中获得多少杠杆效应——  
**[01:12:47] Speaker A:** And actual leverage. Yeah, that's such a good point.  
真正的杠杆效应。对,这个观点太好了。  
**[01:12:49] Speaker B:** I also think there's a lot of people who spend a lot of time customizing their workflow. So there's like two ends of the spectrum. One is people who never customize or never build automations, but there's this polar opposite end of people who obsess around customizing their tool, like adding a ton of skills and—  
我还觉得有很多人花大量时间定制他们的工作流。这就像光谱的两端,一端是从不定制、从不构建自动化的人,另一端则是那些痴迷于定制工具的人,比如添加大量技能和——  
**[01:13:07] Speaker A:** MCPs and these like workflow improvements, and I think sometimes that can even distract from your core goal of like launching some product or building some feature.  
MCP 以及各种工作流改进,我觉得有时候这甚至会让你偏离核心目标,比如发布某个产品或开发某个功能。  
**[01:13:17] Speaker A:** I think there's a lot of fun in customizing and we definitely want to make our products very hackable so that you can make it work really well for you, but there is a limit to how much it's useful.  
定制确实很有趣,我们也确实希望让产品非常可定制,这样你可以让它真正适合自己,但这种定制是有限度的。  
**[01:13:29] Speaker A:** And I think there's a camp of people who maybe spend so much time customizing that they're like not sleeping and not doing the like core task that they originally set out to do.  
我觉得有一群人可能在定制上花了太多时间,以至于他们都不睡觉了,也没在做他们最初想做的核心任务。  
**[01:13:41] Speaker B:** I see a lot of that on Twitter, just like look at my setup, it's out of control, it's so optimized. Then what are you, what are you actually building? No, but my setup is so awesome, like it gets so much done.  
我在 Twitter 上经常看到这种情况,就是「看我的配置,太疯狂了,优化得太好了」。然后你实际在做什么呢?「没啊,但我的配置超棒,能完成好多事」。  
**[01:13:52] Speaker A:** I think the simple setups actually work better.  
我觉得简单的配置其实效果更好。  
**[01:13:56] Speaker B:** It's a power-up, getting to take level up a little bit.  
这就像一个增益道具,让你稍微升个级。  
**[01:13:58] Speaker A:** Yeah. Yeah.  
对,对。  
**[01:13:59] Speaker B:** There's this Karpathy tweet that just came out yesterday where he talked about this divide that's interesting between people that tried ChatGPT and Claude back in the day. It was like okay and they're like nah this is terrible and they kind of gave up on like what AI could do for them and they're just like so cynical of like no way it's not actually that big of a deal and then there's people that are using it to code essentially who see the full intense power of it and how good it is and people on both sides don't understand the other side and why they like how much they see the world and so your advice is really good here just like actually use it for real things and see how good it actually has gotten.  
Karpathy 昨天发了条推文,讲了一个很有意思的分化现象,就是有些人当年试用 ChatGPT 和 Claude 时觉得「还行吧」,然后就觉得「这太糟糕了」,基本放弃了 AI 能为他们做什么,对此非常悲观,觉得「不可能,没那么厉害」;而另一群人用它来写代码,他们看到了它全部的强大力量,知道它有多好用。两边的人都不理解对方,不明白对方为什么会那样看待世界。所以你的建议真的很好,就是真正用它做实际的事情,看看它现在到底有多强。  
**[01:14:38] Speaker A:** Yeah I think the big shift is that the 2024 generation of products were chat-based and the current code generation of products is action-based.  
对,我觉得最大的转变是 2024 年那一代产品是基于对话的,而现在这一代代码生成产品是基于行动的。  
**[01:14:50] Speaker A:** The big aha moment people have is when Claude can just do things on your behalf. It is an amazing feeling to know that the agent is capable of doing so much more than telling you what to do. The agent can actually just do it itself. And when people feel that, I think that's the eye-opening moment.  
人们的顿悟时刻就是当 Claude 能代表你直接做事的时候。知道这个 agent 能做的远不止告诉你该做什么,它实际上可以自己去做,这种感觉太棒了。当人们体验到这一点时,我觉得那就是开眼的时刻。  
**[01:15:10] Speaker B:** Shout out to the Chrome extension, the Claude Chrome extension, which you can just watch it doing stuff and you'd be like, "Fill out this form for me," and it's like, "All right, here I go."  
要夸一下 Chrome 扩展,就是 Claude 的 Chrome 扩展,你可以看着它做事,你说「帮我填这个表单」,它就「好的,我来了」。  
**[01:15:18] Speaker A:** Exactly.  
没错。  
**[01:15:19] Speaker B:** Okay. Anything else before we get to our very exciting lightning round?  
好的。在进入我们非常激动人心的快问快答环节之前,还有什么要补充的吗?  
**[01:15:22] Speaker A:** No, let's do it.  
没有了,开始吧。  
**[01:15:24] Speaker B:** Let's do it. Kat, I've got five questions for you. Welcome to the lightning round. There's this animation that plays. I have to make sure to say it. Are you ready?  
开始吧。Kat,我有五个问题要问你。欢迎来到快问快答环节。这里会播放一个动画,我得说一下。准备好了吗?  
**[01:15:32] Speaker A:** I'm ready.  
准备好了。  
**[01:15:34] Speaker B:** First question, what are two or three  
第一个问题,有哪两三本  
**[01:15:36] Speaker A:** Books that you find yourself recommending most to other people?  
你最常推荐给别人的书?  
**[01:15:38] Speaker B:** I really like How Asia Works. It's a story about economic development and what are the policies and governments that make long-lasting successful economies.  
我很喜欢《亚洲是如何运作的》(How Asia Works)。这本书讲的是经济发展的故事,以及哪些政策和政府能造就长久成功的经济体。  
**[01:15:51] Speaker B:** The other book that I'm really into is The Technology Trap. So this is actually about the past few technology revolutions, so the Industrial Revolution and the computer revolution, and how this has affected workers.  
我很喜欢的另一本书是《技术陷阱》。这本书讲的是过去几次技术革命,比如工业革命和计算机革命,以及这些革命如何影响了工人阶层。  
**[01:16:06] Speaker B:** The reason that I really like this is because I think there's a lot we can learn from history to make sure that this transition goes well.  
我特别喜欢这本书的原因是,我认为我们可以从历史中学到很多东西,以确保这次技术转型能够顺利进行。  
**[01:16:17] Speaker B:** And maybe on a fun note, I really like The Paper Menagerie. It's just a book of short stories about coming of age and AI and self-discovery.  
说点轻松的,我很喜欢《纸动物园》。这是一本短篇小说集,讲述了关于成长、人工智能和自我发现的故事。  
**[01:16:29] Speaker A:** Favorite recent movie or TV show you have really enjoyed?  
你最近很喜欢的电影或电视剧是什么?  
**[01:16:34] Speaker B:** I really like Drive to Survive. There's no deeper meaning to it. I just—  
我很喜欢《极速求生》。没什么深层含义,就是——  
**[01:16:41] Speaker A:** There's just something very satisfying about people being so obsessed with like a singular engineering goal and just like the purity of their pursuit. And I also really love Free Solo, which is about Alex Honnold climbing El Capitan without a harness.  
看到人们如此痴迷于一个单一的工程目标,以及他们追求的纯粹性,就是让人很满足。我也很喜欢《徒手攀岩》,讲的是 Alex Honnold 在没有安全绳的情况下攀登 El Capitan。  
**[01:16:58] Speaker A:** And I think similarly, it's just such a pure achievement to be able to climb this extremely challenging, dangerous route and to be able to have the mental focus to do it knowing that if you make a single mistake, you die.  
我觉得同样地,能够攀登这条极具挑战性和危险性的路线,并且能够保持精神高度集中,明知道一个失误就会丧命,这是一种非常纯粹的成就。  
**[01:17:17] Speaker B:** It's insane. Yeah, that movie is out of control. And it's interesting how these relate in some way to the work you do.  
太疯狂了。是的,那部电影简直不可思议。有意思的是,这些在某种程度上都和你的工作有关。  
**[01:17:22] Speaker A:** I actually am a rock climber. I first watched Free Solo before I climbed rocks and so I thought it was impressive. I didn't understand how impressive it was.  
我其实是个攀岩爱好者。我第一次看《徒手攀岩》的时候还没开始攀岩,所以当时觉得很厉害,但不明白到底有多厉害。  
**[01:17:31] Speaker A:** It's one of the rare movies where like the more you know about it, the more you're blown away by how insane this is. Like the kinds of moves he's doing on the wall are things  
这是少数几部你越了解就越被震撼到的电影之一。他在岩壁上做的那些动作  
**[01:17:40] Speaker A:** That like I don't think I will ever be able to do in my lifetime if it were set in a gym like one foot off the ground.  
我觉得即使是在攀岩馆里,离地面只有一英尺高,我这辈子可能都做不到。  
**[01:17:47] Speaker B:** With a rope.  
还有安全绳的情况下。  
**[01:17:48] Speaker A:** With a rope.  
对,有安全绳的情况下。  
**[01:17:50] Speaker B:** Did you see the documentary on that other guy, the younger one that went on like ice mountain?  
你看过那个关于另一个人的纪录片吗,那个年轻人去爬冰山的?  
**[01:17:54] Speaker A:** I did. That one was very sad.  
看过。那部很悲伤。  
**[01:17:56] Speaker B:** But that was wild. Okay, uh, favorite product you recently discovered that you really love?  
但那部也很震撼。好的,呃,你最近发现的、真心喜欢的产品是什么?  
**[01:17:59] Speaker A:** The product that has most changed my life outside of cloud products is probably Waymo.  
除了云产品之外,对我生活改变最大的产品可能是 Waymo。  
**[01:18:07] Speaker A:** Like I'm a diehard Waymo user. I use it twice a day, get to and from work.  
我是 Waymo 的铁杆用户。我每天用两次,上下班都坐。  
**[01:18:12] Speaker A:** So the two things that I really like about it are one, I don't feel bad if a Waymo is waiting for me.  
我特别喜欢它的两点,第一,如果 Waymo 在等我,我不会感到内疚。  
**[01:18:17] Speaker A:** And so I feel less pressure to be right at the curbside the moment it arrives.  
所以我不会有那种压力,觉得必须在它到达的那一刻就站在路边。  
**[01:18:25] Speaker A:** And the second thing is I feel like it lets me be a bit more productive.  
第二点是,我觉得它让我的效率更高一些。  
**[01:18:28] Speaker A:** When I'm in the car with another human, I typically try not to do any work.  
当车里有另一个人的时候,我通常会尽量不工作。  
**[01:18:35] Speaker A:** calls. I feel a little rude if I'm like on my laptop the whole time. But one thing I really appreciate about the Waymo is I can call into a work call. I'm not worried about someone overhearing me. I'm not worried about, hey, is this like rude? Am I talking too loud? Do I need to ask someone to like change the music? And so this has been like I feel like this has given me back like 30 minutes every day.  
打电话。如果我一直在用笔记本电脑,我会觉得有点不礼貌。但 Waymo 让我很欣赏的一点是,我可以接入工作电话。我不用担心有人会听到我说话。我不用担心,嘿,这样是不是不礼貌?我说话声音是不是太大了?我需不需要让司机换个音乐?所以这个让我觉得每天多出了 30 分钟。  
**[01:18:55] Speaker B:** All these second order effects of technology. It's so interesting.  
技术带来的这些二阶效应,真的很有意思。  
**[01:18:59] Speaker A:** Yeah. I always thought Waymo needed to be priced lower than Uber and Lyft to succeed, but actually I'm like very happy to pay a 2x premium for it.  
是的。我一直以为 Waymo 需要比 Uber 和 Lyft 便宜才能成功,但实际上我很愿意为它支付两倍的价格。  
**[01:19:06] Speaker B:** I love Waymo. It's just like once you see it, you're just like, "Wow, this is insane." And then you get used to it. Like you get in there, you're like, "This is crazy." And then you forget about it.  
我很喜欢 Waymo。就是你第一次看到它的时候会觉得「哇,太疯狂了」。然后你就习惯了。你坐进去的时候还在想「这太不可思议了」,但很快就忘了这回事。  
**[01:19:17] Speaker A:** Totally. And I think it's also changed the vernacular. Like a lot of people at  
确实是这样。而且我觉得它还改变了大家的说话方式。比如  
**[01:19:21] Speaker A:** Anthropic love Waymo. And I think in the past you would be like, "Hey, like let's call like blah blah ride share app." And now like everyone's just like, "Okay, is the Waymo here?"  
Anthropic 有很多人都喜欢 Waymo。以前大家会说「嘿,我们叫个某某打车软件吧」,现在大家都直接说「Waymo 到了吗?」  
**[01:19:30] Speaker B:** Okay, two more questions. Do you have a favorite life motto that you often come back to in work or in life?  
好的,还有两个问题。你有没有一句最喜欢的人生格言,在工作或生活中经常想起的?  
**[01:19:35] Speaker A:** Just do things.  
就是去做事情。  
**[01:19:37] Speaker B:** That's right.  
没错。  
**[01:19:38] Speaker A:** I think there's a lot of value in first principles thinking and if you know what you're optimizing for and you have strong first principles, then you can normally deduce what the right course of action is and be able to clearly articulate that to all the stakeholders and then you should just do it. I think jobs are fake. If you understand the constraints, you can figure out what you can do and then just try to do it quickly, learn from the mistakes and apologize or fix them if you did something wrong.  
我觉得第一性原理思维很有价值,如果你知道自己在优化什么,并且有坚实的第一性原理,那你通常就能推导出正确的行动方案,并能向所有利益相关方清楚地阐述,然后你就应该去做。我认为工作角色其实是虚的。如果你理解了限制条件,你就能想清楚自己能做什么,然后快速去做,从错误中学习,如果做错了就道歉或修正。  
**[01:20:08] Speaker A:** You could just do things, whoever said that.  
你就是可以去做事情,不管是谁说的这句话。  
**[01:20:10] Speaker B:** I think it's liberating actually to tell people this. I think in a lot of companies, roles are very strictly defined like, okay, this is what the PM does, this is what the designer does, this is what the engineer does, and then even team scopes are very rigidly defined. So, hey, this corner of the codebase we touch and this corner we're not allowed to touch. And I think what 'just do things' lets people do is they feel empowered to make these decisions, empowered to operate across team boundaries just to get something done.  
我觉得告诉人们这一点其实很解放人。在很多公司里,角色定义得非常严格,比如这是 PM 做的,这是设计师做的,这是工程师做的,甚至团队范围也划分得很死板。所以就变成了,嘿,代码库的这个角落我们负责,那个角落我们不能碰。而我觉得「just do things」让人们感到有权力做这些决定,有权力跨越团队边界去把事情做成。  
**[01:20:38] Speaker A:** That feels like a big important skill to be good at. People call it agency. Just like, do the things.  
这感觉是一项很重要的技能。人们称之为主动性。就是,去做那些事情。  
**[01:20:46] Speaker B:** Bias towards action. All these ways of describing just like, don't wait for permission.  
偏向行动。所有这些说法都是在描述同一件事:不要等待许可。  
**[01:20:50] Speaker A:** Yeah. I think this is my favorite reason to work at a startup at some point in your career.  
对。我觉得这是我最喜欢在职业生涯中某个阶段去创业公司工作的原因。  
**[01:20:54] Speaker A:** Your life, because like one thing that was like very life-changing for me was actually working at Scale when we were 20 people.  
在你的人生中,因为有一件事对我来说真的改变了人生,就是在 Scale 只有 20 个人的时候去那里工作。  
**[01:21:00] Speaker A:** And so there was just no process and we had like really big problems that we needed to solve.  
那时候完全没有流程,而我们有非常大的问题需要解决。  
**[01:21:04] Speaker A:** And it was like I really appreciate Alex and the rest of the team for like empowering me and the rest of the team to just like figure things out without any boundaries for what sales is supposed to do, what ops is supposed to do, what engineer is supposed to do.  
我真的很感激 Alex 和团队其他成员,他们赋予我和团队权力,让我们在没有任何边界的情况下去解决问题——不用管销售应该做什么、运营应该做什么、工程师应该做什么。  
**[01:21:18] Speaker A:** Just like you have all the tools at your disposal, you have some like ambitious hairy problem statement and you can do whatever you need to like get to a good solution.  
就是你手头有所有工具,面前有一个雄心勃勃的棘手问题,你可以做任何需要做的事情来找到好的解决方案。  
**[01:21:28] Speaker B:** Like you almost need that experience to build that skill to feel comfortable doing that because a lot of people, you know, they go through school or in college and it's all these like do the thing we tell you to do and then you will get a good grade.  
你几乎需要那种经历来培养这种技能,让自己习惯这样做,因为很多人在学校或大学里,都是「做我们让你做的事,然后你就能得到好成绩」。  
**[01:21:38] Speaker B:** And you have to kind of unlearn that of like, okay, I'm just  
你必须摆脱那种思维模式,转变成「好吧,我就是要  
**[01:21:41] Speaker A:** I'm going to do the thing that needs to be done, and even if people think it's dumb, I think it's the right thing to do.  
去做需要做的事情,即使别人觉得这很蠢,我也认为这是正确的事。」  
**[01:21:46] Speaker B:** Yeah, exactly.  
对,就是这样。  
**[01:21:47] Speaker A:** Okay, I actually have two more quick questions. Two more final questions. One is, when Claude thinks, there's all these—I don't know if you call them verbs. What's the term for these things?  
好,我其实还有两个快速问题。最后两个问题。一个是,当 Claude 在思考的时候,会有各种——我不知道该叫它们动词还是什么。这些东西叫什么?  
**[01:21:55] Speaker B:** Uh, thinking words.  
呃,思考词汇。  
**[01:21:56] Speaker A:** Thinking words. And interestingly, these all leaked in the source code. Do you have a favorite thinking word?  
思考词汇。有意思的是,这些都在源代码里泄露了。你有最喜欢的思考词汇吗?  
**[01:22:03] Speaker B:** I really like manifesting. It's also like the sticker that I have on my favorite.  
我特别喜欢 manifesting(显化)。这也是我最喜欢的贴纸上的词。  
**[01:22:10] Speaker A:** Clearly the winner. Okay, final question. Asked Boris this too. With AGI potentially arriving in our lifetime, when you don't potentially have to work, what are you going to do? What are you going to do with all your time?  
显然是赢家。好,最后一个问题。也问过 Boris 这个问题。如果 AGI 可能在我们有生之年到来,当你可能不需要工作的时候,你会做什么?你会怎么安排你的时间?  
**[01:22:23] Speaker B:** I think it will take a long time for AGI to diffuse across society. So, I—  
我认为 AGI 要在整个社会中普及还需要很长时间。所以,我——  
**[01:22:28] Speaker A:** I think the immediate thing is actually just like helping bring the world along.  
我觉得当下最重要的事情其实就是帮助整个世界跟上这个进程。  
**[01:22:32] Speaker A:** I think my like non-serious answer for after this happens is I'll probably just do a lot of rock climbing.  
我那个不太正经的答案是,等这一切发生之后,我可能就会去大量攀岩。  
**[01:22:38] Speaker A:** I'll probably just like live in some—I'll probably move to like Fontainebleau and just like live amongst 10,000 boulders and climb for a bit.  
我可能会搬到 Fontainebleau 那样的地方,就住在一万块巨石中间,爬一阵子。  
**[01:22:48] Speaker A:** There's also so many books I want to read that my goal is to be able to read one or two books a week and I'm currently at probably like 0.5.  
还有太多书我想读了,我的目标是能做到每周读一到两本书,而我现在大概只能读 0.5 本。  
**[01:23:02] Speaker A:** The backlog is pretty big. I think there's just like so much we can learn from history and so much that I don't understand as well as I would love to.  
积压的书单很长。我觉得我们能从历史中学到太多东西,而有太多东西我还没有理解到我希望达到的程度。  
**[01:23:09] Speaker A:** Like I don't know anything about physics or like robotics or like any hardware or like aerospace or there's just so many interesting topics.  
比如我对物理学、机器人学、任何硬件或者航空航天都一无所知,有太多有趣的话题了。  
**[01:23:20] Speaker A:** So I'm excited to learn even knowing that the AI will already know it.  
所以我很期待去学习这些,即使知道 AI 已经掌握了这些知识。  
**[01:23:26] Speaker B:** Cat, this was amazing. You're awesome. Two follow questions.  
Cat,这次对话太棒了。你真厉害。还有两个后续问题。  
**[01:23:29] Speaker B:** Where can folks find you online if they want to reach out?  
如果大家想联系你,可以在哪里找到你?  
**[01:23:32] Speaker A:** Out and just follow what you're up to? And how can listeners be useful to you?  
关注你在做什么?还有听众怎样能帮到你?  
**[01:23:35] Speaker B:** The best way to reach out is I am Catwoo on Twitter. Feel free to like tag me in things. Feel free to DM me. I read all my DMs. I don't always respond to every single one, but I will read them all.  
最好的联系方式是我在 Twitter 上的账号 Catwoo。欢迎在帖子里 @ 我。欢迎给我发私信。我会读所有私信,虽然不一定每条都回复,但我都会看。  
**[01:23:49] Speaker B:** And then the thing that is most helpful is tell us where Claude Code and Cowork aren't working well for you. We are very grateful for the amount of positive feedback. But the things that we thrive on is edge cases, errors, like specific tasks that we can reproduce where Claude Code or Cowork fail.  
然后最有帮助的是告诉我们 Claude Code 和 Cowork 在哪些地方对你不好用。我们非常感激大家的正面反馈,但我们真正需要的是边缘案例、错误,以及我们能复现的、Claude Code 或 Cowork 失败的具体任务。  
**[01:24:13] Speaker B:** Because if you're able to share that with us and we're able to reproduce it, then this is something that we're able to actively improve for our next generations of models and for our next harnesses.  
因为如果你能分享给我们,而我们能复现它,那这就是我们能在下一代模型和下一代工具中主动改进的东西。  
**[01:24:25] Speaker A:** Extremely cool. Everyone on Twitter are not shy with sharing this feedback. So keep it coming.  
太酷了。Twitter 上的大家在分享反馈这方面从不害羞,所以请继续。  
**[01:24:30] Speaker B:** Share us, please, please share the  
请分享给我们,拜托了,请分享——  
**[01:24:33] Speaker A:** Problems that you're having with us.  
你们遇到的问题。  
**[01:24:34] Speaker B:** Yeah, and it's really cool to see all your team being so active on Twitter and responding to people, and so like what I'm hearing, like this is actually stuff you guys actually see and react to, so...  
是的,看到你们整个团队在 Twitter 上这么活跃、回复大家,真的很酷。所以我听到的意思是,这些反馈你们团队确实会看到并采取行动,对吧……  
**[01:24:44] Speaker A:** Yeah, we appreciate everyone being so engaged with us. It gives the team a ton of energy. We have this channel of like user love, and so whenever you guys share a success story, we post it there, and whenever you guys share like issues with our product, we put it into our feedback channel. That way our broader team is able to act on it.  
是的,我们很感激大家这么积极地和我们互动。这给团队带来了巨大的能量。我们有一个「用户喜爱」频道,所以每当你们分享成功案例时,我们会发到那里;每当你们分享产品问题时,我们会放进反馈频道。这样我们更广泛的团队就能据此采取行动。  
**[01:25:02] Speaker B:** That is so cool to know. Thanks for sharing that. Well, C, thank you so much for being here.  
知道这个太好了。谢谢分享。好的,Cat,非常感谢你来做客。  
**[01:25:07] Speaker A:** Thanks for having me.  
谢谢邀请我。  
**[01:25:09] Speaker B:** Bye everyone.  
大家再见。  
**[01:25:11] Speaker B:** Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also...  
非常感谢收听。如果你觉得有价值,可以在 Apple Podcasts、Spotify 或你喜欢的播客应用上订阅本节目。还有……  
**[01:25:19] Speaker A:** Please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast.  
也请考虑给我们评分或留下评论,因为这真的能帮助其他听众找到这个播客。  
**[01:25:25] Speaker A:** You can find all past episodes or learn more about the show at lennispodcast.com.  
你可以在 lennispodcast.com 找到所有往期节目或了解更多关于本节目的信息。  
**[01:25:30] Speaker A:** See you in the next episode.  
下期节目见。  

---

## Deep Dive Summary

### Topic 1: The challenge of building for current AI capabilities vs AGI
为当前AI能力构建产品的挑战
_[00:00]_

**Q:** What is the core challenge in building AI products today?
**问：** 当今构建AI产品的核心挑战是什么？

**A:** The fundamental tension in AI product development lies in calibrating expectations between future AGI potential and present-day model limitations. While it's tempting to design for hypothetical "super AGI strong model" capabilities, the real engineering challenge is "figuring out for the current model, how do you elicit the maximum capability" from systems that exist today. This requires being "the right amount of AGI pilled"—optimistic enough to build ambitious products, but grounded enough to work within actual constraints rather than imagined future ones.
**答：** AI产品开发的核心矛盾在于如何平衡对未来AGI潜力的期待和当前模型能力的现实。虽然为假想的"super AGI strong model"设计产品很诱人，但真正的工程挑战是如何从现有系统中"elicit the maximum capability"。这需要保持"the right amount of AGI pilled"——既要有足够的雄心构建创新产品，又要脚踏实地地在实际约束而非想象的未来能力范围内工作。

### Topic 2: Anthropic's accelerated shipping velocity
Anthropic的快速迭代策略
_[00:13]_

**Q:** How has Anthropic reduced product development timelines?
**问：** Anthropic如何实现产品开发提速？

**A:** Anthropic has dramatically compressed product development cycles by systematically "removing every single barrier to shipping things," achieving timeline reductions from 6 months down to 1 month or even a single day for many features. This acceleration represents a fundamental shift in their operational philosophy, prioritizing rapid iteration over traditional product management approaches. The interviewer's observation about PM candidates "approaching it very incorrectly" suggests Anthropic may be deliberately selecting for or cultivating a different mindset that enables this velocity.
**答：** Anthropic通过系统性地"remove every single barrier to shipping"，将产品功能的开发周期从6个月压缩到1个月，有些甚至缩短到一天。这种提速不仅是流程优化，更是运营理念的根本转变，优先考虑快速迭代而非传统的产品管理方式。访谈中提到很多PM候选人的思路不对，暗示Anthropic在刻意寻找或培养一种能支撑这种速度的特殊工作方式。

### Topic 3: The evolving PM role in AI product development
AI产品开发中PM角色的演变
_[00:27]_

**Q:** How is the product manager role changing with AI?
**问：** 产品经理角色如何随AI而变化？

**A:** The PM role is fundamentally shifting toward rapid iteration and product taste as AI native products require "launching features every single week" rather than traditional development cycles. The core skill transformation centers on decision-making over execution—as implementation "becomes much cheaper to write," the critical value lies in "deciding what to write." This represents a strategic elevation where PMs must develop stronger product intuition and judgment to navigate the accelerated pace of AI product development, as exemplified by Kat Woo's work leading Claude Code and Co-work at Anthropic.
**答：** PM角色正在发生根本性转变，核心是快速迭代和产品品味。AI原生产品要求"每周发布新功能"而非传统开发周期。最关键的技能转型在于决策能力超越执行能力——当实现变得"更便宜"时，真正有价值的是"决定写什么"。这意味着PM需要培养更强的产品直觉和判断力，才能驾驭AI产品开发的加速节奏，正如Anthropic的Claude Code和Co-work产品负责人Kat Woo所展示的那样。

### Topic 4: Introduction to Kat Woo and episode setup
Kat Woo介绍和节目开场
_[01:00]_

**Q:** Who is the guest and what makes this episode important?
**问：** 嘉宾是谁，为什么这期节目重要？

**A:** The host positions Kat Woo as a central figure in the intersection of AI and product development, emphasizing that her team is building a product that is fundamentally transforming how others build their own products. The introduction frames Kat as someone "full of insights and wisdom and lessons," signaling that the episode will focus on practical knowledge from someone actively shaping the AI product landscape. The host's enthusiasm and claim that this is "an episode you cannot miss" sets high expectations for actionable insights from someone at the forefront of AI-driven product innovation.
**答：** 主持人将Kat Woo定位为AI和产品开发交叉领域的核心人物，强调她的团队正在打造一款从根本上改变其他人构建产品方式的产品。开场介绍将Kat描述为充满"insights and wisdom and lessons"的人，预示着这期节目将聚焦于来自AI产品领域前沿实践者的实用知识。主持人的热情和"不容错过"的表述为听众设定了高期待，暗示将获得AI驱动产品创新的可操作性见解。

### Topic 5: Kat's PM role and collaboration dynamics with Boris on Claude Code
Kat的PM角色及与Boris在Claude Code上的协作模式
_[01:37]_

**Q:** How does Kat work with Boris and what does her PM role entail on the Claude Code team?
**问：** Kat如何与Boris协作，她在Claude Code团队的PM角色具体是什么？

**A:** Kat describes a complementary partnership where Boris serves as tech lead and "product visionary" who defines the "AGI pill version" and 3-6 month roadmap, while she focuses on translating that vision into execution paths and managing cross-functional alignment with marketing, sales, finance, and capacity teams to remove shipping blockers. The collaboration operates on what she calls an "80% mind meld" model where they're largely aligned, with each driving the remaining 20% of initiatives they individually care more about. This division allows Boris to maintain his high-velocity technical leadership while Kat ensures organizational readiness and coordination across the company.
**答：** Kat描述了一种互补的合作模式：Boris作为技术负责人和"产品愿景者"，定义"AGI pill版本"和3-6个月的产品路线图，而她的重点是将愿景转化为具体执行路径，并协调市场、销售、财务、容量等跨职能团队，确保功能发布时没有阻碍。两人的协作基于"80% mind meld"模式——大部分时候高度一致，各自主导剩余20%自己更关注的事项。这种分工让Boris能保持高速的技术输出，同时Kat确保组织层面的准备度和跨部门协同。

### Topic 6: Common mistakes PMs make when interviewing for AI roles
PM面试AI岗位时的常见错误
_[04:30]_

**Q:** What are PMs getting wrong about AI product management?
**问：** PM对AI产品管理有哪些误解？

**A:** PMs interviewing for AI roles often fail to grasp the fundamental shift in velocity that AI has created, still approaching the role with pre-AI assumptions about planning horizons and coordination overhead. The core mistake is overemphasizing "aligning multi-quarter roadmaps with partner teams" when AI has compressed product timelines from 6 months down to "one month and sometimes to one week or even one day." Successful AI PMs prioritize speed-to-user over coordination, focusing on "how can I shorten the time from having this idea to actually getting the product in the hands of users" and creating "concept corners" where ideas can ship by week's end. The role has fundamentally transformed from coordination-heavy planning to execution-focused rapid iteration, yet many candidates haven't internalized that "how much of the job now is just moving, is helping the team move fast."
**答：** 面试AI产品岗位的PM往往没有意识到AI带来的速度变革有多根本，仍然用传统思维来理解这个角色。最大的误区是过度强调跨团队协调和长期规划对齐，而AI已经把产品周期从6个月压缩到了"一个月，有时一周甚至一天"。成功的AI PM优先考虑的是如何"缩短从想法到用户手中的时间"，建立能让工程师或PM的想法在一周内上线的"concept corner"机制。这个角色已经从协调密集型转变为执行导向的快速迭代，但很多候选人还没理解"现在这份工作的核心就是推动，帮助团队快速前进"。

### Topic 7: Tactics for enabling fast shipping: clear goals and user focus
实现快速发布的策略：明确目标和用户聚焦
_[06:36]_

**Q:** What specific practices help teams move faster in AI product development?
**问：** 哪些具体实践帮助团队在AI产品开发中更快行动？

**A:** The speaker argues that setting clear goals is the first critical practice because "LLMs are so general" which creates significant ambiguity around target users, problems, and use cases. A concrete example illustrates this: defining the goal as enabling "professional developers at enterprises to safely get to zero permission prompts" immediately "rules out a lot of potential approaches" and focuses the team's efforts. The speaker emphasizes that great PMs must explicitly specify the key user, the core problem (like "too many permission prompts" causing fatigue), and the precise use case to cut through the generality of LLM capabilities. Beyond goal-setting, the speaker mentions establishing "some repeatable process for getting these features shipped" as the second essential element.
**答：** 嘉宾认为设定明确目标是第一个关键实践，因为 LLM 能力太通用，会在目标用户、问题定义和使用场景上造成大量模糊性。他用一个具体例子说明：将目标定义为让企业的专业开发者安全地实现零权限提示，这样的明确目标会立即排除掉很多可能的方案，让团队聚焦。优秀的 PM 必须明确指出关键用户、核心问题（比如权限提示过多导致疲劳）以及精确的使用场景，才能穿透 LLM 的通用性。除了目标设定，嘉宾还提到建立可重复的功能发布流程是第二个重要因素。

### Topic 8: Research preview strategy and cross-functional processes
研究预览策略与跨职能协作流程
_[07:42]_

**Q:** How does Anthropic use research previews and streamline launches?
**问：** Anthropic 如何利用研究预览机制并简化产品发布流程？

**A:** Anthropic's Cloudcode team ships "almost all" features in research preview, clearly branding them as early products to "reduce our commitment" and enable shipping "in a week or two" rather than months. The PM's role is to establish frameworks that define when to involve cross-functional partners and set their expectations, creating what the speaker describes as a "really tight process between engineering, marketing, and docs." This system allows engineers to post ready features in an "Evergreen launch room" where docs, PMM, and DevRel teams can "turn around the marketing announcement" within a day, dramatically lowering friction for any engineer to ship.
**答：** Anthropic 的 Cloudcode 团队将"几乎所有"功能都以研究预览形式发布，明确标注为早期产品以"降低承诺"，使团队能在"一两周内"而非数月完成发布。PM 的核心职责是建立框架，明确何时引入跨职能伙伴及其预期，形成"engineering、marketing 和 docs 之间非常紧密的流程"。工程师将准备好的功能发布到"Evergreen launch room"，docs、PMM 和 DevRel 团队能在次日完成营销公告，大幅降低任何工程师发布功能的摩擦。

### Topic 9: The role of PRDs, metrics, and team principles
PRD、指标和团队原则的作用
_[08:59]_

**Q:** How do PRDs and documentation fit into fast-moving AI product development?
**问：** PRD和文档如何适应快速发展的AI产品开发？

**A:** Anthropic uses a lightweight, selective approach to PRDs rather than making them mandatory for all work. The team maintains alignment through "very rigorous metrics" with weekly readouts and a documented list of "team principles" that articulate key users and tradeoffs, enabling engineers to "make decisions by themselves without feeling blocked on PM." PRDs are reserved for "particularly ambiguous" features or "projects that require heavy infrastructure that do take many months," typically as one-pagers covering goals, delightful use cases, and current failure modes. This approach balances speed with clarity by embedding shared understanding in metrics and principles rather than upfront documentation.
**答：** Anthropic对PRD采用轻量化、选择性的方式，而非强制要求所有工作都写PRD。团队通过每周的"very rigorous metrics"复盘和明确的"team principles"文档来保持一致性，后者阐明了核心用户和权衡取舍，让工程师能够"make decisions by themselves without feeling blocked on PM"。PRD只用于"particularly ambiguous"的功能或"需要数月时间的重基础设施项目"，通常是一页纸，涵盖目标、理想用例和当前失败模式。这种做法通过指标和原则嵌入共识，而非前置文档，在速度和清晰度之间取得平衡。

### Topic 10: Anthropic's shipping pace and use of Mythos model
Anthropic的快速发布节奏与Mythos模型的应用
_[10:29]_

**Q:** How does Anthropic maintain such a rapid shipping cadence, and are they using their powerful Mythos model internally?
**问：** Anthropic如何保持快速的产品发布节奏？他们是否在内部使用强大的Mythos模型？

**A:** The interviewer observes that Anthropic has achieved an unprecedented shipping velocity, with "literally every day" featuring a major feature or product launch according to a community-created calendar. The question specifically probes whether Anthropic's internal use of Mythos—described as an "incredible model" that remains "in preview because it's so powerful people are a little afraid of what it can do"—might be accelerating their development capabilities. The transcript slice ends with the question posed but does not contain the respondent's answer, leaving the connection between Mythos usage and shipping speed unconfirmed in this segment.
**答：** 访谈者观察到Anthropic实现了前所未有的产品发布速度，根据社区制作的日历显示"literally every day"都有重要功能或产品推出。问题特别探讨了Anthropic内部使用Mythos的情况——这个模型被描述为"incredible"且因为"so powerful people are a little afraid of what it can do"而仍处于预览阶段——是否可能加速了他们的开发能力。这段对话以提问结束，但未包含回答者的答复，因此Mythos使用与发布速度之间的联系在此片段中未得到证实。

### Topic 11: Shipping velocity and internal tooling at Anthropic
Anthropic 的发布速度和内部工具使用
_[11:03]_

**Q:** How much does using Anthropic's own models internally contribute to their fast shipping pace?
**问：** 在内部使用 Anthropic 自己的模型对其快速发布节奏有多大贡献？

**A:** While Anthropic does use their models internally and it has "increased our rate of shipping a little bit," the speaker emphasizes this doesn't explain "the bulk of the increase" in velocity. The primary driver is organizational culture: they're "very low on process" and deliberately "remove every single barrier to shipping things." The goal is empowering every team member to take an idea "from just an idea to out in the world in less than a week, sometimes" even in a single day, suggesting that process minimization and team autonomy matter more than tooling advantages.
**答：** 虽然 Anthropic 内部确实使用自家模型，并且"稍微提升了发布速度"，但发言人强调这并不能解释速度提升的"主要原因"。真正的驱动力是组织文化：他们"流程极少"，刻意"移除一切发布障碍"。目标是让每个团队成员都能在"不到一周，有时甚至一天内"把想法变成现实产品，这表明流程精简和团队自主权比工具优势更重要。

### Topic 12: Claude source code leak incident
Claude 源代码泄露事件
_[11:49]_

**Q:** What happened with the Claude code leak and how did Anthropic respond?
**问：** Claude 代码泄露事件是怎么发生的，Anthropic 如何应对？

**A:** Anthropic's Claude source code was accidentally leaked due to "human error" when an employee was working with Claude to write a PR for updating package releases. The incident passed through "two layers of human review" before the leak occurred, indicating a systemic process failure rather than individual negligence. Anthropic treated it as a "process failure" and responded by hardening their safeguards, with the employee remaining at the company as the focus shifted to learning and preventing recurrence.
**答：** Anthropic 的 Claude 源代码因为「人为失误」而意外泄露，当时一名员工正在使用 Claude 协助编写更新包发布流程的 PR。这个事件经过了「两层人工审查」才发生泄露，说明这是系统性的流程问题而非个人疏忽。Anthropic 将其定性为「流程失败」，通过加固安全措施来应对，涉事员工继续留任，公司重点放在吸取教训和预防再次发生上。

### Topic 13: OpenClaude API access changes
OpenClaude API 访问限制调整
_[12:54]_

**Q:** Why did Anthropic restrict Claude subscription access for third-party products like OpenClaude?
**问：** 为什么 Anthropic 限制了 Claude 订阅对 OpenClaude 等第三方产品的访问？

**A:** Anthropic restricted third-party access because the $200/month subscription "wasn't designed for third party products which have different usage patterns" than first-party offerings, creating unsustainable demand on infrastructure. The company made "the hard decision" to "prioritize our first party products and our API" while working to scale infrastructure and improve token efficiency. As a transition measure, Anthropic provided subscription users with API credits, attempting to find "the most seamless transition" possible. The interviewer validated this as a necessary business decision, noting that companies "can't just give away compute when it's so in demand" and that users don't understand the economics of subsidizing essentially unlimited usage.
**答：** Anthropic 限制第三方访问是因为每月 200 美元的订阅"wasn't designed for third party products"，第三方产品的使用模式与自家产品不同，给基础设施带来了不可持续的需求压力。公司做出了"hard decision"，决定"prioritize our first party products and our API"，同时努力扩展基础设施和提高 token 效率。作为过渡方案，Anthropic 为订阅用户提供了 API credits，试图找到"most seamless transition"。访谈者认为这是必要的商业决策，指出公司"can't just give away compute when it's so in demand"，用户往往不理解补贴几乎无限使用量背后的经济成本。

### Topic 14: PM team structure at Anthropic
Anthropic 的 PM 团队结构
_[14:17]_

**Q:** How is the product management team organized at Anthropic and what do different PM teams focus on?
**问：** Anthropic 的产品管理团队是如何组织的，不同的 PM 团队关注什么？

**A:** Anthropic has approximately 30-40 PMs organized into five specialized teams aligned with different product surfaces and customer needs. The research PM team, led by Diane, acts as a bridge "understanding all of the feedback from our customers for our models" and shepherding model launches, while the Cloud developer platform team maintains the API infrastructure and releases capabilities like "managed agents" for hosted solutions. The consumer-facing teams include Claude (covering Claude.ai and core products), Enterprise (focused on adoption enablers like "cost controls, RBAC, security controls"), and a cross-functional Growth team that drives expansion across the entire product suite including both consumer products and the Cloud API.
**答：** Anthropic 有大约 30-40 名 PM，分为五个专业团队，分别对应不同的产品层和客户需求。研究 PM 团队由 Diane 领导，负责收集客户对模型的反馈并推动模型发布，充当研究团队和用户之间的桥梁；Cloud 开发者平台团队维护 API 基础设施，并发布 managed agents 等托管能力。面向用户的团队包括 Claude 团队（负责 Claude.ai 和核心产品）、Enterprise 团队（专注于成本控制、RBAC、安全控制等企业采用所需的功能），以及一个跨职能的 Growth 团队，负责推动包括消费者产品和 Cloud API 在内的整个产品线的增长。

### Topic 15: Future of PM role and role convergence
PM 角色的未来与职能融合
_[15:42]_

**Q:** Will companies need more or fewer PMs in the AI era, and how are PM, engineering, and design roles merging?
**问：** AI 时代公司需要更多还是更少的 PM，以及 PM、工程和设计角色如何融合？

**A:** The speakers argue that roles are fundamentally merging, with "PMs doing some engineering work, engineers doing PM work, designers are PMing and also landing code," creating two viable strategies: hiring more PMs to guide fast-moving engineers, or hiring engineers with strong product taste who can ship end-to-end. At Anthropic, they prioritize the latter approach, where engineers can go "from seeing user feedback on Twitter through to shipping a product at the end of the week with almost no product involvement," and notably all their PMs have engineering backgrounds while designers are former front-end engineers. The core skill that becomes most valuable is "product taste"—deciding what to build from "tens of thousands of GitHub issues"—though engineering background remains particularly useful in the near term because it provides "a better sense for how hard something should be," enabling faster prioritization decisions between quick implementations versus costly builds.
**答：** 两位嘉宾认为职能边界正在消失，PM 在写代码、工程师在做产品决策、设计师既做 PM 又提交代码，形成了两种策略：要么多招 PM 来指导快速迭代的工程师，要么招有产品品味的工程师实现端到端交付。Anthropic 选择后者，工程师可以「从 Twitter 上看到用户反馈到一周内发布产品，几乎不需要 PM 介入」，团队所有 PM 都有工程背景，设计师也是前端工程师出身。最核心的能力是「product taste」——从「成千上万的 GitHub issues」中判断该做什么，不过工程背景在短期内仍然特别有用，因为能「更好地判断某件事有多难」，帮助快速区分哪些可以一小时搞定、哪些成本高昂。

### Topic 16: Evolving skill requirements in AI-accelerated development
AI 加速开发中不断变化的技能要求
_[19:27]_

**Q:** What skills will remain valuable as AI coding capabilities advance, and how should product people adapt?
**问：** 随着 AI 编码能力的提升，哪些技能将保持价值，产品人员应该如何适应？

**A:** Speaker A emphasizes that valued skill sets are shifting rapidly with "large increases in coding capability" occurring every few months, making it difficult to predict beyond that timeframe. The most critical capability is "first principles thinking" to understand how the tech landscape is changing and identify what the team needs, as the work is becoming "more amorphous." The current environment rewards people who can "wear a lot of hats," swap between roles with "very low ego," and quickly learn new skill sets to fill the highest priority gaps. Speaker B synthesizes this as humans remaining essential for picking what to work on, understanding market direction, prioritizing effectively, and evaluating whether what's been built "is good and right."
**答：** Speaker A 强调技能需求变化很快，每隔几个月 AI 的编码能力就会大幅提升，导致很难预测更远的未来。最关键的能力是「first principles thinking」，能够理解技术格局的变化并识别团队真正需要什么，因为工作正在变得越来越模糊和流动。当前环境青睐那些能够「wear a lot of hats」、在不同角色间灵活切换、对具体做什么工作没有「ego」、并能快速学习新技能来填补最高优先级缺口的人。Speaker B 总结认为人类的价值在于：选择做什么、理解市场方向、确定优先级，以及判断构建出来的东西是否「good and right」。

### Topic 17: Human common sense vs AI limitations in product launches
产品发布中人类常识与 AI 的局限性
_[21:43]_

**Q:** What gaps do AI models still have compared to human judgment in managing complex product launches?
**问：** 在管理复杂产品发布时，AI 模型相比人类判断还有哪些差距？

**A:** Speaker B identifies a critical gap in AI's ability to navigate the social and organizational complexity of product launches, which involve "a thousand moving pieces" where many things could go wrong. The core limitation is that models lack "tacit common sense" and "EQ kind of knowledge"—they struggle to map stakeholder relationships, understand individual preferences, and choose appropriate communication channels to maintain alignment. While B expresses optimism that models will improve in these areas, the current deficiency highlights that human judgment remains essential for managing the interpersonal and contextual nuances that determine launch success.
**答：** Speaker B 指出 AI 在处理产品发布的社交和组织复杂性方面存在关键缺陷，产品发布涉及"a thousand moving pieces"，很多环节可能出错。核心局限在于模型缺乏"tacit common sense"和"EQ kind of knowledge"——它们难以理解利益相关者之间的关系、个人偏好，以及选择合适的沟通渠道来保持团队协调。虽然 B 对模型未来的改进持乐观态度，但当前的不足表明，人类判断在管理决定发布成功的人际和情境细微差别方面仍然不可或缺。

### Topic 18: Thriving in constant chaos and change
在持续混乱和变化中保持状态
_[22:24]_

**Q:** How do you stay sane and effective when working inside the tornado of rapid AI development?
**问：** 在快速 AI 开发的漩涡中，如何保持理智和高效？

**A:** The team deliberately hires people who "lean into the chaos" and face challenges with optimism rather than stress, recognizing that perfectionism leads to burnout in an environment where priorities escalate rapidly—sometimes from "P0 to P0000" within a single day. They practice brutal prioritization focused on core use cases, accepting that shipping "products that aren't as polished as I wish they were" is necessary to maintain velocity and gather real feedback. The approach requires acknowledging limits, sleeping well to make good decisions, and being "okay letting things go," which represents a significant mindset shift from traditional software development where buggy releases would be unacceptable. They specifically seek experienced industry veterans who understand their own energy management and have weathered previous cycles of chaos.
**答：** 团队刻意招募那些能够 "lean into the chaos" 的人，用乐观而非焦虑来面对挑战，因为他们认识到在优先级快速升级的环境中（有时一天内从 "P0 升级到 P0000"），完美主义只会导致倦怠。他们实行残酷的优先级排序，聚焦核心用例，接受发布 "不够完善的产品" 来保持速度并获取真实反馈。这种方法要求承认自身局限，保证睡眠以做出好决策，并且 "okay letting things go"，这与传统软件开发中不能接受有 bug 的发布形成鲜明对比。他们特别寻找有丰富行业经验的老手，这些人懂得管理自己的精力，经历过多轮混乱周期。

### Topic 19: Trade-offs of blurred roles and rapid shipping
角色模糊和快速迭代的代价
_[25:05]_

**Q:** What do we sacrifice when engineers become PMs and teams ship features at unprecedented speed?
**问：** 当工程师兼任 PM 角色、团队以前所未有的速度发布功能时，我们牺牲了什么？

**A:** The primary sacrifice is product consistency—when "code was expensive to write," teams carefully planned how products related to each other, but AI's rapid pace now leads to overlapping features as teams test multiple form factors simultaneously. This creates confusion for new users who "might not know what is the best path to accomplish X," requiring more education around core features and best practices. Beyond product architecture, users struggle with the psychological burden of keeping up, feeling pressure to "check Twitter every single day" rather than the traditional monthly or quarterly update cycle, creating an "ever-increasingly fast treadmill" that demands better onboarding and in-product education.
**答：** 最大的牺牲是产品一致性。过去代码开发成本高时，团队会仔细规划产品间的关系，但 AI 快速发展导致现在会同时测试多个功能形态，造成功能重叠。新用户因此不清楚「完成某个任务的最佳路径是什么」，需要更多关于核心功能和最佳实践的引导。除了产品架构层面，用户还面临心理负担——需要「每天刷 Twitter」了解最新动态，而不是传统的按月或按季度更新节奏，这种「越来越快的跑步机」需要更好的产品内教育来缓解。

### Topic 20: PowerUp feature and onboarding challenges
PowerUp 功能与用户引导的挑战
_[27:48]_

**Q:** How does the /powerup feature address the challenge of too many features and user overwhelm?
**问：** /powerup 功能如何解决功能过多导致的用户困扰？

**A:** The team initially resisted building onboarding flows because they believed "the product should be intuitive enough that you don't actually need to go through any tutorial," reflecting a design philosophy prioritizing inherent usability. However, they eventually diverged from this principle due to overwhelming user demand for guidance through their feature-rich product, with users specifically asking "what are the 10 that I absolutely need to use" out of "100 features." The /powerup feature emerged as a pragmatic compromise, providing a curated walkthrough of best practices and essential features to help users navigate the complexity without requiring them to discover everything independently.
**答：** 团队最初坚持不做用户引导流程，因为他们相信产品应该足够直观，用户不需要教程就能上手。但随着产品功能越来越多，用户强烈要求内置引导体验，特别是希望从"100 个功能"中知道"哪 10 个是必须掌握的"。最终团队妥协推出了 /powerup 功能，通过精选的最佳实践和核心功能演示，帮助用户应对复杂度，而不是让他们自己摸索。

### Topic 21: Anthropic's come-from-behind success story
Anthropic 的逆袭成功故事
_[28:32]_

**Q:** What ingredients allowed Anthropic to go from underdog to 11 billion ARR with massive growth?
**问：** 是什么因素让 Anthropic 从弱者成长为 110 亿美元 ARR 并实现大规模增长？

**A:** Anthropic's remarkable turnaround from being "the least funded" and lacking distribution to achieving $11 billion ARR stems from two core ingredients. First, their "unifying mission" of "bringing safe AGI to all of humanity" serves as the decisive framework for prioritization—when competing priorities arise, teams explicitly discuss "which one is more important for Anthropic's mission" and everyone aligns behind that decision, even if it means deprioritizing individual product lines like Claude code features. Second, this mission-driven approach translates into exceptional focus, where "teams are willing to make sacrifices that hurt their own goals and their own KPIs in service of Anthropic's goals," preventing scope creep into unrelated ventures like social networks or content feeds that other AI companies pursued. The speaker emphasizes this unified decision-making at scale is "something that I've never seen at a company of our scale," suggesting it's a rare organizational capability that enables faster execution across the entire organization.
**答：** Anthropic 从「资金最少」、缺乏分发渠道的弱势地位逆袭到 110 亿美元 ARR，核心在于两个要素。首先是「统一使命」——「为全人类带来安全的 AGI」成为决策框架，当出现优先级冲突时，团队会明确讨论「哪个对 Anthropic 的使命更重要」，所有人都会支持最终决定，即使这意味着要推迟 Claude code 等产品功能的发布。其次，这种使命驱动转化为极致的专注力，「团队愿意牺牲自己的目标和 KPI 来服务 Anthropic 的整体目标」，避免像其他 AI 公司那样分散精力去做社交网络或内容推荐等不相关的产品。受访者强调这种规模下的统一决策能力「在同等规模公司中从未见过」，这是实现快速执行的罕见组织能力。

### Topic 22: Mission-driven decision making and focus
使命驱动的决策与专注力
_[31:10]_

**Q:** How does Anthropic's safety mission enable fast, unified decisions across the organization?
**问：** Anthropic 的安全使命如何推动组织内快速统一的决策？

**A:** Anthropic's mission creates organizational alignment by prioritizing company-level goals over individual team metrics, enabling teams to "make sacrifices that hurt their own goals and their own KPIs in service of Anthropic's goals." This manifests in concrete product decisions: the speaker states "if Claude failed but Anthropic succeeded, I would be extremely happy," illustrating willingness to subordinate specific products to broader mission objectives. The recent decision to discontinue open source efforts exemplifies this framework—when third-party distribution conflicted with growing direct user reach through "Claude subscriptions with our first-party products," the company chose to "double down" on the latter despite trade-offs.
**答：** Anthropic 的使命通过将公司整体目标置于团队指标之上来实现组织协同，团队愿意"牺牲自己的目标和 KPI 来服务 Anthropic 的目标"。这体现在具体产品决策中：发言人表示"如果 Claude 失败但 Anthropic 成功，我会非常高兴"，说明他们愿意让具体产品服从更大的使命目标。停止开源项目的决策就是这一框架的实例——当第三方分发与通过"Claude 订阅和第一方产品"直接触达用户的目标冲突时，公司选择"加倍投入"后者，即使这意味着权衡取舍。

### Topic 23: When to use Claude Code, Desktop, and Cowork
Claude Code、Desktop 和 Cowork 的使用场景
_[32:34]_

**Q:** What are the best use cases for each of Anthropic's three main product surfaces?
**问：** Anthropic 三个主要产品的最佳使用场景分别是什么？

**A:** The speakers frame product selection around output type and context: Claude Code CLI is positioned as "the most powerful" with features landing first, ideal for "one-off coding tasks" at your desk, while Desktop adds a graphical layer with real-time preview that's "really great for people who want something a bit more graphical" and serves as a "one-stop control plane" across all sessions. Mobile/web fills the mobility gap for users who are "out and about" and don't want to be "tethered to their phone" with an open laptop, enabling task initiation on the go. Cowork addresses a fundamentally different need: "if the output is anything that's not code" like Slack management, slide decks, or planning docs, making it the non-coding companion that's "growing incredibly fast" despite being underappreciated.
**答：** 产品选择的核心逻辑是输出类型和使用场景：Claude Code CLI 是"最强大的"工具，新功能最先上线，适合在电脑前处理"一次性编码任务"；Desktop 增加了图形界面和实时预览功能，对"想要图形化界面的人"更友好，同时作为"一站式控制面板"统一管理所有会话。移动端和 web 版解决了移动办公的痛点，让用户在"外出时"不必"拴着手机"带着打开的笔记本电脑，可以随时启动任务。Cowork 则针对完全不同的需求："如果输出不是代码"，比如处理 Slack 消息、制作幻灯片或撰写规划文档，它正在"快速增长"但仍被低估。

### Topic 24: Getting started with Cowork: connecting data sources
开始使用 Cowork：连接数据源
_[35:58]_

**Q:** What are the first steps and use cases for using Cowork as a PM?
**问：** 作为产品经理使用 Cowork 的第一步和用例是什么？

**A:** The foundational step for using Cowork effectively is connecting all relevant data sources, as the system "can only do a great job if it has access to all the context" needed to curate outputs. For a PM role, this means integrating Google Calendar, Slack, Gmail, and Google Drive to give Cowork "the flexibility to find relevant context, to ask questions, to pull in threads," which substantially improves result quality. The speaker illustrates practical usage with preparing for an upcoming Cowork Cloud conference, specifically working on a talk about Cowork's evolution "from an assistant to like a full-on agent."
**答：** 使用 Cowork 的关键第一步是连接所有相关数据源，因为系统需要完整的上下文才能生成高质量的输出。对于产品经理来说，这意味着要集成 Google Calendar、Slack、Gmail 和 Google Drive，让 Cowork 能够灵活地查找相关信息、提问和整合对话线索，从而显著提升结果质量。演讲者以准备 Cowork Cloud 大会演讲为例，说明了实际应用场景——特别是关于 Cowork 从助手演变为完整 agent 的主题演讲。

### Topic 25: Using Cowork to build a conference presentation deck
用 Cowork 自动生成会议演讲幻灯片
_[36:42]_

**Q:** How did Speaker A use Cowork to create a 20-page deck for the Cowork Cloud conference?
**问：** Speaker A 如何用 Cowork 为 Cowork Cloud 会议生成 20 页演讲幻灯片？

**A:** Speaker A connected Cowork to Google Drive, Slack, and other data sources, then provided a prompt describing the desired narrative along with the PMM's draft outline and constraints about avoiding overlap with the keynote. Cowork autonomously worked for hours, synthesizing information from Twitter launches, internal Slack channels, and codebase announcements to produce a 20-page deck that was "pretty good" but required feedback because it was "a little too wordy" compared to A's preference for minimal text. The result leveraged Anthropic's design system to look "incredibly polished," transforming what would have been hours of manual work into a strong draft that allowed A to focus on selecting the best demos and making final decisions about content.
**答：** Speaker A 把 Google Drive、Slack 等数据源连接到 Cowork，然后输入了一段 prompt，描述了演讲的叙事方向、PMM 提供的大纲草稿，以及不要和主题演讲重复的约束条件。Cowork 自主工作了几个小时，从 Twitter 发布记录、内部 Slack 频道、代码库公告等渠道综合信息，生成了一份 20 页的幻灯片，质量"相当不错"，但因为文字"有点啰嗦"需要反馈调整（A 更喜欢极简文字风格）。最终成果利用了 Anthropic 的设计系统，看起来"非常精致"，把原本需要数小时的手工劳动变成了一份高质量草稿，让 A 可以专注于挑选最佳 demo 和做最终内容决策。

### Topic 26: The PM's role in AI-assisted work: decision-making over execution
产品经理在 AI 辅助工作中的角色：决策而非执行
_[40:24]_

**Q:** What is the PM's role when working with AI tools like Claude?
**问：** 在使用 Claude 等 AI 工具时，产品经理的角色是什么？

**A:** The PM's role shifts to high-level strategic decision-making while AI handles execution, as demonstrated when the speaker decided on a talk outline covering "the progression from making local tasks successful to making every PR green to helping engineers land more PRs" and selecting which demos would be most compelling for each section. After making these structural decisions, the AI tool "just went off for a few hours and built the whole slide deck," eliminating tedious execution work. This represents a fundamental shift where PMs function like directors working with a "deck designer that also has actual knowledge about what you've worked on," focusing on content strategy rather than production mechanics.
**答：** 产品经理的角色转变为高层战略决策，而 AI 负责执行。演讲者举例说明了这一点：他决定演讲大纲要涵盖"从让本地任务成功到让每个 PR 通过，再到帮助工程师提交更多 PR 的演进过程"，并为每个部分选择最有说服力的 demo。做完这些结构性决策后，AI 工具"就自己跑了几个小时，把整个幻灯片都做出来了"，省去了繁琐的执行工作。这代表了一种根本性转变：产品经理像导演一样与一个"既懂设计又真正了解你工作内容的幻灯片设计师"协作，专注于内容策略而非制作细节。

### Topic 27: How Cowork accesses and uses design systems
Cowork 如何获取和应用设计系统
_[41:08]_

**Q:** How does Cowork know and apply Anthropic's design system to slide decks?
**问：** Cowork 如何了解并将 Anthropic 的设计系统应用到幻灯片中？

**A:** Cowork learns Anthropic's design system by being given direct access to their "standardized deck" used across external engagements, which contains "20 of these example slides" showing colors, fonts, and different slide formats. The system works through example-based learning rather than explicit design rules—Claude examines the template slides to understand the visual patterns. There's also flexibility to "connect to like your Figma MCP" if slide formats are stored there, allowing the tool to pull design specifications from multiple sources.
**答：** Cowork 通过直接访问 Anthropic 用于对外沟通的标准化幻灯片模板来学习设计系统，这个模板包含约 20 个示例页面，展示了颜色、字体和各种幻灯片格式。系统采用基于示例的学习方式，而非明确的设计规则——Claude 通过分析模板幻灯片来理解视觉模式。此外还支持连接 Figma MCP，如果幻灯片格式保存在 Figma 中，工具可以从那里提取设计规范。

### Topic 28: PM tech stack at Anthropic
Anthropic 产品经理的技术栈
_[41:48]_

**Q:** What tools does a PM at Anthropic use in their daily work?
**问：** Anthropic 的产品经理在日常工作中使用哪些工具？

**A:** The PM's core stack centers on Claude, Code, and co-work, with Slack serving as "the core OS" of Anthropic's daily operations. Notably, about 30% of their time is spent "pushing the boundaries of what co-work can do" to identify product limitations and understand model mistakes. The most significant shift has been Claude enabling a "surge in personalized work software" where employees build custom apps for specific use cases rather than adapting to generic tools that "don't perfectly fit the use case." This represents a fundamental change in how internal tooling is approached—lowering the barrier to custom software development across the entire company.
**答：** 产品经理的核心工具是 Claude、Code 和 co-work，而 Slack 是 Anthropic 日常运作的"核心操作系统"。值得注意的是，他们大约 30% 的时间用于"探索 co-work 的边界"，以发现产品不足并理解模型错误。最重要的变化是 Claude 带来了"个性化工作软件的激增"——员工为特定场景构建定制应用，而不是迁就那些"不完全契合使用场景"的通用工具。这代表了内部工具开发方式的根本转变，大幅降低了全公司定制软件开发的门槛。

### Topic 29: Custom internal tools: sales deck generator example
定制内部工具：销售演示文稿生成器案例
_[43:06]_

**Q:** What custom tools have people at Anthropic built with Claude?
**问：** Anthropic 员工用 Claude 构建了哪些定制工具？

**A:** An Anthropic sales team member built a web app that automates the creation of customized sales decks by pulling customer context from Salesforce and Gong to tailor presentations. The tool starts with proven templates like "101, 201, and mastering Claude" and intelligently adapts them based on customer-specific factors: deployment platform (Bedrock, Claude for Enterprise, Console), use case concerns ("code review stage of the SDLC"), and compliance requirements ("HIPAA compliant or needs XYZ security controls"). This automation reduces what was "manual work that could take 20 to 30 minutes" down to "a few seconds," solving the problem where salespeople would either spend the time customizing or skip it entirely and "use the general deck."
**答：** Anthropic 的一位销售人员开发了一个 web 应用，通过从 Salesforce 和 Gong 提取客户信息来自动生成定制化的销售演示文稿。这个工具基于已验证有效的模板（"101, 201, and mastering Claude"），根据客户的具体情况智能调整内容：部署平台（Bedrock、Claude for Enterprise、Console）、关注的使用场景（"code review stage of the SDLC"）以及合规要求（"HIPAA compliant or needs XYZ security controls"）。这个自动化工具将原本需要 "20 到 30 分钟" 的手动工作缩短到 "几秒钟"，解决了销售人员要么花时间定制、要么直接使用通用版本的两难问题。

### Topic 30: Why Slack remains dominant as communication infrastructure
Slack 作为通信基础设施的持久优势
_[44:42]_

**Q:** Why does Slack continue to be the preferred communication tool despite AI advances?
**问：** 在 AI 快速发展的背景下，为什么 Slack 仍然是首选的通信工具？

**A:** Slack has achieved a unique position as "communications infrastructure" that companies treat as an operating system rather than replaceable SaaS software. Unlike Salesforce, which faces pressure from custom-built alternatives, Slack excels at its "core task of helping everyone get real-time updates" so effectively that even cutting-edge teams remain dependent on it. The platform's durability stems from both its execution excellence and its "hackability"—the ease of building custom Slack bots and integrations allows teams to adapt it to their specific workflows rather than seeking alternatives.
**答：** Slack 已经成为企业的"通信基础设施"，地位类似操作系统，而不是可替换的 SaaS 软件。与面临定制化替代压力的 Salesforce 不同，Slack 在"帮助所有人获得实时更新"这个核心任务上做得极其出色，即使是最前沿的团队也离不开它。Slack 的持久性来自两方面：卓越的执行力，以及平台的"可定制性"——团队可以轻松构建 Slack bot 和集成，按自己的方式使用它，而不是寻找替代品。

### Topic 31: Token usage across teams: Applied AI as heavy users
各团队的 token 使用情况：Applied AI 团队是重度用户
_[46:48]_

**Q:** Which teams at Anthropic are the biggest token spenders after engineering?
**问：** 在 Anthropic，除了工程团队之外，哪些团队消耗的 token 最多？

**A:** The Applied AI team emerges as the second-largest token consumer at Anthropic, functioning as "very technical go-to-market" professionals who help customers adopt API and model features. Their heavy usage stems from dual demands: building customer prototypes with Claude Code, which has made this work "so much faster," and managing extensive customer communications including "inbound and historical context call notes" through Cowork. The team's workload intensity is substantial, with individual members handling "five to ten customer engagements on a high day" across multiple accounts, pushing the boundaries of what both tools can accomplish.
**答：** Applied AI 团队是 Anthropic 内部第二大 token 消费群体，他们扮演着"非常技术化的 go-to-market"角色，帮助客户采用 API 和模型功能。他们的高使用量来自两方面需求：一是用 Claude Code 为客户构建原型，这让工作速度大幅提升；二是通过 Cowork 管理大量客户沟通，包括客户咨询和历史通话记录。团队工作强度很高，单个成员在繁忙时一天要处理"五到十个客户对接"，覆盖多个客户账户，不断探索这两个工具的能力边界。

### Topic 32: Applied AI workflows: automated customer meeting prep
Applied AI 团队的自动化客户会议准备工作流
_[48:19]_

**Q:** How does the Applied AI team use Cowork for customer engagement workflows?
**问：** Applied AI 团队如何使用 Cowork 来优化客户互动流程？

**A:** Applied AI team members handling multiple customers use Cowork to automate pre-meeting preparation by requesting overnight summaries of "all my customer meetings that are coming up the next day" along with historical context like past requests and action items. Cowork generates a comprehensive "dossier" or brief that includes real-time research capabilities, such as checking Slack for "the latest ETA" on feature launches so PMs have current information during calls. These workflows are organically created by individual team members and then shared across the team, demonstrating bottom-up adoption where users "push the boundaries of what Cowork can do" rather than following top-down mandates.
**答：** Applied AI 团队成员通常需要管理多个客户，每天可能有 5-10 场客户会议，他们会在前一晚让 Cowork 自动生成第二天所有会议的准备材料，包括客户的历史需求、待办事项和关注重点。Cowork 不仅能整理历史信息形成一份完整的客户档案，还能实时去 Slack 等渠道查询最新的功能发布时间等信息，确保 PM 在客户电话中掌握最新动态。这些工作流程都是团队成员自发创建并相互分享的，体现了用户主动探索 Cowork 能力边界的使用模式。

### Topic 33: Token spend trends: costs increasing with model improvements
Token 支出趋势：模型改进推动成本上升
_[49:25]_

**Q:** How much are employees spending on tokens and how does it compare to salaries?
**问：** 员工在 token 上的花费有多少，与薪资相比如何？

**A:** While Anthropic doesn't provide specific dollar figures, they observe a clear pattern where "token cost per engineer" and "per any knowledge worker" increases substantially with each model improvement. The mechanism is behavioral: as models become more capable, users "delegate far more tasks" to AI tools and "spend a lot more hours" in products like Claude Code and CoWork. This creates a direct correlation between model quality jumps and rising per-employee token costs, though whether costs actually exceed salaries remains unaddressed in this response.
**答：** 虽然 Anthropic 没有给出具体数字，但他们观察到一个明显趋势：每当模型升级或产品大幅改进时，"每个工程师的 token 成本"和"每个知识工作者"的成本都会显著增加。背后的机制是用户行为变化：模型能力越强，用户就会"委托更多任务"给 AI，并在 Claude Code 和 CoWork 等工具上"花费更多时间"。这形成了模型质量提升与员工 token 成本上升之间的直接关联，但成本是否真的超过薪资这个问题并未得到回应。

### Topic 34: Token usage limits at Anthropic
Anthropic 的 token 使用限制
_[50:12]_

**Q:** Do Anthropic employees have unlimited token access and how is usage managed?
**问：** Anthropic 员工是否有无限的 token 使用权限，以及如何管理使用？

**A:** Anthropic employees have generous but not unlimited token access, with some users hitting limits despite high allowances. The company operates on a trust-based system that empowers teams to "build as fast as possible" while relying on employees' understanding of serving costs to use tokens responsibly. Rather than strict enforcement, the culture makes token waste "very frowned upon" and delegates judgment calls to individuals, balancing rapid internal development with cost awareness.
**答：** Anthropic 员工有很高的 token 额度但并非无限制，部分用户仍会触及上限。公司采用基于信任的管理方式，让团队能够"尽可能快地构建"，同时依赖员工对模型服务成本的理解来负责任地使用 token。公司不采用严格管控，而是通过文化让浪费 token 成为"非常不受欢迎的行为"，将判断权交给个人，在快速内部开发和成本意识之间取得平衡。

### Topic 35: Essential PM skills for AI products
AI 产品 PM 的核心技能
_[51:14]_

**Q:** What emerging skills do PMs need when working on AI products?
**问：** 在 AI 产品上工作的 PM 需要哪些新兴技能？

**A:** The most critical skill for AI product managers is navigating the tension between building for current model capabilities versus an AGI future. While it's tempting to be "AGI-pilled" and design simple text-box interfaces for hypothetical superintelligent models, the real challenge is "eliciting the maximum capability" from today's models by guiding users toward strengths and patching weaknesses. The best PMs can "define what the product should look like a month from now" by observing how users push existing limits, then steadily execute while adapting to rapidly changing model performance. This requires being the "right amount of AGI-pilled"—maintaining long-term vision while solving immediate usability problems that help users get onto "the golden path."
**答：** AI 产品经理最关键的能力是在当前模型能力和 AGI 未来之间找到平衡。虽然很容易被 AGI 愿景吸引，设计简单的文本框界面来应对未来的超级智能模型，但真正的挑战在于如何从当前模型中"激发最大能力"——引导用户发挥模型优势，同时弥补其弱点。优秀的 PM 能够通过观察用户如何突破现有产品边界，来"定义一个月后产品应该是什么样子"，然后稳步执行并根据快速变化的模型性能调整方向。这需要保持"适度的 AGI 信念"——既要有长远愿景，又要解决当下的可用性问题，帮助用户走上"黄金路径"。

### Topic 36: Building PM intuition through model introspection
通过模型自省建立 PM 直觉
_[53:19]_

**Q:** How do you develop the skill to understand model capabilities and limitations?
**问：** 如何培养理解模型能力和局限性的技能？

**A:** The speaker advocates for three core practices to build model intuition: first, actively asking models to "introspect on its own behaviors" when they make unexpected decisions, which reveals what "misled it" in the system prompt or task structure. Second, identifying a small group of trusted users—"five people you trust"—who can articulate what makes specific model configurations effective, since "not everyone's feedback is as qualified." Third, building a focused set of evaluations, emphasizing that "just building 10 great evals" can help teams quantify progress and identify gaps, making evals an "underappreciated" tool that more PMs and engineers should prioritize.
**答：** 演讲者提出三个核心实践来建立模型直觉：首先，当模型做出意外决策时，主动让模型"introspect on its own behaviors"，这能揭示 system prompt 或任务结构中"misled it"的地方。其次，找到一小群可信用户——"five people you trust"——他们能准确表达特定模型配置的优劣，因为"not everyone's feedback is as qualified"。第三，构建聚焦的评估集，强调"just building 10 great evals"就能帮助团队量化进展和发现差距，让 evals 成为更多 PM 和工程师应该重视的"underappreciated"工具。

### Topic 37: Role of evals in product development
Eval 在产品开发中的作用
_[55:03]_

**Q:** How important are evals for AI product management and how much time should PMs spend on them?
**问：** Eval 对 AI 产品管理有多重要，PM 应该花多少时间在上面？

**A:** Evals are described as an "underappreciated" tool that doesn't require massive scale to be valuable—"just building 10 great evals" can help teams quantify goals and track progress effectively. The time investment varies significantly by feature type: some teams have dedicated pods collaborating closely with research to measure code behaviors, while the speaker personally jumps into evals when features "need a bit more product definition," producing concrete outputs like five evals with prompts that demonstrate success rates. The approach is pragmatic rather than universal—"not every feature needs it"—but certain features like memory "benefit a lot from this" structured evaluation work.
**答：** Eval 被认为是一个"被低估"的工具，不需要大规模投入就能产生价值——"只需构建 10 个优质 eval"就能帮助团队量化目标和追踪进展。投入时间因功能类型差异很大：有些团队设有专门小组与研究团队紧密协作来精确测量代码行为，而嘉宾本人会在功能"需要更多产品定义"时介入 eval 工作，产出具体成果如五个 eval 及能提升成功率的 prompt。这种方法是务实而非普适的——"并非每个功能都需要"——但像 memory 这样的特定功能"从中获益很多"。

### Topic 38: Identifying expert model evaluators
识别专业的模型评估者
_[56:45]_

**Q:** Who are the people best at evaluating model performance and providing useful feedback?
**问：** 谁最擅长评估模型性能并提供有用的反馈？

**A:** Two groups stand out as exceptional model evaluators: Amanda, who "molds Claude's character" with a rare ability to articulate success criteria for the inherently ambiguous task of character design, and the Claude Code team, whose rapid feedback during team lunches surfaces specific behavioral patterns like models being "too abrupt" or "loves writing a ton of memories." The Code team's qualitative observations serve as hypothesis generators that guide quantitative analysis—they identify potential issues that the team then validates with data, solving the challenge that "we have a ton of data, but it is very hard to extract insights." This human-in-the-loop evaluation process is particularly valuable because character crafting "requires a very strong sense of conviction" that can't be verified through automated testing like coding tasks can.
**答：** 两类人在模型评估上表现突出：Amanda 擅长"塑造 Claude 的性格"，能够为这种本质上模糊的任务清晰地定义成功标准；Claude Code 团队在团队午餐时能快速发现具体的行为模式，比如模型"太突兀"或"喜欢写大量 memories"。Code 团队的定性观察充当了假设生成器，引导定量分析——他们识别潜在问题，然后团队用数据验证，解决了"我们有大量数据，但很难提取洞察"的难题。这种人机协同的评估流程特别有价值，因为性格塑造"需要非常强的信念感"，无法像编码任务那样通过自动化测试来验证。

### Topic 39: Claude's character and personality design
Claude 的性格和个性设计
_[58:45]_

**Q:** Why is Claude's character and personality so important to its success?
**问：** 为什么 Claude 的性格和个性对其成功如此重要？

**A:** Claude's personality is deliberately designed to embody traits of an ideal coworker rather than being a trivial side feature, which fundamentally drives its effectiveness across tasks. The speakers emphasize that users appreciate Claude being "lighthearted and fun" while "extremely competent," creating an approachable yet capable working dynamic. Key personality traits include being "low ego" (responding to corrections with genuine apology and collaboration), maintaining positivity when users face "insurmountable tasks," and providing "earnest feedback" rather than blind agreement. This intentional character design—combining "bias towards action," supportiveness, and honest collaboration—makes Claude "a lot more enjoyable to work with" and explains why users mourned personality changes in OpenClaude, viewing personality as core to Claude's multi-domain success rather than superficial polish.
**答：** Claude 的性格设计刻意模仿理想同事的特质，而非表面功能，这从根本上驱动了它在各类任务中的有效性。用户喜欢 Claude 既"轻松有趣"又"极其胜任"的特点，营造出平易近人但能力强的协作氛围。关键性格特征包括"低自我"（对纠错真诚道歉并协作改进）、在用户面对"难以逾越的任务"时保持积极、提供"真诚反馈"而非盲目附和。这种有意的性格设计——结合"行动偏向"、支持性和诚实协作——让 Claude "工作起来愉快得多"，也解释了为什么用户对 OpenClaude 性格变化感到遗憾，将性格视为 Claude 多领域成功的核心而非表面修饰。

### Topic 40: Removing features as models improve
模型能力提升后的功能精简
_[1:00:45]_

**Q:** How do product features change when new, more capable models are released?
**问：** 当发布更强大的新模型时，产品功能会发生什么变化？

**A:** The team frequently removes features with each model upgrade because many features were originally added "as a crutch for the model" when it couldn't perform tasks naturally. The to-do list feature exemplifies this pattern: early Claude models needed explicit reminders to complete all 20 call sites in a refactor, but Opus 4 "naturally thinks to do everything" without prompting. With each model release, the team systematically reviews the entire system prompt to identify and remove unnecessary "prompting interventions," simplifying the product as model capabilities expand. While some scaffolding features like the to-do list remain for user visibility, they've become "such a deemphasized part of the product" that models may or may not use them.
**答：** 团队在每次模型升级时都会移除功能，因为许多功能最初是作为"模型的拐杖"添加的，用来弥补模型无法自然完成任务的不足。待办清单功能就是典型例子：早期 Claude 模型需要明确提醒才能完成重构中的全部 20 个调用点，但 Opus 4 已经能"自然地想到完成所有事项"而无需提示。团队会在每次发布新模型时系统性地审查整个 system prompt，识别并移除不必要的"提示干预"，随着模型能力增强而简化产品。虽然像待办清单这样的辅助功能因为用户可见性而保留，但它们已经变成"产品中非常不重要的部分"，模型可用可不用。

### Topic 41: Code review product development and model improvements
代码审查产品开发与模型改进
_[01:03:27]_

**Q:** How did newer AI models enable reliable code review features?
**问：** 新的AI模型如何实现可靠的代码审查功能？

**A:** The team had attempted to build code review features multiple times with earlier models, but the accuracy wasn't sufficient to launch a production-ready product beyond "simpler versions" like the slash code review command. Only with the latest models—specifically "Opus 4.5 and 4.6 and Sonnet 4.6"—did they achieve code review quality good enough that "our engineering team relies on this code review to pass before we merge PRs." The breakthrough capability was running "multiple code review agents simultaneously" that could "traverse the entirety of the codebase and synthesize a set of real issues" rather than surface false positives. This represents a pattern of building at the edge of what's currently possible, waiting for model improvements to catch up, and then having a mature product ready when the technology becomes reliable.
**答：** 团队此前多次尝试开发代码审查功能，但早期模型的准确率不足以支撑正式产品，只能推出像 slash code review command 这样的简化版本。直到最新的 Opus 4.5、4.6 和 Sonnet 4.6 模型出现，代码审查质量才达到了工程团队在合并 PR 前必须通过审查的可靠程度。关键突破在于能够同时运行多个代码审查 agent，遍历整个代码库并综合出真正需要解决的问题，而不是产生大量误报。这体现了一种产品策略：在技术边界上提前布局，等待模型能力追上，最终获得成熟可用的产品。

### Topic 42: Building ahead of model capabilities
提前构建超越当前模型能力的产品
_[01:04:39]_

**Q:** What is the strategy for building products before models are fully capable?
**问：** 在模型完全成熟之前构建产品的策略是什么？

**A:** The strategy involves deliberately building products "at the edge of what's working" with the expectation that model improvements will arrive within six months to make them viable. By creating prototypes that "don't necessarily work yet," teams can precisely identify capability gaps and immediately test whether new model releases close those gaps by swapping them into existing prototypes. This approach positions builders ahead of competitors by having infrastructure and product experience ready when the technology catches up.
**答：** 这个策略是有意识地构建处于"当前能力边缘"的产品，预期模型会在六个月内改进到足以支撑产品运行。通过创建"暂时还无法正常工作"的原型，团队能够精确识别能力缺口，并在新模型发布时立即将其替换到现有原型中测试是否弥补了这些缺口。这种方法让构建者在技术成熟时已经具备了基础设施和产品经验，从而领先于竞争对手。

### Topic 43: Vision for Claude Code and Cowork
Claude Code 和 Cowork 的产品愿景
_[01:05:12]_

**Q:** What is the long-term vision for scaling from single tasks to hundreds of concurrent agents?
**问：** 从单任务扩展到数百个并发 agent 的长期愿景是什么？

**A:** The vision follows a progression from "individual tasks" as the "core building block" to massive parallelization as models improve. The team observed users moving from single tasks to "multi-coding" with six tasks simultaneously in late 2024, and extrapolates this will scale to "50 Claudes" or "hundreds of Claudes at a time" as models get smarter. This scaling requires fundamental infrastructure shifts: moving computation off local machines due to RAM constraints, building interfaces that help humans identify "which tasks you need to look into," ensuring agents "fully verify work" so users can quickly trust completed tasks, and creating "self-improving" systems that incorporate feedback to prevent repeated mistakes. The focus is on maintaining task success rates while enabling users to manage massively parallel agent workflows with minimal friction.
**答：** 产品愿景是随着模型能力提升，从单个任务这个"核心构建块"逐步扩展到大规模并行化。团队观察到用户已经从单任务发展到 2024 年底的"multi-coding"（同时处理六个任务），并预测未来会扩展到"同时运行 50 个或数百个 Claude"。这种规模化需要基础设施的根本性转变：因为本地机器 RAM 限制，计算需要迁移到远程；需要构建界面帮助用户识别"哪些任务需要关注"；确保 agent "完全验证工作"让用户能快速信任完成的任务；以及建立"自我改进"系统，能吸收反馈避免重复错误。核心是在保持任务成功率的同时，让用户能以最小摩擦管理大规模并行的 agent 工作流。

### Topic 44: Career advice for thriving in an AI-driven world
AI时代的职业发展策略
_[01:07:23]_

**Q:** How can people leverage AI to focus on creative work and automate tedious tasks?
**问：** 如何利用AI专注于创造性工作并自动化重复性任务？

**A:** The speaker argues that AI provides "a ton more leverage" by automating repetitive manual tasks, allowing professionals to focus on the creative aspects of their work they "absolutely love" rather than the "tedious parts" they hate. The key strategy is to identify tasks you're doing multiple times, use tools like Claude Code or Copilot to automate them with high success rates, and then redirect that freed-up bandwidth—potentially "20% time"—toward high-impact projects that previously lacked resources. The core advice centers on problem-solving: actively look for automation opportunities in your daily workflow and tackle "pet projects" or ideas that have been floating around but never had time to execute.
**答：** 嘉宾认为AI能够大幅提升个人效能，关键在于识别并自动化那些重复性的手动任务，把精力集中在自己真正热爱的创造性工作上。具体做法是找出多次重复的任务，用Claude Code或Copilot等工具实现自动化并不断迭代提高成功率，然后利用释放出的时间（可能有20%的额外带宽）去做那些一直想做但没时间做的高价值项目。核心建议是主动发现问题：关注日常工作中可以自动化的环节，以及那些一直在脑海中但从未实现的想法。

### Topic 45: Finding problems to solve with AI
寻找用AI解决的问题
_[01:09:19]_

**Q:** How do you identify automation opportunities and achieve 100% reliability?
**问：** 如何识别自动化机会并实现100%的可靠性？

**A:** The core strategy is to identify problems by observing repetitive tasks you perform constantly and ideas you haven't had time to implement, essentially "solve a problem for yourself." However, the critical challenge is pushing automations beyond 90-95% accuracy to "100% of the time" reliability, because "if an automation doesn't work 100% of the time, it's not really an automation." This final 5-10% requires significant investment in teaching the system your preferences and providing iterative feedback, even though building the automation is initially slower than doing tasks manually. Both speakers acknowledge struggling with this—Speaker A uses a Gmail spam filter that's "95% great" but occasionally misses important emails, demonstrating that "there's just not much value in a 95% automation."
**答：** 核心策略是观察自己反复执行的任务和一直没时间做的想法，本质上就是"解决自己的问题"。但关键挑战在于把自动化从90-95%的准确率推进到"100%可靠"，因为"如果自动化不是100%有效，它就不算真正的自动化"。最后这5-10%需要大量投入来教会系统你的偏好并持续反馈，尽管搭建自动化初期比手动完成任务更慢。两位嘉宾都承认自己在这方面有困难——Speaker A用Gmail垃圾邮件过滤器"95%很好用"，但偶尔会漏掉重要邮件，说明"95%的自动化没什么价值"。

### Topic 46: Improving customization workflows in Cowork
简化Cowork的自定义工作流程
_[01:11:02]_

**Q:** How is the team simplifying the process of customizing skills and commands?
**问：** 团队如何简化自定义技能和命令的流程？

**A:** The team recognizes that the current customization workflow requires users to understand "too many concepts" including defining skills, providing feedback, explicitly telling Cowork to update based on feedback, and verifying the incorporation of changes. Speaker A illustrates the practical friction with their email filtering workflow that's "95% great" but occasionally misclassifies important messages as spammy, acknowledging this as motivation to refine the system. The team views making this flow "really seamless" as their responsibility, aiming to reduce the cognitive overhead and pain points in the customization process.
**答：** 团队意识到当前的自定义流程要求用户理解"太多概念"，包括定义技能、提供反馈、明确告诉Cowork根据反馈更新，以及验证修改是否正确应用。Speaker A用自己的邮件过滤工作流举例说明实际问题——虽然"95%都很好"，但偶尔会把重要邮件误判为垃圾邮件，这促使他们改进系统。团队认为让整个流程"真正无缝"是他们的职责，目标是降低认知负担和使用痛点。

### Topic 47: Building apps for daily use vs prototypes
构建日常使用的应用而非原型
_[01:11:57]_

**Q:** Why should people focus on building AI apps they use daily rather than just prototypes?
**问：** 为什么人们应该专注于构建日常使用的AI应用而不仅仅是原型？

**A:** Speaker A argues that building AI apps you use "every single day" is essential because only through sustained usage do you extract real value and learning, whereas prototype apps that you "one-shot" and never revisit provide limited insight and no actual leverage. Speaker B extends this by identifying a second trap: people who obsessively customize their workflows with "a ton of skills" and MCPs can become so focused on optimization that they neglect their "core goal of launching some product or building some feature." The speakers acknowledge that customization and hackability have value, but warn there's "a limit to how much it's useful"—the goal is shipping and doing real work, not endlessly perfecting your setup.
**答：** Speaker A 认为构建「每天都在用」的 AI 应用至关重要，因为只有持续使用才能获得真正的价值和学习，而那些「one-shot」后就不再回访的原型应用提供的洞察和实际杠杆都很有限。Speaker B 进一步指出第二个陷阱：有些人痴迷于用「大量 skills」和 MCPs 定制工作流，结果过度专注于优化，反而忽略了「推出产品或构建功能」这个核心目标。两位嘉宾承认定制和可扩展性有价值，但警告「有用程度是有限的」——真正的目标是交付和完成实际工作，而不是无休止地完善你的配置。

### Topic 48: The divide in AI perception and action-based products
AI认知的分歧与基于行动的产品
_[01:13:41]_

**Q:** What is the shift from chat-based to action-based AI products?
**问：** 从基于聊天到基于行动的AI产品的转变是什么？

**A:** Speaker A identifies a fundamental shift where 2024's chat-based AI products have evolved into action-based agents that can execute tasks autonomously rather than just provide instructions. The key insight is that people experience an "aha moment" when they realize "the agent is capable of doing so much more than telling you what to do" - it can actually perform actions independently. This divide in perception stems from early ChatGPT/Claude users who found chat interfaces underwhelming versus developers using AI for coding who witnessed its "full intense power," creating two groups that fundamentally misunderstand each other's experiences. The Chrome extension example illustrates this shift: users can now watch Claude autonomously "fill out this form" rather than being told how to do it.
**答：** Speaker A 指出了一个根本性转变：2024年基于聊天的AI产品已经演变为能够自主执行任务的基于行动的agent，而不仅仅是提供指令。关键洞察在于，当人们意识到"agent能做的远不止告诉你该做什么"——它能真正独立执行操作时，会产生顿悟时刻。这种认知分歧源于早期ChatGPT/Claude用户觉得聊天界面不够好，而用AI编程的开发者见证了它的"full intense power"，导致两个群体根本无法理解彼此的体验。Chrome扩展的例子说明了这一转变：用户现在可以看着Claude自主地"填写表单"，而不是被告知如何操作。

### Topic 49: Lightning round: Book recommendations
快问快答：书籍推荐
_[01:15:22]_

**Q:** What books does Kat recommend about economics, technology transitions, and fiction?
**问：** Kat推荐哪些关于经济学、技术转型和小说的书籍？

**A:** Kat recommends three books spanning economics, technology history, and fiction. She highlights "How Asia Works" as a compelling narrative about "economic development" and the "policies and governments that make long-lasting successful economies." Her second recommendation, "The Technology Trap," examines "past few technology revolutions" including the Industrial and computer revolutions and their impact on workers, which she values because "there's a lot we can learn from history to make sure that this transition goes well." For fiction, she recommends "The Paper Menagerie," describing it as "short stories about coming of age and AI and self-discovery."
**答：** Kat推荐了三本书，涵盖经济学、技术史和小说。她推荐《How Asia Works》，认为这本书讲述了经济发展的故事，探讨了哪些政策和政府能够创造长期成功的经济体。第二本《The Technology Trap》回顾了工业革命和计算机革命等过去几次技术革命对工人的影响，她认为从历史中可以学到很多，以确保当前的技术转型顺利进行。小说方面，她推荐《The Paper Menagerie》，这是一本关于成长、AI和自我发现的短篇小说集。

### Topic 50: Lightning round: Favorite recent show
快问快答：最喜欢的近期节目
_[01:16:29]_

**Q:** What TV show does Kat enjoy watching?
**问：** Kat最喜欢看什么电视节目？

**A:** Kat identifies Drive to Survive as a recent favorite show, emphasizing her enjoyment is straightforward entertainment rather than intellectual engagement. She explicitly notes "there's no deeper meaning to it," suggesting she values the show for pure entertainment value without needing to justify it through educational or artistic merit. This response reveals a comfort with consuming media simply for enjoyment, without the need to frame every viewing choice as culturally significant or professionally relevant.
**答：** Kat提到她很喜欢看Drive to Survive这部剧，并特别强调这只是单纯的娱乐享受，"there's no deeper meaning to it"。她的回答显示出一种坦然的态度——不需要为每个观看选择赋予深层意义或专业价值，纯粹的娱乐本身就足够了。这种直白的表达反映出她在快节奏工作之余，能够放松地享受不需要思考的内容。

### Topic 51: Favorite documentaries about singular pursuits
关于专注追求的纪录片推荐
_[01:16:41]_

**Q:** What documentaries does Speaker A appreciate and why?
**问：** Speaker A 喜欢哪些纪录片，为什么？

**A:** Speaker A is drawn to documentaries that showcase "the purity of their pursuit" around singular, obsessive goals. They particularly admire Free Solo, which documents Alex Honnold's harness-free climb of El Capitan, because it represents "such a pure achievement" requiring extreme mental focus under life-or-death stakes. The appeal lies in watching people pursue engineering or physical challenges with total dedication, where "if you make a single mistake, you die."
**答：** Speaker A 喜欢展现人们对单一目标极致专注的纪录片，欣赏这种追求的纯粹性。他们特别推崇 Free Solo，记录了 Alex Honnold 无保护攀登 El Capitan 的过程，认为这是一项需要极致精神专注的 "pure achievement"，因为 "一个失误就会丧命"。这类纪录片的吸引力在于展现人们在工程或体能挑战上的全情投入和极限追求。

### Topic 52: Personal rock climbing experience and Free Solo appreciation
个人攀岩经历与对 Free Solo 的认知升级
_[01:17:17]_

**Q:** How does Speaker A's climbing experience change their view of Free Solo?
**问：** Speaker A 的攀岩经历如何改变了对 Free Solo 的看法？

**A:** Speaker A describes Free Solo as "one of the rare movies where the more you know about it, the more you're blown away," explaining that their initial impression before climbing was merely that it was impressive, but lacked true comprehension of the difficulty. After becoming a rock climber themselves, they gained visceral understanding of the technical demands, noting that the moves Alex Honnold performs are so advanced that they don't expect to ever execute them "in a gym like one foot off the ground" with safety equipment. This experiential knowledge transformed their appreciation from surface-level admiration to profound awe at the physical and mental achievement.
**答：** Speaker A 认为 Free Solo 是「越了解越震撼」的罕见电影类型。在自己开始攀岩之前，他只是觉得这部片子很厉害，但并不真正理解其难度。成为攀岩者之后，他对技术难度有了切身体会——Alex Honnold 在岩壁上完成的动作，即使是「在离地一英尺的攀岩馆里、有绳索保护」的情况下，他也认为自己这辈子可能都做不到。这种亲身经历将他的认知从表面的钦佩升级为对身心成就的深层敬畏。

### Topic 53: Waymo as a life-changing product
Waymo 如何改变日常生活
_[01:17:56]_

**Q:** What product has most changed Speaker A's life and why?
**问：** 什么产品最大程度改变了 Speaker A 的生活，为什么？

**A:** Speaker A identifies Waymo as the most life-changing non-cloud product, using it twice daily for commuting and valuing two specific benefits that challenge conventional pricing assumptions. The service eliminates social pressure since "I don't feel bad if a Waymo is waiting for me," allowing more flexible timing at pickup. More significantly, it reclaims productivity by creating a judgment-free workspace where A can take work calls without the awkwardness of having a human driver overhear conversations or feeling "rude" using a laptop, effectively giving back "30 minutes every day." Contrary to A's initial belief that autonomous rides needed lower pricing to compete, these psychological and productivity benefits justify paying "a 2x premium" over Uber and Lyft, demonstrating how second-order effects can outweigh direct cost comparisons.
**答：** Speaker A 认为 Waymo 是改变生活最大的非云产品，每天通勤使用两次，并且愿意支付比 Uber 和 Lyft 高一倍的价格。Waymo 带来两个核心价值：一是消除了社交压力，不用担心让车等待或准时到达路边；二是创造了无干扰的移动工作空间，可以自由打工作电话、用电脑而不会感到 "rude"，每天找回约 30 分钟生产力时间。这个案例说明产品的心理价值和效率提升可以超越价格敏感度，颠覆了 A 原本认为自动驾驶必须更便宜才能成功的假设。

### Topic 54: Life motto: Just do things
人生格言：Just do things
_[01:19:30]_

**Q:** What is Speaker A's favorite life motto and how does it apply to work?
**问：** Speaker A 最喜欢的人生格言是什么，如何应用到工作中？

**A:** Speaker A's core philosophy is "just do things," which combines first principles thinking with rapid execution and learning from mistakes. The approach rejects rigid role definitions because "jobs are fake"—if you understand constraints and what you're optimizing for, you can deduce the right action, articulate it to stakeholders, and execute quickly. This mindset, which others call "agency" or "bias towards action," empowers people to operate across team boundaries without waiting for permission, making it particularly valuable in startup environments where formal structures shouldn't block progress.
**答：** Speaker A 的核心理念是 "just do things"，将第一性原理思考与快速执行结合起来，从错误中学习。这种方法拒绝僵化的角色定义，因为 "jobs are fake"——只要理解约束条件和优化目标，就能推导出正确行动，向利益相关者清晰表达，然后快速执行。这种思维方式被称为 "agency" 或 "bias towards action"，让人们能够跨团队边界行动而不必等待许可，在创业公司环境中尤其有价值。

### Topic 55: Value of startup experience for building agency
创业公司经历对培养主动性的价值
_[01:20:50]_

**Q:** Why does Speaker A value startup experience, particularly at Scale?
**问：** 为什么 Speaker A 重视创业公司经历，特别是在 Scale 的经历？

**A:** Speaker A considers early-stage startup experience "life-changing" because it forces you to develop agency by removing traditional role boundaries and processes. At Scale when it was 20 people, the team was empowered to solve "ambitious hairy problems" using "all the tools at your disposal" without being constrained by what sales, ops, or engineering is "supposed to do." This contrasts sharply with traditional education where success means "do the thing we tell you to do," requiring people to actively unlearn that conditioning. The key skill developed is the confidence to "do the thing that needs to be done" even when others might think it's unconventional, which Speaker B notes almost requires direct experience to internalize.
**答：** Speaker A 认为早期创业公司的经历是「改变人生的」，因为它通过打破传统的职能边界和流程，迫使你培养主动性。在 Scale 只有 20 人时，团队被赋权去解决「ambitious hairy problems」，可以使用「all the tools at your disposal」，不受销售、运营或工程「应该做什么」的限制。这与传统教育形成鲜明对比——在学校里成功意味着「do the thing we tell you to do」，需要人们主动去 unlearn 这种思维定式。培养的核心能力是有信心去「do the thing that needs to be done」，即使别人可能觉得不合常规，而 Speaker B 指出这种能力几乎必须通过直接经验才能内化。

### Topic 56: Favorite Claude thinking word
最喜欢的 Claude 思考词
_[01:21:47]_

**Q:** What is Speaker B's favorite Claude thinking word?
**问：** Speaker B 最喜欢的 Claude 思考词是什么？

**A:** Speaker B's favorite thinking word from Claude's internal vocabulary is "manifesting," which they describe with notable enthusiasm. The preference appears to extend beyond just the technical context, as Speaker B mentions having this word as "the sticker that I have on my favorite," suggesting a personal connection to the term. The conversation reveals that these thinking words were discovered through leaked source code, indicating they are part of Claude's internal processing mechanisms that users don't typically see.
**答：** Speaker B 最喜欢的 Claude 思考词是 "manifesting"，并且表现出明显的热情。这个偏好不仅限于技术层面，Speaker B 提到这个词是自己最喜欢的贴纸上的内容，说明对这个词有个人情感联系。对话透露这些思考词是通过泄露的源代码被发现的，表明它们是 Claude 内部处理机制的一部分，用户通常看不到。

### Topic 57: Plans for post-AGI life
AGI 之后的生活规划
_[01:22:10]_

**Q:** What would Speaker A do when AGI arrives and work becomes optional?
**问：** 当 AGI 到来、工作不再是必需时，Speaker A 会做什么？

**A:** Speaker A envisions a two-phase approach: first focusing on "helping bring the world along" during AGI's gradual diffusion across society, then pursuing personal passions once the transition stabilizes. The "non-serious answer" involves relocating to Fontainebleau to rock climb "amongst 10,000 boulders" and dramatically increasing reading from the current "0.5" books per week to "one or two books a week" to tackle a substantial backlog. Despite acknowledging that "the AI will already know it," Speaker A remains excited to learn about domains like physics, robotics, hardware, and aerospace, emphasizing the intrinsic value of human learning and understanding history beyond mere information access.
**答：** Speaker A 的规划分两个阶段：先是在 AGI 逐步扩散到社会各领域时帮助世界适应这个转变，之后再追求个人兴趣。具体计划包括搬到 Fontainebleau 专心攀岩，把阅读量从现在的每周 0.5 本提升到 1-2 本来消化积压的书单。尽管知道 AI 已经掌握了所有知识，Speaker A 依然对学习物理、robotics、硬件、航空航天等领域充满热情，强调人类学习和理解历史的内在价值不只是获取信息本身。

### Topic 58: How to reach out and provide feedback
如何联系团队并提供反馈
_[01:23:26]_

**Q:** How can listeners contact Speaker A and help improve Claude?
**问：** 听众如何联系 Speaker A 并帮助改进 Claude？

**A:** Speaker A is reachable on Twitter as @Catwoo and reads all DMs, though doesn't always respond to each one. The most valuable feedback for the team is not praise but specific "edge cases, errors, like specific tasks" where Claude Code or Cowork fail that can be reproduced, as these directly inform improvements to "next generations of models and our next harnesses." The team actively monitors user feedback through internal channels, posting success stories to a "user love" channel and routing product issues to a feedback channel where the broader team can act on them.
**答：** Speaker A 在 Twitter 上的账号是 @Catwoo，会阅读所有私信但不一定每条都回复。对团队最有价值的反馈不是表扬，而是能够复现的具体失败案例——"edge cases, errors"——这些会直接用于改进下一代模型和工具。团队内部有专门的反馈渠道系统：成功案例会发到 "user love" 频道激励团队，产品问题则进入反馈频道供更广泛的团队采取行动。

### Topic 59: Team engagement with user feedback
团队对用户反馈的参与
_[01:24:44]_

**Q:** How does the Anthropic team handle user feedback and success stories?
**问：** Anthropic 团队如何处理用户反馈和成功案例？

**A:** Anthropic has established a systematic internal feedback loop where user engagement directly energizes the team through dedicated channels. Success stories are collected in a "user love" channel that celebrates positive outcomes, while product issues flow into a separate feedback channel, ensuring the "broader team is able to act on it." This dual-channel approach creates visibility across the organization and transforms external user input into actionable internal momentum.
**答：** Anthropic 建立了系统化的内部反馈机制，用户的积极参与通过专门渠道直接为团队注入能量。成功案例会被收集到 "user love" 频道来庆祝积极成果，而产品问题则进入单独的反馈频道，确保 "broader team is able to act on it"（更广泛的团队能够采取行动）。这种双渠道方式在组织内部创造了可见性，将外部用户输入转化为可执行的内部动力。

---

## Vocabulary (CEFR B2+)

### pilled  /pɪld/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** having adopted a particular mindset or worldview (internet slang, often used as a suffix)  
**CN:** 持有某种特定思维方式或世界观的(网络俚语,常作后缀使用)

**Original examples:**
- [00:00] I think it is very hard to be the right amount of AGI **pilled**.  
  我认为很难对AGI持有恰到好处的信念程度。

**Extra example:**
- After reading that book, he became completely **pilled** on the benefits of remote work.  
  读完那本书后,他完全接受了远程工作的好处这一观念。

### elicit  /ɪˈlɪsɪt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to draw out or bring forth (a response, information, or capability)  
**CN:** 引出，激发，获取（回应、信息或能力）

**Original examples:**
- [00:07] The hard thing is figuring out for the current model, how do you **elicit** the maximum capability?  
  难点在于弄清楚对于当前的模型，你如何引出最大的能力？
- [52:53] I think the hard thing is figuring out for the current model: How do you **elicit** the maximum capability?  
  我认为难点在于弄清楚对于当前的模型：你如何引出最大的能力？

**Extra example:**
- The interviewer used open-ended questions to **elicit** detailed responses from candidates.  
  面试官使用开放式问题来引出候选人的详细回答。

### barrier  /ˈbæriər/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** an obstacle or circumstance that prevents progress or makes it difficult  
**CN:** 障碍，阻碍

**Original examples:**
- [00:17] We want to remove every single **barrier** to shipping things.  
  我们想要移除交付产品的每一个障碍。
- [11:20] I think a lot of it is the process and the expectation on the team. So we're very low on process. We want to remove every single **barrier** to shipping things. We want to make sure every single person on the team feels empowered to take their idea from just an idea to out in the world in less than a week, sometimes.  
  我认为很大程度上是流程和团队的期望。我们的流程非常精简。我们想要移除每一个阻碍发布的障碍。我们希望确保团队中的每个人都有能力把自己的想法从构思阶段推进到实际发布,有时候不到一周就能完成。

**Extra example:**
- Language can be a significant **barrier** to effective communication in international teams.  
  语言可能是国际团队有效沟通的重大障碍。

### iterate  /ˈɪtəreɪt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 3

**EN:** to perform or repeat a process multiple times, making improvements each time  
**CN:** 迭代，反复改进

**Original examples:**
- [00:32] The thing that is extremely important for building AI native products is **iterating** so quickly, figuring out a way for you to actually launch features every single week.  
  构建AI原生产品极其重要的一点是快速迭代，找到一种方法让你真正能够每周发布功能。
- [08:01] This is just an idea, this is just something that we're trying to get feedback on and **iterating** on, and that this might not be supported forever.  
  这只是一个想法,只是我们想要获取反馈并迭代的东西,而且这可能不会永久支持。
- [38:58] You've **iterated** on it, so just to help people try this for themselves. So step one is connect their—what did you say? Slack. What else do you suggest they connect?  
  你已经迭代过了,所以为了帮助大家自己尝试这个。第一步是连接他们的——你刚才说什么来着?Slack。你还建议他们连接什么?

**Extra example:**
- The team **iterates** on the design based on user feedback every sprint.  
  团队根据用户反馈在每个冲刺中对设计进行迭代。

### vision  /ˈvɪʒən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a clear idea or mental image of what the future should be like  
**CN:** 愿景，远见

**Original examples:**
- [02:37] To like that **vision** 3 to 6 months from now. And I spend more of my time on the cross functional, so making sure that our marketing team, sales team, finance, capacity, etc. are like bought in on the plan and that we're all rowing in the same direction and that once the feature is ready that there aren't any blockers to shipping it.  
  一步步实现三到六个月后的愿景。我把更多时间花在跨职能协调上,确保市场、销售、财务、产能等团队都认同这个计划,大家朝着同一个方向努力,并且在功能准备好后不会有任何阻碍发布的因素。

**Extra example:**
- The CEO shared her **vision** for transforming the company into a leader in sustainable technology.  
  CEO分享了她将公司转变为可持续技术领导者的愿景。

### alignment  /əˈlaɪnmənt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** arrangement in a straight line or in correct relative positions; agreement or cooperation  
**CN:** 对齐，一致性，协调

**Original examples:**
- [30:12] And so just to make sure that's clear, so essentially having the number one mission is safety **alignment**, making sure AI is good for the world.  
  为了确保理解清楚,本质上就是把第一使命定为安全对齐,确保 AI 对世界有益。

**Extra example:**
- Achieving **alignment** between different departments is crucial for executing the company strategy.  
  实现不同部门之间的协调对于执行公司战略至关重要。

### ambiguity  /ˌæmbɪˈɡjuːəti/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** the quality of being open to more than one interpretation; uncertainty or vagueness  
**CN:** 模糊性，不确定性

**Original examples:**
- [06:56] I think the first thing is to set clear goals because LLMs are so general that actually creates a lot of **ambiguity** in who we're building for, what problems we're trying to solve, what the top use cases are.  
  我认为首先要设定清晰的目标,因为大语言模型太通用了,这实际上会在我们为谁构建、要解决什么问题、最重要的用例是什么这些方面造成很多模糊性。
- [51:35] I think there's a lot of **ambiguity** in what models are capable of in that timeline and how user behavior will change.  
  我认为在那个时间线内模型能做什么以及用户行为会如何变化存在很多不确定性。

**Extra example:**
- The contract's **ambiguity** led to disputes between the two parties.  
  合同的模糊性导致了双方之间的争议。

### evergreen  /ˈevərɡriːn/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** remaining fresh, relevant, or continuously active over time  
**CN:** 持续保持新鲜、相关或活跃的;长期有效的

**Original examples:**
- [08:31] So when engineers have a feature that they feel is ready and that we've dogfooded internally, they post it in our **Evergreen** launch room.  
  当工程师有一个他们认为已经准备好并且我们内部已经试用过的功能时,他们会把它发布在我们的 Evergreen 发布频道里。

**Extra example:**
- They created an **evergreen** content strategy that continues to attract new users.  
  他们创建了一个长期有效的内容策略,持续吸引新用户。

### friction  /ˈfrɪkʃən/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** resistance or difficulty that slows down progress or makes something harder to do  
**CN:** 摩擦，阻力

**Original examples:**
- [08:51] And because we have this really tight process, it lowers the **friction** for any engineer to ship something, and PM is the role that should be setting this up.  
  因为我们有这个非常严密的流程，它降低了任何工程师交付产品的阻力，而PM就是应该建立这个流程的角色。

**Extra example:**
- The new onboarding process reduced **friction** for new users signing up for the service.  
  新的入职流程减少了新用户注册服务的阻力。

### trade off  /treɪd ɒf/
**CEFR:** C1 | **Part of speech:** n./v. | **Occurrences:** 1

**EN:** a balance achieved between two desirable but incompatible features; to exchange one thing for another  
**CN:** 权衡，取舍

**Original examples:**
- [09:29] The second thing that we do is we have this list of team principles. And this includes who our key users are, why those are our key users. And the reason that we articulate all of this is so that everybody on the team feels like they understand how our business works. They understand what's important to us and what we're willing to **trade off**. And it lets people make decisions by themselves without feeling like they're  
  我们做的第二件事是有一份团队原则清单。这包括我们的核心用户是谁、为什么他们是我们的核心用户。我们阐明所有这些的原因是让团队中的每个人都觉得他们理解我们的业务如何运作、什么对我们重要、我们愿意做出什么权衡。这让人们可以自己做决定,而不会觉得被——

**Extra example:**
- There's always a **trade-off** between speed and quality in product development.  
  在产品开发中，速度和质量之间总是需要权衡。

### empowered  /ɪmˈpaʊərd/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** having the authority, confidence, or ability to do something independently  
**CN:** 被赋权的，有能力独立行动的

**Original examples:**
- [11:40] Our top goal is to help empower professional developers, making everyone feel **empowered** to ship independently.  
  我们的首要目标是帮助赋能专业开发者，让每个人都感到有能力独立发布产品。

**Extra example:**
- The new policy made employees feel more **empowered** to make decisions.  
  新政策让员工感到更有权力做决策。

### subsidize  /ˈsʌbsɪdaɪz/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to support financially by paying part of the costs  
**CN:** 补贴，资助

**Original examples:**
- [14:02] Like you guys are **subsidizing** this usage at like 200 bucks a month and there's like it's like basically unlimited use of this and like...  
  你们在以每月200美元的价格**补贴**这种使用，而且基本上是无限使用...

**Extra example:**
- The government decided to **subsidize** renewable energy projects.  
  政府决定补贴可再生能源项目。

### taste  /teɪst/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 5

**EN:** the ability to make good judgments about quality, style, or what is appropriate  
**CN:** 品味，鉴赏力，判断力

**Original examples:**
- [16:26] You can either hire a lot more engineers who have great product **taste** or you can keep your engineering hiring the same and hire a lot more PMs to help guide some of their work.  
  你可以雇佣更多有出色产品**品味**的工程师，或者保持工程师招聘不变，雇佣更多PM来指导他们的工作。
- [16:36] On our team we're pretty focused on hiring engineers with great product **taste**.  
  在我们团队，我们非常注重招聘有出色产品**品味**的工程师。
- [17:14] I think product **taste** is still a very rare skill to have and we'll pretty much hire anyone who we feel has demonstrated this strongly.  
  我认为产品**品味**仍然是一项非常稀缺的技能，我们基本上会雇佣任何能强烈展现这一点的人。
- [18:16] I still think it comes back to product **taste**.  
  我仍然认为这归结到产品**品味**。
- [18:28] We get tens of thousands of GitHub issues asking for every single thing under the sun, and it takes a lot of care and **taste** to figure out, okay, which of these is worth building and what is the right way to build it.  
  我们收到成千上万个GitHub issue，要求各种各样的功能，需要大量的用心和**品味**来判断，好吧，哪些值得开发，以及正确的开发方式是什么。

**Extra example:**
- Her **taste** in design helped the company create more appealing products.  
  她在设计上的**品味**帮助公司创造了更吸引人的产品。

### overhead  /ˈoʊvərhɛd/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the ongoing costs or extra work required to maintain a process or system  
**CN:** 管理费用，额外开销，负担

**Original examples:**
- [16:36] On our team we're pretty focused on hiring engineers with great product taste. This way we can reduce the amount of **overhead** for shipping any product.  
  在我们团队，我们非常注重招聘有出色产品品味的工程师。这样我们可以减少发布任何产品的**额外开销**。

**Extra example:**
- The new system reduced administrative **overhead** by 30%.  
  新系统将行政管理**开销**减少了30%。

### prioritization  /praɪˌɔrɪtəˈzeɪʃən/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the process of deciding the order of importance of tasks or goals  
**CN:** 优先级排序，确定优先顺序

**Original examples:**
- [19:23] So it helps a bit with the **prioritization**.  
  所以这对**优先级排序**有一些帮助。

**Extra example:**
- Effective **prioritization** is crucial when managing multiple projects.  
  在管理多个项目时，有效的**优先级排序**至关重要。

### tacit  /ˈtæsɪt/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** understood or implied without being directly stated; implicit knowledge gained through experience  
**CN:** 心照不宣的，隐性的，默会的

**Original examples:**
- [22:09] I think a lot of this more **tacit** common sense, like EQ kind of knowledge, is still very valuable.  
  我认为很多这种更**隐性的**常识，比如情商类的知识，仍然非常有价值。

**Extra example:**
- Much of the team's expertise is **tacit** knowledge that's hard to document.  
  团队的很多专业知识是难以记录的**隐性**知识。

### brutally  /ˈbruːtəli/
**CEFR:** C1 | **Part of speech:** adv. | **Occurrences:** 1

**EN:** in an extremely direct, harsh, or uncompromising manner  
**CN:** 残酷地，毫不留情地，极其严格地

**Original examples:**
- [23:40] But I think you just have to acknowledge that there's only so much that you can do, that you need to sleep well so that you can make good decisions the next day and just like **brutally** prioritize where you spend your time.  
  但我认为你必须承认你能做的有限，你需要睡好觉才能在第二天做出好决策，并且要**毫不留情地**确定时间花在哪里的优先级。

**Extra example:**
- We need to **brutally** cut costs to survive this downturn.  
  我们需要**毫不留情地**削减成本才能度过这次低迷期。

### polished  /ˈpɑːlɪʃt/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** refined and perfected in quality or appearance; showing high level of finish  
**CN:** 精致的，完善的，打磨过的

**Original examples:**
- [24:16] Like there's products that we ship that aren't as **polished** as I wish they were.  
  有些我们发布的产品并不像我希望的那样**精致**。

**Extra example:**
- The presentation was highly **polished** and professional.  
  这个演示非常**精致**和专业。

### consistency  /kənˈsɪstənsi/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** the quality of always behaving or performing in a similar way, or of always happening in a similar way  
**CN:** 一致性，连贯性

**Original examples:**
- [25:42] We're sacrificing product **consistency**.  
  我们正在牺牲产品的一致性。
- [25:42] We're sacrificing product **consistency**. Historically, when code was expensive to write, you would carefully plan out everything in your product suite, how every product relates to each other, what the use case for every single one is, how they integrate, and you would...  
  我们牺牲的是产品一致性。以前,当写代码成本很高的时候,你会仔细规划产品套件中的所有内容,每个产品之间的关系、每个产品的使用场景、它们如何集成,然后你会

**Extra example:**
- The team struggled to maintain **consistency** across different platforms.  
  团队在不同平台上难以保持一致性。

### agentic  /eɪˈdʒentɪk/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** relating to AI systems that can act autonomously and make decisions independently to achieve goals  
**CN:** 智能体的，自主行动的（指能够自主决策和行动的AI系统）

**Original examples:**
- [27:09] I think with these **agentic** tools, not just Coda Code and Cowork, but like across the whole ecosystem, people feel this need to like check Twitter every single day to see what the absolute latest thing is.  
  我认为有了这些智能体工具，不仅仅是Coda Code和Cowork，而是整个生态系统，人们感到需要每天查看Twitter，看看最新的进展是什么。

**Extra example:**
- **Agentic** AI systems can plan and execute complex tasks without constant human supervision.  
  智能体AI系统可以在没有持续人工监督的情况下规划和执行复杂任务。

### unify  /ˈjuːnɪfaɪ/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to bring together different elements or groups into a single coherent whole  
**CN:** 统一，使一致

**Original examples:**
- [29:29] The two most important things are, one, this **unifying** mission.  
  最重要的两件事是，第一，这个统一的使命。
- [29:29] The two most important things are, one, this **unifying** mission. It's hard to state how important this is. We hire people who care most about bringing safe AGI to all of humanity, and this is actually something that we reference frequently in our decisions about what our entire product org should focus on shipping. And because we put this mission above—  
  最重要的两点是:第一,这个统一的使命。很难说清楚这有多重要。我们招聘的都是最关心为全人类带来安全 AGI 的人,而且我们在决定整个产品组织应该专注交付什么时,会经常参考这个使命。正因为我们把这个使命放在——

**Extra example:**
- The new platform will **unify** all customer data into one system.  
  新平台将把所有客户数据统一到一个系统中。

### deprioritize  /diːpraɪˈɒrɪtaɪz/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to reduce the importance or urgency of something relative to other tasks or goals  
**CN:** 降低优先级，不再优先考虑

**Original examples:**
- [30:43] And so we **deprioritize** shipping this and we just wait until later.  
  所以我们降低了发布这个功能的优先级，等到以后再说。

**Extra example:**
- The company decided to **deprioritize** minor bug fixes to focus on the new feature launch.  
  公司决定降低小bug修复的优先级，专注于新功能的发布。

### tradeoff  /ˈtreɪdɒf/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a balance achieved between two desirable but incompatible features; a compromise involving the sacrifice of one thing for another  
**CN:** 权衡，取舍

**Original examples:**
- [31:10] And people are very happy to make those **trade-offs**.  
  人们非常乐意做出这些权衡。

**Extra example:**
- There's always a **tradeoff** between speed and quality in software development.  
  在软件开发中，速度和质量之间总是存在权衡。

### sacrifice  /ˈsækrɪfaɪs/
**CEFR:** B2 | **Part of speech:** n./v. | **Occurrences:** 1

**EN:** to give up something valuable or important for the sake of other considerations  
**CN:** 牺牲，舍弃

**Original examples:**
- [31:10] Mission means that teams are willing to make **sacrifices** that hurt their own goals and their own KPIs in service of Anthropic's goals and Anthropic's KPIs.  
  使命意味着团队愿意做出牺牲，损害自己的目标和KPI，以服务于Anthropic的目标和KPI。

**Extra example:**
- Success often requires **sacrificing** short-term gains for long-term benefits.  
  成功往往需要牺牲短期收益来换取长期利益。

### synthesize  /ˈsɪnθəsaɪz/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 3

**EN:** to combine different ideas, information, or elements into a coherent whole  
**CN:** 综合，合成

**Original examples:**
- [37:53] And it **synthesized** all this together to this 20-page deck that I woke up to this morning, and I read through it and it was like pretty good.  
  它把所有这些综合成了一份20页的文档，我今天早上醒来看到的，读完后觉得相当不错。
- [40:02] It's able to **synthesize** a massive amount of information really quickly and present all of the possibilities to you.  
  它能够非常快速地综合大量信息，并向你展示所有可能性。
- [01:04:21] Code review agents simultaneously traverse the entirety of the codebase and **synthesize** a set of real issues that an engineer needs to address before merge.  
  代码审查智能体同时遍历整个代码库，并综合出工程师在合并前需要解决的一系列实际问题。

**Extra example:**
- The report **synthesizes** findings from multiple research studies.  
  这份报告综合了多项研究的发现。

### durable  /ˈdʊrəbl/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** able to last or remain effective for a long time  
**CN:** 持久的;耐用的;能长期保持有效的

**Original examples:**
- [44:53] It's so interesting like people talk about Salesforce as just like SaaS. We don't need SaaS software anymore. We're going to build our own. It's like Slack is a **durable** tool that nobody wants to try to compete with and build a better version.  
  特别有意思的是,大家都在说 Salesforce 只是 SaaS 软件,我们不再需要 SaaS 了,我们要自己做。但 Slack 是个持久耐用的工具,没人想去竞争或者做个更好的版本。

**Extra example:**
- Building a **durable** competitive advantage requires continuous innovation.  
  建立持久的竞争优势需要持续创新。

### automate  /ˈɔːtəmeɪt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make a process or system operate automatically without human intervention  
**CN:** 使自动化

**Original examples:**
- [46:19] Vanta **automates** compliance and risk management with over 35 security and privacy frameworks including SOC 2, ISO 27001 and HIPAA.  
  Vanta自动化了合规和风险管理，支持超过35个安全和隐私框架，包括SOC 2、ISO 27001和HIPAA。

**Extra example:**
- The company plans to **automate** repetitive tasks to improve efficiency.  
  公司计划将重复性任务自动化以提高效率。

### delegate  /ˈdelɪɡeɪt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to give a particular job, duty, right, etc. to someone else so that they do it for you  
**CN:** 委派，授权；把（工作、任务等）交给他人处理

**Original examples:**
- [49:50] It is clear to us that as the models get better, people **delegate** far more tasks to it and they spend a lot more hours in tools like Claude Code and CoWork.  
  我们很清楚，随着模型变得更好，人们会把更多任务委派给它，并在 Claude Code 和 CoWork 这样的工具上花费更多时间。
- [54:04] I **delegated** the verification to this sub-agent and the sub-agent didn't do the test and I didn't check its work.  
  我把验证工作委派给了这个子代理，但子代理没有做测试，而我也没有检查它的工作。

**Extra example:**
- As a manager, you need to learn to **delegate** responsibilities effectively.  
  作为管理者，你需要学会有效地委派职责。

### flywheel  /ˈflaɪwiːl/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a heavy wheel in a machine that helps the machine work at a steady speed; in business, a self-reinforcing cycle where success builds on itself  
**CN:** 飞轮；（商业中的）自我强化循环，良性循环

**Original examples:**
- [50:42] It's so interesting how many advantages come from having the most advanced model. It's such an interesting **flywheel** that starts to kick in.  
  拥有最先进的模型能带来如此多的优势，真是太有意思了。这是一个非常有趣的飞轮效应开始发挥作用。

**Extra example:**
- Amazon's **flywheel** effect means lower prices attract more customers, which attracts more sellers, leading to even lower prices.  
  Amazon 的飞轮效应意味着更低的价格吸引更多顾客，从而吸引更多卖家，进而带来更低的价格。

### introspect  /ˌɪntrəˈspekt/
**CEFR:** C2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to examine and consider your own ideas, thoughts, and feelings  
**CN:** 内省，反省；自我审视

**Original examples:**
- [53:32] One of the things I really like to do is to ask the model to **introspect** on its own behaviors.  
  我真正喜欢做的一件事是让模型对自己的行为进行内省。

**Extra example:**
- Taking time to **introspect** can help you understand your motivations better.  
  花时间内省可以帮助你更好地理解自己的动机。

### harness  /ˈhɑːrnɪs/
**CEFR:** C1 | **Part of speech:** n./v. | **Occurrences:** 3

**EN:** a set of straps for controlling an animal; a system or framework for controlling something; to control and use the force or strength of something  
**CN:** （控制动物的）挽具；（控制系统的）框架，工具；驾驭，利用

**Original examples:**
- [54:17] A lot of times, just like being very curious about why the model made the decision that it did will show you what misled it so that you can fix the **harness** in order to close this gap.  
  很多时候，对模型为什么做出这样的决定保持好奇心，会让你看到是什么误导了它，这样你就可以修复这个控制框架来弥补这个差距。
- [54:42] Usually there's like a handful of people who are much better than others at articulating what makes a specific model or model **harness** combination good.  
  通常只有少数人比其他人更擅长阐述是什么让特定的模型或模型控制框架组合变得优秀。
- [01:02:44] I forget who said this on the podcast that the model will eat your **harness** for breakfast. And what I'm hearing here is essentially you remove things over time that you've had to add on top of the model where it was not operating the way you wanted. And essentially as the models get smarter, it just becomes simpler and simpler for it to do the thing you want it to do.  
  我忘了是谁在播客里说过「模型会把你的脚手架当早餐吃掉」。我现在听到的本质上就是,你们会逐渐移除那些为了让模型按预期工作而不得不添加的东西。随着模型变得越来越智能,让它做你想做的事情就变得越来越简单。

**Extra example:**
- Companies are learning to **harness** the power of AI to improve customer service.  
  公司正在学习如何驾驭 AI 的力量来改善客户服务。

### eval  /ɪˈvæl/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 5

**EN:** short for evaluation; a test or assessment to measure performance or quality, especially in AI/ML contexts  
**CN:** 评估，测评（evaluation 的缩写）；（尤指 AI/机器学习领域的）性能测试

**Original examples:**
- [55:03] I think **evals** is this underappreciated thing that more PMs, more engineers should be working on.  
  我认为评估测试是一个被低估的东西，更多的产品经理和工程师应该在这方面投入工作。
- [55:33] There's this trend of just like that is the future of product management is writing **evals** because essentially it's what does success look like?  
  有这样一个趋势，就是编写评估测试是产品管理的未来，因为本质上这就是在定义成功是什么样子。
- [55:46] I think the importance of **evals** varies a bit based on the feature that you're working on and or like what the problem you're trying to solve is.  
  我认为评估测试的重要性会根据你正在开发的功能或你试图解决的问题而有所不同。
- [56:15] I personally jump into **evals** when there's a feature that I think needs a bit more product definition and often the output of this is okay here are like five **evals** that I made.  
  当我认为某个功能需要更多产品定义时，我个人会投入到评估测试中，通常输出结果是，好的，这里有我做的大约五个评估测试。
- [56:45] Point you made about people being very good at evaluating models so interesting. It's almost like a human **eval** of just like okay they understand where it's spiking or it's maybe lacking. Is there  
  你提到的那个观点很有意思——人们非常擅长评估模型。这几乎就像是一种人工评估,他们能理解模型在哪些方面表现突出,哪些方面可能有所欠缺。是否有

**Extra example:**
- We run comprehensive **evals** to ensure the model performs well across different scenarios.  
  我们运行全面的评估测试以确保模型在不同场景下都表现良好。

### conviction  /kənˈvɪkʃn/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a strong opinion or belief; the feeling of being certain about something  
**CN:** 坚定的信念；确信，深信

**Original examples:**
- [57:15] Even coding is easier because you can verify the success, whereas crafting the character requires a very strong sense of **conviction** in who Claude should be.  
  即使是编程也更容易，因为你可以验证成功与否，而塑造角色需要对 Claude 应该是什么样子有非常坚定的信念。

**Extra example:**
- She spoke with **conviction** about the need for educational reform.  
  她坚定地谈到了教育改革的必要性。

### vibe  /vaɪb/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a general feeling or impression about something (informal)  
**CN:** 对某事的总体感觉或印象(非正式用语)

**Original examples:**
- [57:49] One of the fastest ways for us to get feedback is to just like at these team lunches just like go to every single person and just be like, "Hey, what is your **vibe** on the model?"  
  我们获取反馈最快的方式之一就是在团队午餐时,去问每个人,'嘿,你对这个模型的感觉如何?'

**Extra example:**
- I'm getting a positive **vibe** from the client about our proposal.  
  我从客户那里感受到了对我们提案的积极反馈。

### crutch  /krʌtʃ/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a stick that you put under your arm to help you walk when you have hurt your leg; something that provides support or help, especially when used to compensate for a weakness  
**CN:** 拐杖；（弥补弱点的）支撑物，依赖物

**Original examples:**
- [01:01:10] So a lot of times we add features to the product as a **crutch** for the model because it's not naturally doing it itself.  
  所以很多时候我们会在产品中添加功能作为模型的拐杖，因为它本身不会自然地做到这一点。

**Extra example:**
- Don't use the calculator as a **crutch**—try to do the math in your head first.  
  不要把计算器当作拐杖——先试着在脑子里算一算。

### leverage  /ˈlevərɪdʒ/
**CEFR:** C1 | **Part of speech:** n./v. | **Occurrences:** 2

**EN:** the ability to influence or achieve more with the same amount of effort or resources; to use something to maximum advantage  
**CN:** 杠杆作用；影响力；（用更少资源）实现更多的能力；充分利用

**Original examples:**
- [01:07:49] I think AI gives everybody a ton more **leverage** than they used to.  
  我认为 AI 给每个人带来了比以前多得多的杠杆作用。
- [01:12:45] And you're not getting much **leverage** from it—  
  而你并没有从中获得多少杠杆作用——

**Extra example:**
- By automating routine tasks, you can **leverage** your time to focus on strategic work.  
  通过自动化日常任务，你可以充分利用时间专注于战略性工作。

### tedious  /ˈtiːdiəs/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** too long, slow, or dull; tiresome or monotonous  
**CN:** 冗长乏味的，单调枯燥的

**Original examples:**
- [01:08:04] Most people have like creative parts of their job that they absolutely love and then like **tedious** parts of their job that they really hate doing.  
  大多数人的工作中既有他们非常热爱的创造性部分，也有他们真的很讨厌做的**枯燥乏味**的部分。
- [01:08:14] I think the beauty of AI is that it can do those **tedious** parts for you.  
  我认为AI的美妙之处在于它可以为你完成那些**枯燥乏味**的部分。

**Extra example:**
- Data entry can be a **tedious** task that requires patience.  
  数据录入可能是一项需要耐心的**枯燥乏味**的任务。

### bandwidth  /ˈbændwɪdθ/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** (figurative) the capacity or time available to deal with a situation  
**CN:** （比喻）处理事务的能力、时间或精力

**Original examples:**
- [01:08:50] It can learn from every time that you've done that manual task and generalize and then run it automatically so that you can focus on the creative parts and that means you can do a lot more than you used to be able to do with your **bandwidth**.  
  它可以从你每次完成手动任务中学习并归纳，然后自动运行，这样你就可以专注于创造性部分，这意味着你可以用你的**精力**做比以前多得多的事情。
- [01:09:01] AI enables you to expand your **bandwidth** and accomplish tasks you couldn't before.  
  AI使你能够扩展你的**能力范围**，完成以前无法完成的任务。

**Extra example:**
- I don't have the **bandwidth** to take on another project right now.  
  我现在没有**精力**再接另一个项目了。

### cynical  /ˈsɪnɪkəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** believing that people are motivated purely by self-interest; distrustful of human sincerity or integrity  
**CN:** 愤世嫉俗的，怀疑的，不信任的

**Original examples:**
- [01:14:38] Some people remain **cynical** about AI's ability to truly understand context and deliver meaningful results.  
  有些人对AI真正理解上下文并提供有意义结果的能力持**怀疑态度**。

**Extra example:**
- He has a **cynical** view of politics and doesn't trust any politicians.  
  他对政治持**怀疑态度**，不信任任何政客。

### deduce  /dɪˈduːs/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** arrive at a conclusion by reasoning; infer from available information  
**CN:** 推断，演绎，推论

**Original examples:**
- [01:19:38] I think there's a lot of value in first principles thinking and if you know what you're optimizing for and you have strong first principles, then you can normally **deduce** what the right course of action is and be able to clearly articulate that to all the stakeholders.  
  我认为第一性原理思维很有价值，如果你知道自己在优化什么，并且有强大的第一性原理，那么你通常可以**推断出**正确的行动方案，并能够向所有利益相关者清楚地表达出来。

**Extra example:**
- From the evidence presented, the detective was able to **deduce** who committed the crime.  
  根据呈现的证据，侦探能够**推断出**是谁犯下了罪行。

### articulate  /ɑːrˈtɪkjuleɪt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** express an idea or feeling fluently and coherently  
**CN:** 清楚表达，明确阐述

**Original examples:**
- [01:19:38] If you know what you're optimizing for and you have strong first principles, then you can normally deduce what the right course of action is and be able to clearly **articulate** that to all the stakeholders.  
  如果你知道自己在优化什么，并且有强大的第一性原理，那么你通常可以推断出正确的行动方案，并能够向所有利益相关者清楚地**阐述**出来。

**Extra example:**
- She was able to **articulate** her vision for the company in a way that inspired the entire team.  
  她能够以一种激励整个团队的方式**阐述**她对公司的愿景。

### empower  /ɪmˈpaʊər/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 2

**EN:** give someone the authority or power to do something; make someone stronger and more confident  
**CN:** 授权，使能够，赋予力量

**Original examples:**
- [01:20:10] I think what 'just do things' lets people do is they feel **empowered** to make these decisions, **empowered** to operate across team boundaries just to get something done.  
  我认为'just do things'让人们能够做的是，他们感到被**赋予权力**去做这些决定，被**赋予权力**跨越团队边界去完成某件事。
- [01:21:04] I really appreciate Alex and the rest of the team for like **empowering** me and the rest of the team to just like figure things out without any boundaries.  
  我真的很感激Alex和团队其他成员**授权**我和团队其他人在没有任何界限的情况下去解决问题。

**Extra example:**
- Education **empowers** individuals to make informed decisions about their future.  
  教育**赋予**个人对自己未来做出明智决定的能力。

### agency  /ˈeɪdʒənsi/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the capacity of individuals to act independently and make their own free choices  
**CN:** 自主性，能动性，行动力

**Original examples:**
- [01:20:38] That feels like a big important skill to be good at. People call it **agency**.  
  这感觉是一项很重要的技能。人们称之为**能动性**。

**Extra example:**
- The program aims to give students more **agency** over their learning process.  
  该项目旨在让学生对自己的学习过程拥有更多**自主权**。

### bias  /ˈbaɪəs/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a tendency, inclination, or prejudice toward or against something  
**CN:** 倾向，偏好，偏见

**Original examples:**
- [01:20:46] **Bias** towards action. All these ways of describing just like, don't wait for permission.  
  行动**偏好**。所有这些描述方式都像是在说，不要等待许可。

**Extra example:**
- Successful entrepreneurs often have a **bias** toward taking calculated risks.  
  成功的企业家往往有承担经过计算的风险的**倾向**。

### diffuse  /dɪˈfjuːz/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to spread or cause to spread over a wide area or among a large number of people  
**CN:** 扩散，传播（到广泛区域或大量人群中）

**Original examples:**
- [01:22:23] I think it will take a long time for AGI to **diffuse** across society.  
  我认为AGI要扩散到整个社会需要很长时间。

**Extra example:**
- New technologies often take years to **diffuse** throughout an industry.  
  新技术往往需要数年时间才能在整个行业中扩散开来。

### cross-functional  /ˌkrɒs ˈfʌŋkʃənl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** involving or combining different departments, skills, or areas of expertise within an organization  
**CN:** 跨职能的，涉及不同部门或专业领域的

**Original examples:**
- [02:37] And I spend more of my time on the **cross-functional**, so making sure that our marketing team, sales team, finance, capacity, etc. are like bought in on the plan.  
  我把更多时间花在跨职能协调上，确保我们的营销团队、销售团队、财务、产能等都认同这个计划。
- [08:17] And then the third thing that a PM should do is help create the framework for the team so that they know when to pull in **cross-functional** partners and what those **cross-functional** partners' expectations are.  
  PM应该做的第三件事是帮助团队建立框架，让他们知道何时引入跨职能合作伙伴，以及这些跨职能合作伙伴的期望是什么。

**Extra example:**
- The project requires a **cross-functional** team with members from engineering, design, and business.  
  这个项目需要一个跨职能团队，成员来自工程、设计和商务部门。

### commitment  /kəˈmɪtmənt/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a promise or firm decision to do something; an obligation or responsibility  
**CN:** 承诺，保证；义务，责任

**Original examples:**
- [08:10] It reduces our **commitment** for shipping something.  
  这降低了我们发布产品的承诺程度。

**Extra example:**
- The company made a **commitment** to reduce carbon emissions by 50% within five years.  
  公司承诺在五年内将碳排放减少50%。

### dogfood  /ˈdɒɡfuːd/
**CEFR:** C2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** (in tech industry) to use one's own product internally before releasing it to customers, in order to test and improve it  
**CN:** （科技行业）内部试用自己的产品（在向客户发布前进行测试和改进）

**Original examples:**
- [08:31] So when engineers have a feature that they feel is ready and that we've **dogfooded** internally, they post it in our Evergreen launch room.  
  当工程师有一个他们认为已经准备好并且我们内部已经试用过的功能时,他们会把它发布在我们的 Evergreen 发布频道里。

**Extra example:**
- We always **dogfood** our software for at least two weeks before the public release.  
  我们总是在公开发布前至少内部试用我们的软件两周。

### harden  /ˈhɑːrdən/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make a system, process, or security measure stronger and more resistant to failure or attack  
**CN:** 加固，强化（系统、流程或安全措施）

**Original examples:**
- [12:33] And so this was a result of human error and we've **hardened** our processes to make sure that it doesn't happen in the future.  
  这是人为失误造成的，我们已经加固了流程以确保未来不会再发生。

**Extra example:**
- The IT team worked to **harden** the network against potential cyber attacks.  
  IT团队努力加固网络以防范潜在的网络攻击。

### safeguard  /ˈseɪfɡɑːrd/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a measure or action taken to protect someone or something or to prevent something undesirable  
**CN:** 保护措施，防护措施

**Original examples:**
- [12:42] Yes, yes. It's a process failure and the most important thing is to just learn from it and to add more **safeguards** so that doesn't happen again.  
  是的,还在。这是一个流程失败,最重要的是从中吸取教训,并增加更多的保障措施,这样就不会再发生了。

**Extra example:**
- The new policy includes several **safeguards** to protect user privacy.  
  新政策包含了几项保护用户隐私的防护措施。

### seamless  /ˈsiːmləs/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** smooth and continuous, with no apparent gaps or spaces between one part and the next  
**CN:** 无缝的，流畅的，平滑过渡的

**Original examples:**
- [13:36] We spent a bunch of time trying to figure out what is the most **seamless** transition that we can offer.  
  我们花了很多时间试图找出我们能提供的最流畅的过渡方案。

**Extra example:**
- The app provides a **seamless** experience across all devices.  
  这款应用在所有设备上都提供了无缝的使用体验。

### shepherd  /ˈʃepərd/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to guide or direct a project, process, or group of people carefully through various stages  
**CN:** 引导，指导（项目、流程或团队经历各个阶段）

**Original examples:**
- [14:31] So we have the research PM team who Diane leads, and this team is responsible for understanding all of the feedback from our customers for our models and then feeding that to the research team to act on it, and they also **shepherd** the model launch.  
  我们有Diane领导的研究PM团队，这个团队负责了解客户对我们模型的所有反馈，然后将反馈传递给研究团队采取行动，他们还负责引导模型发布。

**Extra example:**
- The project manager will **shepherd** the initiative through all approval stages.  
  项目经理将引导这项计划通过所有审批阶段。

### amorphous  /əˈmɔːrfəs/
**CEFR:** C2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** without a clearly defined shape or form; vague or ill-organized  
**CN:** 无定形的；模糊的，无组织的

**Original examples:**
- [20:28] The nature of work is becoming increasingly **amorphous**, requiring people who can adapt and wear many hats.  
  工作的性质正变得越来越模糊不定，需要能够适应并身兼多职的人。

**Extra example:**
- The company's strategy remained **amorphous**, lacking clear direction or priorities.  
  公司的战略仍然模糊不清，缺乏明确的方向或优先事项。

### stakeholder  /ˈsteɪkhoʊldər/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a person or group with an interest or concern in something, especially a business  
**CN:** 利益相关者，股东

**Original examples:**
- [21:59] I think the model doesn't always have a great sense of who all the **stakeholders** are, how they relate to each other, what their preferences are, what are the right venues to communicate with them to keep them on board.  
  我认为模型并不总能很好地了解所有利益相关者是谁，他们之间的关系如何，他们的偏好是什么，以及与他们沟通以保持他们支持的正确渠道是什么。
- [01:19:38] I think there's a lot of value in first principles thinking and if you know what you're optimizing for and you have strong first principles, then you can normally deduce what the right course of action is and be able to clearly articulate that to all the **stakeholders** and then you should just do it.  
  我认为第一性原理思维很有价值，如果你知道自己在优化什么，并且有强大的第一性原理，那么你通常可以推断出正确的行动方案，并能够向所有利益相关者清楚地阐述，然后就应该去做。

**Extra example:**
- The project manager scheduled a meeting with all key **stakeholders** to discuss the timeline.  
  项目经理安排了一次会议，与所有关键利益相关者讨论时间表。

### overlap  /ˌoʊvərˈlæp/
**CEFR:** B2 | **Part of speech:** n./v. | **Occurrences:** 1

**EN:** extend over so as to cover partly; a part or amount which overlaps  
**CN:** 重叠；重复部分

**Original examples:**
- [26:11] A lot of the times it's because there's two form factors that we love internally and we want the external audience to tell us which one is better, leading to feature **overlap**.  
  很多时候是因为我们内部喜欢两种形式，我们想让外部受众告诉我们哪一个更好，这导致了功能重叠。

**Extra example:**
- There's significant **overlap** between the marketing and sales teams' responsibilities.  
  市场营销团队和销售团队的职责之间存在显著重叠。

### intuitive  /ɪnˈtuːɪtɪv/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** easy to understand or operate without explicit instruction; based on instinct  
**CN:** 直观的；凭直觉的

**Original examples:**
- [27:59] We didn't actually want to do something like PowerUp because we felt like the product should be **intuitive** enough that you don't actually need to go through any tutorial.  
  我们其实不想做像PowerUp这样的东西，因为我们觉得产品应该足够直观，你实际上不需要经过任何教程。

**Extra example:**
- The app's interface is so **intuitive** that even first-time users can navigate it easily.  
  这个应用的界面非常直观，即使是首次使用的用户也能轻松操作。

### diverge  /daɪˈvɜːrdʒ/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** separate from another route and go in a different direction; differ or deviate  
**CN:** 分歧；偏离

**Original examples:**
- [28:08] And over time, we've just realized that there's just so many features and there's so much demand for a built-in onboarding experience that we **diverged** a bit from our original principle saying no onboarding flow and added this because there's just so many users who wanted to know there's 100 features.  
  随着时间的推移，我们意识到有太多功能，对内置引导体验的需求也很大，所以我们稍微偏离了最初不做引导流程的原则，添加了这个功能，因为有太多用户想知道有100个功能。

**Extra example:**
- Their opinions began to **diverge** on how to approach the problem.  
  他们对如何处理这个问题的意见开始出现分歧。

### constraint  /kənˈstreɪnt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a limitation or restriction  
**CN:** 限制，约束

**Original examples:**
- [01:19:38] I think there's a lot of value in first principles thinking and if you know what you're optimizing for and you have strong first principles, then you can normally deduce what the right course of action is and be able to clearly articulate that to all the stakeholders and then you should just do it. I think jobs are fake. If you understand the **constraints**, you can figure out what you can do and then just try to do it quickly, learn from the mistakes and apologize or fix them if you did something wrong.  
  我觉得第一性原理思维很有价值,如果你知道自己在优化什么,并且有坚实的第一性原理,那你通常就能推导出正确的行动方案,并能向所有利益相关方清楚地阐述,然后你就应该去做。我认为工作角色其实是虚的。如果你理解了限制条件,你就能想清楚自己能做什么,然后快速去做,从错误中学习,如果做错了就道歉或修正。
- [01:19:38] I think jobs are fake. If you understand the **constraints**, you can figure out what you can do and then just try to do it quickly, learn from the mistakes and apologize or fix them if you did something wrong.  
  我认为工作是虚假的。如果你理解了限制条件，你就能弄清楚自己能做什么，然后快速去做，从错误中学习，如果做错了就道歉或修正。

**Extra example:**
- Budget **constraints** forced the team to scale back their ambitious plans.  
  预算限制迫使团队缩减了他们雄心勃勃的计划。

### curate  /kjʊˈreɪt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** select, organize, and present items or information, typically using professional or expert knowledge  
**CN:** 策划；精选

**Original examples:**
- [36:23] So what that means for me is I connect it to my Google Calendar, I connect it to my Slack, to my Gmail, to my Google Drive, so that it just knows it has the flexibility to find relevant context, to ask questions, to pull in threads, and this substantially improves the quality of the result by **curating** the right information.  
  所以对我来说，这意味着我把它连接到我的Google Calendar、Slack、Gmail和Google Drive，这样它就知道自己有灵活性去找到相关背景、提出问题、提取线索，通过精选正确的信息，这大大提高了结果的质量。

**Extra example:**
- The museum **curates** exhibitions that showcase contemporary art from emerging artists.  
  博物馆策划展览，展示新兴艺术家的当代艺术作品。

### compelling  /kəmˈpelɪŋ/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** evoking interest, attention, or admiration in a powerfully irresistible way  
**CN:** 令人信服的；引人注目的

**Original examples:**
- [40:24] So for this what I ended up deciding was that I wanted the talk to cover the progression from making local tasks successful to making every PR green to like helping engineers land more PRs and for each of these which demo would be the most **compelling**.  
  所以对于这个，我最终决定的是，我希望演讲涵盖从使本地任务成功到使每个PR通过，再到帮助工程师完成更多PR的进展，以及对于每一个阶段，哪个演示会是最有说服力的。

**Extra example:**
- She presented a **compelling** argument for why the company should invest in renewable energy.  
  她提出了一个令人信服的论点，说明公司为什么应该投资可再生能源。

### progression  /prəˈɡreʃən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the process of developing or moving gradually towards a more advanced state  
**CN:** 进展,发展,演进

**Original examples:**
- [40:24] So for this what I ended up deciding was that I wanted the talk to cover the **progression** from making local tasks successful to making every PR green to like helping engineers land more PRs.  
  所以最终我决定,我希望这次演讲涵盖从成功完成本地任务到让每个 PR 通过,再到帮助工程师提交更多 PR 的**演进过程**。

**Extra example:**
- The **progression** of the disease can be slowed with early treatment.  
  通过早期治疗可以减缓疾病的**进展**。

### compliance  /kəmˈplaɪəns/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** the practice of obeying rules, laws, or standards  
**CN:** 合规,遵守(规则、法律或标准)

**Original examples:**
- [46:19] Having to make do with outdated solutions. Vanta automates **compliance** and risk management with over 35 security and privacy frameworks including SOC 2, ISO 27001 and HIPAA.  
  不得不凑合使用过时的解决方案。Vanta 自动化了合规和风险管理,支持超过 35 个安全和隐私框架,包括 SOC 2、ISO 27001 和 HIPAA。
- [46:19] Vanta automates **compliance** and risk management with over 35 security and privacy frameworks including SOC 2, ISO 27001 and HIPAA.  
  Vanta 通过超过 35 个安全和隐私框架(包括 SOC 2、ISO 27001 和 HIPAA)自动化**合规**和风险管理。

**Extra example:**
- The company must ensure **compliance** with all environmental regulations.  
  公司必须确保**遵守**所有环境法规。

### tailor  /ˈteɪlər/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make or adapt something for a particular purpose or person  
**CN:** 定制,使适应(特定目的或人)

**Original examples:**
- [44:34] Time doing it or they'll just decide not to do it and use the general deck. With this it takes like a few seconds and you get a **tailored** deck.  
  要么就干脆不做,直接用通用版本。有了这个工具,几秒钟就能生成一份定制化的演示文稿。

**Extra example:**
- We **tailor** our training programs to meet each client's specific needs.  
  我们**定制**培训项目以满足每个客户的具体需求。

### infrastructure  /ˈɪnfrəstrʌktʃər/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the basic physical and organizational structures and facilities needed for the operation of a society or enterprise  
**CN:** 基础设施,基础架构

**Original examples:**
- [45:04] I think it's pretty important communications **infrastructure** and I think they do the core task of helping everyone get real-time updates incredibly well.  
  我认为这是非常重要的通信**基础设施**,我认为他们在帮助每个人获得实时更新这一核心任务上做得非常出色。

**Extra example:**
- The country needs to invest heavily in its digital **infrastructure**.  
  这个国家需要在其数字**基础设施**上进行大量投资。

### compliant  /kəmˈplaɪənt/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** meeting or agreeing with rules, standards, or laws  
**CN:** 合规的,符合(规则、标准或法律)的

**Original examples:**
- [46:31] This helps companies get **compliant** fast and stay **compliant** more than ever before.  
  这帮助公司快速实现**合规**并比以往任何时候都更好地保持**合规**状态。

**Extra example:**
- All our products are fully **compliant** with international safety standards.  
  我们所有的产品都完全**符合**国际安全标准。

### prototype  /ˈproʊtətaɪp/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** an early sample or model built to test a concept or process  
**CN:** 原型,雏形

**Original examples:**
- [47:14] And so sometimes our applied team will for example make **prototypes** on behalf of these customers which Claude Code makes so much faster than it used to be.  
  所以有时我们的应用团队会代表这些客户制作**原型**,而 Claude Code 让这个过程比以前快得多。

**Extra example:**
- The engineers built a working **prototype** to demonstrate the new technology.  
  工程师们制作了一个可运行的**原型**来展示这项新技术。

### inbound  /ˈɪnbaʊnd/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** traveling towards or arriving at a place, especially incoming communications or requests  
**CN:** 入站的,进来的(尤指通信或请求)

**Original examples:**
- [47:26] They also have the dual goal of needing to manage a lot of customer comms, a lot of like customer **inbound** and historical context call notes.  
  他们还有双重目标,需要管理大量客户沟通,大量客户**来电**以及历史背景通话记录。

**Extra example:**
- Our team handles over 500 **inbound** customer inquiries every day.  
  我们的团队每天处理超过 500 个**来电**客户咨询。

### dossier  /ˈdɒsieɪ/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a collection of documents containing detailed information about a particular person or subject  
**CN:** 档案,卷宗(关于某人或某事的详细资料汇编)

**Original examples:**
- [48:45] And Cowork will just put together this **dossier**, this brief of what they should be aware of going into the next meeting.  
  Cowork 会整理出这份**档案**,这份简报会列出他们在进入下次会议前应该了解的内容。

**Extra example:**
- The intelligence agency compiled a comprehensive **dossier** on the suspect.  
  情报机构编制了一份关于嫌疑人的全面**档案**。

### imbue  /ɪmˈbjuː/
**CEFR:** C2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to fill or inspire with a particular quality or feeling  
**CN:** 使充满(某种特质或感觉),灌输

**Original examples:**
- [01:00:40] And so we try to **imbue** this into Claude because we think it makes it a lot more enjoyable to work with.  
  所以我们试图将这种特质**灌输**到Claude中,因为我们认为这会让它用起来更愉快。

**Extra example:**
- The teacher tried to **imbue** her students with a love of learning.  
  这位老师试图向学生**灌输**对学习的热爱。

### refactor  /ˌriːˈfæktər/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to restructure existing code without changing its external behavior, improving its internal structure  
**CN:** 重构(代码),在不改变外部行为的情况下改进代码内部结构

**Original examples:**
- [01:01:20] When we first launched Claude Code, people would ask it to do these large **refactors** and Claude Code would say, "Okay, cool. I need to change these like 20 call sites," and it would go and change five of them and then stop.  
  当我们首次推出Claude Code时,人们会要求它做这些大规模的**重构**,Claude Code会说,'好的,我需要修改这20个调用点',然后它会修改其中5个就停下来。

**Extra example:**
- We need to **refactor** this module to make it more maintainable.  
  我们需要**重构**这个模块,让它更易于维护。

### traverse  /trəˈvɜːrs/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to move across, through, or over something systematically  
**CN:** 穿越,遍历,系统地移动通过某物

**Original examples:**
- [01:04:21] Code review agents simultaneously **traverse** the entirety of the codebase and synthesize a set of real issues that an engineer needs to address before merge.  
  代码审查agent会同时**遍历**整个代码库,并综合出工程师在合并前需要解决的一系列实际问题。

**Extra example:**
- The algorithm **traverses** the tree structure to find the optimal path.  
  该算法**遍历**树形结构以找到最优路径。

### extrapolate  /ɪkˈstræpəleɪt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to estimate or conclude something by extending known information or trends  
**CN:** 推断,外推,根据已知信息或趋势推测

**Original examples:**
- [01:06:17] As the models get even smarter, the way that we are **extrapolating** this is okay, next maybe you're going to run like 50 Claudes at a time or hundreds of Claudes at a time.  
  随着模型变得更智能,我们**推断**的方式是,接下来你可能会同时运行50个Claude或数百个Claude。

**Extra example:**
- We can **extrapolate** from current sales data to predict next quarter's revenue.  
  我们可以从当前销售数据**推断**出下季度的收入。

### grunt work  /ɡrʌnt wɜːrk/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** routine, tedious, and often menial work that requires little skill  
**CN:** 繁重乏味的工作,基础性重复工作

**Original examples:**
- [01:08:55] If AI can take care of the **grunt work**, then you have this extra 20% time now that you might not have had before.  
  如果AI能处理这些**繁重乏味的工作**,那么你就有了以前可能没有的额外20%的时间。

**Extra example:**
- Interns often start by doing **grunt work** like data entry and filing.  
  实习生通常从做数据录入和归档这类**基础性工作**开始。

### elbow grease  /ˈelboʊ ɡriːs/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** hard physical effort and energy put into a task  
**CN:** 苦干,费力气,卖力工作

**Original examples:**
- [01:10:18] I would encourage listeners to put in that time to scope some automation that you really want to get to 100%. Put in the **elbow grease** to teach it your preferences, to like give it feedback so that it can improve its skill so that it can get to that 100%.  
  我鼓励听众花时间去规划一些你真正想做到100%的自动化。**下功夫**教它你的偏好,给它反馈,这样它才能提升技能,达到那100%。

**Extra example:**
- This old car needs some **elbow grease** to get it running smoothly again.  
  这辆旧车需要**下点功夫**才能让它重新顺畅运转。

### tinker  /ˈtɪŋkər/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make small adjustments or experiments with something in a casual or experimental way  
**CN:** 修修补补,摆弄,做小调整或实验

**Original examples:**
- [01:12:08] I see a lot of people playing around with AI and building prototype apps and **tinkering** with building workflows.  
  我看到很多人在玩AI,构建原型应用,**摆弄**工作流的构建。

**Extra example:**
- He likes to **tinker** with old electronics in his spare time.  
  他喜欢在业余时间**摆弄**旧电子设备。

### hackable  /ˈhækəbl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** designed to be easily modified, customized, or extended by users  
**CN:** 可定制的,可修改的,易于用户扩展的

**Original examples:**
- [01:13:17] I think there's a lot of fun in customizing and we definitely want to make our products very **hackable** so that you can make it work really well for you, but there is a limit to how much it's useful.  
  我认为定制化很有趣,我们确实希望让我们的产品非常**可定制**,这样你就能让它真正为你所用,但这种做法的有用程度是有限的。

**Extra example:**
- The platform is intentionally **hackable** to encourage developer innovation.  
  这个平台被有意设计成**可定制的**,以鼓励开发者创新。

### aha moment  /ɑːˈhɑː ˈmoʊmənt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a moment of sudden realization, insight, or understanding  
**CN:** 顿悟时刻，恍然大悟的瞬间

**Original examples:**
- [01:14:50] The big **aha moment** people have is when Claude can just do things on your behalf.  
  人们的重大**顿悟时刻**是当 Claude 能够代表你直接做事的时候。

**Extra example:**
- I had an **aha moment** when I finally understood how the algorithm worked.  
  当我终于理解这个算法如何运作时，我有了一个**顿悟时刻**。

### eye-opening  /ˈaɪ ˌoʊpənɪŋ/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** surprising and revealing; providing new knowledge or insight  
**CN:** 令人大开眼界的，启发性的

**Original examples:**
- [01:15:10] And when people feel that, I think that's the **eye-opening** moment.  
  当人们感受到这一点时，我认为那就是**大开眼界的**时刻。

**Extra example:**
- The documentary about climate change was truly **eye-opening**.  
  这部关于气候变化的纪录片真的**令人大开眼界**。

### pursuit  /pərˈsuːt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the act of trying to achieve a goal or complete a task with dedication  
**CN:** 追求，追寻

**Original examples:**
- [01:16:41] There's just something very satisfying about people being so obsessed with like a singular engineering goal and just like the purity of their **pursuit**.  
  人们如此痴迷于一个单一的工程目标，以及他们**追求**的纯粹性，这其中有一些非常令人满足的东西。

**Extra example:**
- She dedicated her life to the **pursuit** of scientific knowledge.  
  她将一生奉献给了对科学知识的**追求**。

### premium  /ˈpriːmiəm/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** an additional amount of money paid above the standard price; extra cost for superior quality or service  
**CN:** 溢价，额外费用

**Original examples:**
- [01:18:59] Yeah. I always thought Waymo needed to be priced lower than Uber and Lyft to succeed, but actually I'm like very happy to pay a 2x **premium** for it.  
  是的。我一直以为 Waymo 需要比 Uber 和 Lyft 便宜才能成功,但实际上我很愿意为它支付两倍的价格。

**Extra example:**
- Customers are willing to pay a **premium** for organic products.  
  顾客愿意为有机产品支付**溢价**。

### vernacular  /vərˈnækjələr/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the language or terminology used by people in a particular group, profession, or context  
**CN:** 行话，vernacular，特定群体使用的语言

**Original examples:**
- [01:19:17] And I think it's also changed the **vernacular**.  
  我认为这也改变了**行话**。

**Extra example:**
- The tech industry has its own **vernacular** that outsiders find hard to understand.  
  科技行业有自己的**行话**，外行人很难理解。

### motto  /ˈmɑːtoʊ/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a short sentence or phrase expressing a guiding principle or belief  
**CN:** 座右铭，格言

**Original examples:**
- [01:19:30] Do you have a favorite life **motto** that you often come back to in work or in life?  
  你有没有一个最喜欢的人生**座右铭**，在工作或生活中经常回想起来？

**Extra example:**
- Our company **motto** is 'Innovation through collaboration.'  
  我们公司的**座右铭**是'通过协作实现创新'。

### liberating  /ˈlɪbəreɪtɪŋ/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** giving freedom from restrictions or limitations; empowering  
**CN:** 解放的，使人自由的

**Original examples:**
- [01:20:10] I think it's **liberating** actually to tell people this.  
  我认为告诉人们这一点实际上是**令人解放的**。

**Extra example:**
- She found it **liberating** to finally quit her stressful job.  
  她发现终于辞掉压力大的工作是**令人解放的**。

### rigidly  /ˈrɪdʒɪdli/
**CEFR:** C1 | **Part of speech:** adv. | **Occurrences:** 1

**EN:** in a strict, inflexible manner; without allowing change or adaptation  
**CN:** 严格地，僵化地

**Original examples:**
- [01:20:10] I think in a lot of companies, roles are very strictly defined like, okay, this is what the PM does, this is what the designer does, this is what the engineer does, and then even team scopes are very **rigidly** defined.  
  我认为在很多公司，角色定义非常严格，比如，这是 PM 做的，这是设计师做的，这是工程师做的，甚至团队范围也被**僵化地**定义。

**Extra example:**
- The rules were **rigidly** enforced with no exceptions allowed.  
  规则被**严格地**执行，不允许有任何例外。

### ambitious  /æmˈbɪʃəs/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** having a strong desire to achieve something difficult or requiring great effort  
**CN:** 雄心勃勃的，有野心的；艰巨的，有挑战性的

**Original examples:**
- [01:21:18] Just like you have all the tools at your disposal, you have some like **ambitious** hairy problem statement and you can do whatever you need to like get to a good solution.  
  就像你拥有所有可用的工具一样,你有一些**雄心勃勃的**棘手问题陈述,你可以做任何需要做的事情来获得一个好的解决方案。

**Extra example:**
- The company set an **ambitious** goal to double its revenue within two years.  
  公司设定了一个**雄心勃勃的**目标,要在两年内将收入翻倍。

### disposal  /dɪˈspoʊzl/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the act of getting rid of something; availability for use (especially in 'at one's disposal')  
**CN:** 处理，清除；可供使用（尤其用于短语'at one's disposal'表示可自由支配）

**Original examples:**
- [01:21:18] Just like you have all the tools at your **disposal**, you have some like ambitious hairy problem statement and you can do whatever you need to like get to a good solution.  
  就像你拥有所有**可用的**工具一样,你有一些雄心勃勃的棘手问题陈述,你可以做任何需要做的事情来获得一个好的解决方案。

**Extra example:**
- The team has extensive resources at their **disposal** to complete the project.  
  团队有大量**可支配的**资源来完成这个项目。

### thrive  /θraɪv/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to grow, develop, or be successful, especially in a vigorous way  
**CN:** 茁壮成长，兴旺发达，蓬勃发展

**Original examples:**
- [01:23:49] But the things that we **thrive** on is edge cases, errors, like specific tasks that we can reproduce where Claude Code or Cowork fail.  
  但我们真正**受益于**的是边缘案例、错误,比如我们可以重现的Claude Code或Cowork失败的具体任务。

**Extra example:**
- Children **thrive** in environments where they feel safe and supported.  
  孩子们在感到安全和被支持的环境中会**茁壮成长**。

---

## Useful Phrases

### AGI pilled
**Type:** idiom

**EN:** strongly convinced or obsessed with the idea of Artificial General Intelligence  
**CN:** 对通用人工智能深信不疑或过度痴迷

**Literal:** 服用了AGI药丸  
**Figurative EN:** being overly focused on or convinced by the potential of AGI, often to the point of losing practical perspective  
**Figurative CN:** 过度关注或相信AGI的潜力，以至于失去实际视角

**Original examples:**
- [00:00] I think it is very hard to be the right amount of **AGI pilled**.  
  我认为很难对AGI保持恰当的痴迷程度。
- [02:19] This is what the **AGI pill** version of the product is.  
  这就是产品的AGI痴迷版本。

**Extra example:**
- He's so **AGI pilled** that he thinks all jobs will disappear next year.  
  他对AGI太过痴迷，以至于认为明年所有工作都会消失。

### elicit
**Type:** collocation

**EN:** to draw out or bring forth (a response, capability, or information)  
**CN:** 引出，激发（能力、反应或信息）

**Original examples:**
- [00:07] The hard thing is figuring out for the current model, how do you **elicit** the maximum capability?  
  难点在于弄清楚对于当前模型，如何激发出最大能力？

**Extra example:**
- Good prompting can **elicit** better responses from the AI model.  
  好的提示词可以从AI模型中激发出更好的响应。

### rowing in the same direction
**Type:** idiom

**EN:** working together toward the same goal  
**CN:** 朝着同一个目标努力，齐心协力

**Literal:** 朝同一个方向划船  
**Figurative EN:** all team members working in alignment toward a common objective  
**Figurative CN:** 所有团队成员协调一致地朝着共同目标努力

**Original examples:**
- [02:37] And I spend more of my time on the cross functional, so making sure that our marketing team, sales team, finance, capacity, etc. are like bought in on the plan and that we're all **rowing in the same direction**.  
  我花更多时间在跨职能协作上，确保我们的市场团队、销售团队、财务、产能等都认同这个计划，并且我们都在朝着同一个方向努力。

**Extra example:**
- For the project to succeed, we need everyone **rowing in the same direction**.  
  为了项目成功，我们需要每个人都齐心协力。

### mind meld
**Type:** idiom

**EN:** to be in complete agreement or share the same thoughts  
**CN:** 思想完全一致，心灵相通

**Literal:** 思维融合  
**Figurative EN:** to think alike or be in perfect sync with someone's thinking  
**Figurative CN:** 与某人想法一致或完全同步

**Original examples:**
- [02:58] I think in many ways it works well because we kind of like **mind meld**, but it is actually like remarkably blurry of a line.  
  我认为在很多方面运作良好，因为我们有点像心灵相通，但实际上界限非常模糊。
- [03:20] Like I think we're like 80% **mind melded**.  
  我觉得我们大概有80%的思想是一致的。

**Extra example:**
- After working together for years, they **mind meld** on most decisions.  
  合作多年后，他们在大多数决策上都心有灵犀。

### out of the box
**Type:** idiom

**EN:** immediately available without modification; working from the start  
**CN:** 开箱即用，无需修改即可使用

**Literal:** 从盒子里拿出来  
**Figurative EN:** functioning immediately without additional setup or customization  
**Figurative CN:** 无需额外设置或定制即可立即使用

**Original examples:**
- [06:19] And help define what are the most important tasks that need to work **out of the box** for my product.  
  并帮助定义哪些是我的产品需要开箱即用的最重要任务。

**Extra example:**
- The software should work **out of the box** without complex configuration.  
  这个软件应该开箱即用，无需复杂配置。

### dogfood
**Type:** idiom

**EN:** to use one's own product internally before releasing it to customers  
**CN:** 内部试用自己的产品（在发布给客户之前）

**Literal:** 吃狗粮  
**Figurative EN:** to test and use your own product internally to ensure quality before public release  
**Figurative CN:** 在公开发布前内部测试和使用自己的产品以确保质量

**Original examples:**
- [08:31] So when engineers have a feature that they feel is ready and that we've **dogfooded** internally, they post it in our Evergreen launch room.  
  所以当工程师们觉得某个功能已经准备好，并且我们已经在内部试用过后，他们就会把它发布到我们的常青发布频道。

**Extra example:**
- We always **dogfood** new features for at least a week before launch.  
  我们总是在发布前至少内部试用新功能一周。

### jump in
**Type:** phrasal_verb

**EN:** to start participating or contributing immediately  
**CN:** 立即参与，马上投入

**Original examples:**
- [08:40] And then Sarah, who leads our docs, and Alex, who leads PMM, and Tar and Lydia on DevRel, just **jump in** and can turn around the marketing announcement for it the very next day.  
  然后负责文档的Sarah、负责产品营销的Alex，以及开发者关系团队的Tar和Lydia就会立即投入，第二天就能完成营销公告。

**Extra example:**
- Feel free to **jump in** with your ideas during the meeting.  
  会议期间请随时提出你的想法。

### turn around
**Type:** phrasal_verb

**EN:** to complete or produce something, especially quickly  
**CN:** 完成，制作（尤指快速完成）

**Original examples:**
- [08:40] And then Sarah, who leads our docs, and Alex, who leads PMM, and Tar and Lydia on DevRel, just jump in and can **turn around** the marketing announcement for it the very next day.  
  然后负责文档的Sarah、负责产品营销的Alex，以及开发者关系团队的Tar和Lydia就会立即投入，第二天就能完成营销公告。

**Extra example:**
- We need to **turn around** this report by Friday.  
  我们需要在周五前完成这份报告。

### make the hard decision
**Type:** collocation

**EN:** to choose a difficult or unpopular option after careful consideration  
**CN:** 做出艰难的决定

**Original examples:**
- [13:49] But yeah, we did have to **make the hard decision** that we needed to prioritize our first party products and our API.  
  但是，我们确实不得不做出艰难的决定，我们需要优先考虑我们的第一方产品和API。

**Extra example:**
- The CEO had to **make the hard decision** to lay off 20% of the workforce.  
  首席执行官不得不做出艰难的决定，裁员20%。

### on your behalf
**Type:** collocation

**EN:** acting as a representative for someone; in someone's interest  
**CN:** 代表某人；为某人

**Original examples:**
- [14:49] There's the Cloud developer platform team that maintains the APIs that Claude is built on top of, and they also release things like managed agents, which is a way for you to build your agents and we can host it **on your behalf**.  
  还有云开发者平台团队，他们维护Claude所基于的API，他们还发布托管代理之类的东西，这是一种让你构建代理的方式，我们可以代表你托管它。

**Extra example:**
- Our lawyer will negotiate the contract **on your behalf**.  
  我们的律师将代表你协商合同。

### stay on top of
**Type:** phrasal_verb

**EN:** to remain informed about or in control of something  
**CN:** 掌握最新情况；控制住

**Original examples:**
- [15:59] Designers are squeezed. There's less time to **stay on top of** everything that is happening.  
  设计师们被挤压了。没有足够的时间来掌握正在发生的一切。
- [22:30] How do you just kind of deal as a human going through so much constant change, just like just being on—The inside of the tornado? Maybe it's calm there, but just like how do you **stay on top of** what's going on?  
  作为一个人，你如何应对如此多的持续变化，就像身处龙卷风内部？也许那里很平静，但你如何掌握正在发生的事情？

**Extra example:**
- It's hard to **stay on top of** all the new AI developments.  
  很难掌握所有新的人工智能发展动态。

### end-to-end
**Type:** collocation

**EN:** covering the entire process from beginning to completion  
**CN:** 端到端；从头到尾

**Original examples:**
- [16:47] There are many engineers on our team who are fully able to **end-to-end** go from seeing user feedback on Twitter through to shipping.  
  我们团队中有很多工程师完全能够端到端地从在Twitter上看到用户反馈到发布产品。

**Extra example:**
- We need someone who can manage the project **end-to-end**.  
  我们需要一个能够端到端管理项目的人。

### under the sun
**Type:** idiom

**EN:** everything possible; every conceivable thing  
**CN:** 世上的一切；所有可能的事物

**Literal:** 太阳底下  
**Figurative EN:** everything imaginable or possible  
**Figurative CN:** 所有能想到的事物；一切可能的东西

**Original examples:**
- [18:28] We get tens of thousands of GitHub issues asking for every single thing **under the sun**, and it takes a lot of care and taste to figure out, okay, which of these is worth building and what is the right way to build it.  
  我们收到成千上万的GitHub问题，要求实现世上的一切功能，需要大量的关注和品味来弄清楚，好吧，这些中哪些值得构建，以及构建它的正确方式是什么。

**Extra example:**
- They sell everything **under the sun** at that store.  
  那家商店出售世上的一切东西。

### wear a lot of hats
**Type:** idiom

**EN:** to have many different roles or responsibilities  
**CN:** 身兼多职；承担多种角色

**Literal:** 戴很多帽子  
**Figurative EN:** to perform many different roles or have multiple responsibilities  
**Figurative CN:** 扮演多种角色；承担多项职责

**Original examples:**
- [20:51] So I think the current environment values people who are able to **wear a lot of hats**, are able to swap them, and are like very low ego about what work they do to help the team move faster.  
  所以我认为当前的环境重视那些能够身兼多职、能够切换角色、并且对自己做什么工作来帮助团队更快前进非常低调的人。

**Extra example:**
- In a startup, you have to **wear a lot of hats** - one day you're coding, the next you're doing customer support.  
  在创业公司，你必须身兼多职——今天你在编程，明天你在做客户支持。

### lean into
**Type:** phrasal_verb

**EN:** to embrace or actively engage with something, especially a challenge  
**CN:** 积极应对；拥抱（挑战）

**Original examples:**
- [22:39] I think our team is full of people who **lean into** the chaos.  
  我认为我们的团队充满了积极应对混乱的人。
- [01:09:08] my push is to **lean into** these tools, hand off the work that you're not excited to do  
  我的建议是充分利用这些工具，把你不想做的工作移交出去

**Extra example:**
- Instead of avoiding the difficult conversation, she decided to **lean into** it.  
  她决定积极面对这次艰难的对话，而不是回避它。

### keep someone up at night
**Type:** idiom

**EN:** to cause worry or anxiety that prevents sleep  
**CN:** 让某人担心得睡不着觉

**Literal:** 让某人在夜里保持清醒  
**Figurative EN:** to cause such worry or concern that it affects one's ability to sleep or relax  
**Figurative CN:** 让人担心或焦虑到影响睡眠

**Original examples:**
- [24:16] Launching a feature that is buggy is the kind of thing that would have **kept me up at night**.  
  发布一个有bug的功能是那种会让我彻夜难眠的事情。

**Extra example:**
- The thought of the upcoming presentation **keeps me up at night**.  
  想到即将到来的演讲让我夜不能寐。

### cut across
**Type:** phrasal_verb

**EN:** to affect or involve different groups, areas, or levels  
**CN:** 横跨，涉及（不同群体、领域或层级）

**Original examples:**
- [29:57] We're able to make very fast decisions that **cut across** the entire org and execute on them in a unified way.  
  我们能够做出横跨整个组织的快速决策，并以统一的方式执行。

**Extra example:**
- This policy **cuts across** all departments, from engineering to sales.  
  这项政策涉及所有部门，从工程到销售。

### stand behind
**Type:** phrasal_verb

**EN:** to support or defend someone or something  
**CN:** 支持，拥护

**Original examples:**
- [30:32] And then everyone will **stand behind** the one that we decide.  
  然后每个人都会支持我们决定的那个方案。

**Extra example:**
- The CEO said she would **stand behind** her team's decision no matter what.  
  CEO表示无论如何她都会支持团队的决定。

### double down on
**Type:** phrasal_verb

**EN:** to increase one's commitment or effort toward something  
**CN:** 加倍投入，加大力度

**Original examples:**
- [32:09] One of the ways that we're able to do this is with the Claude subscriptions with our first-party products, and so we just very much want to **double down on** that.  
  我们能够做到这一点的方法之一是通过Claude订阅和我们的第一方产品，所以我们非常想在这方面加大力度。
- [01:11:57] Anything you wanted to **double down on** that we haven't already touched on  
  有什么你想要强调的，我们还没有提到的

**Extra example:**
- After the initial success, the company decided to **double down on** its AI strategy.  
  在初步成功后，公司决定在AI战略上加倍投入。

### kick off
**Type:** phrasal_verb

**EN:** to start or initiate something  
**CN:** 启动，开始

**Original examples:**
- [32:44] So I tend to use Cloud Code in the terminal when I'm just **kicking off** like a one-off coding task.  
  所以当我只是启动一个一次性的编码任务时，我倾向于在终端使用Cloud Code。
- [34:14] I think the benefit of web and mobile is that it's really great for **kicking** things **off** on the go.  
  我认为网页和移动端的好处是非常适合在移动中启动任务。
- [34:57] What mobile lets you do is **kick off** these tasks on the go.  
  移动端让你能够在移动中启动这些任务。

**Extra example:**
- Let's **kick off** the meeting with a quick round of introductions.  
  让我们以快速的自我介绍来开始会议。

### at a glance
**Type:** collocation

**EN:** with a quick look; immediately upon looking  
**CN:** 一眼就能看到，一目了然

**Original examples:**
- [33:57] Desktop is also great for getting an **at-a-glance** view of everything that's happening.  
  桌面版也非常适合一目了然地查看正在发生的所有事情。

**Extra example:**
- The dashboard gives you an **at-a-glance** overview of all your metrics.  
  仪表板让你一目了然地看到所有指标的概览。

### on the go
**Type:** collocation

**EN:** while traveling or moving from place to place; while busy  
**CN:** 在移动中，在忙碌中

**Original examples:**
- [34:14] I think the benefit of web and mobile is that it's really great for kicking things off **on the go**.  
  我认为网页和移动端的好处是非常适合在移动中启动任务。
- [34:39] You're out and about, you're like touching grass, you're going on a walk and you don't have your laptop open.  
  你在外面，比如去户外走走，去散步，而你没有打开笔记本电脑。
- [34:57] What mobile lets you do is kick off these tasks **on the go**.  
  移动端让你能够在移动中启动这些任务。

**Extra example:**
- I usually listen to podcasts **on the go** during my commute.  
  我通常在通勤途中听播客。

### sleep on
**Type:** idiom

**EN:** to fail to appreciate or recognize the value or importance of something  
**CN:** 低估，忽视（某事物的价值或重要性）

**Literal:** 在某物上睡觉  
**Figurative EN:** to underestimate or overlook the significance of something  
**Figurative CN:** 低估或忽视某事物的重要性

**Original examples:**
- [35:48] People are just like **sleeping on** the success that Cowork—it's just like growing incredibly fast.  
  人们只是低估了Cowork的成功——它增长得非常快。

**Extra example:**
- Don't **sleep on** this opportunity - it could change your career.  
  别低估这个机会——它可能会改变你的职业生涯。

### go off
**Type:** phrasal_verb

**EN:** to start working independently or autonomously (in this context)  
**CN:** （在此语境中）开始独立工作，自主运行

**Original examples:**
- [40:24] So for this what I ended up deciding was that I wanted the talk to cover the progression from making local tasks successful to making every PR green to like helping engineers land more PRs and for each of these which demo would be the most compelling and then after this decision about the outline Codeium just like **went off** for a few hours and built the whole slide deck.  
  所以对于这个，我最终决定我想让演讲涵盖从使本地任务成功到使每个PR通过，再到帮助工程师完成更多PR的进展，以及对于每一个哪个演示最有说服力，然后在关于大纲的决定之后，Codeium就自己运行了几个小时，构建了整个幻灯片。

**Extra example:**
- Once I gave the agent clear instructions, it just **went off** and completed the entire task.  
  一旦我给代理明确的指示，它就自己运行并完成了整个任务。

### make or break
**Type:** idiom

**EN:** to cause either total success or total failure  
**CN:** 决定成败

**Literal:** 制造或打破  
**Figurative EN:** to be the factor that determines complete success or complete failure  
**Figurative CN:** 成为决定彻底成功或彻底失败的关键因素

**Original examples:**
- [46:31] Trust has the power to **make or break** your business.  
  信任有能力决定你业务的成败。

**Extra example:**
- This presentation will **make or break** our chances of getting funding.  
  这次演示将决定我们能否获得资金。

### push the boundaries
**Type:** collocation

**EN:** to extend the limits of what is possible or acceptable  
**CN:** 突破界限，拓展极限

**Original examples:**
- [47:04] Oh, applied AI is amazing at **pushing the boundaries** of what Claude Code and Copilot can do.  
  哦，应用AI团队在突破Claude Code和Copilot的能力边界方面做得非常出色。
- [51:52] I think there are patterns that the best PMs can see based on how users are **pushing the limits** of the existing product.  
  我认为最优秀的产品经理能够根据用户如何突破现有产品的极限来发现一些模式。

**Extra example:**
- The startup is **pushing the boundaries** of AI-generated content.  
  这家初创公司正在突破AI生成内容的界限。

### top of mind
**Type:** idiom

**EN:** the most important thing someone is thinking about  
**CN:** 最关心的事，首要考虑的事

**Literal:** 在思维的顶端  
**Figurative EN:** the primary concern or most prominent thought in someone's awareness  
**Figurative CN:** 某人意识中最主要的关注点或最突出的想法

**Original examples:**
- [48:45] What's **top of mind** for them?  
  他们最关心什么？

**Extra example:**
- Security is **top of mind** for all our enterprise customers right now.  
  安全问题是我们所有企业客户目前最关心的事。

### kick in
**Type:** phrasal_verb

**EN:** to start to take effect or become active  
**CN:** 开始生效，开始起作用

**Original examples:**
- [50:42] It's such an interesting flywheel that starts to **kick in**.  
  这是一个非常有趣的飞轮效应开始发挥作用。

**Extra example:**
- The new policy will **kick in** next quarter.  
  新政策将在下个季度开始生效。

### AGI-pilled
**Type:** idiom

**EN:** strongly convinced or influenced by the idea of artificial general intelligence  
**CN:** 深信AGI理念的

**Literal:** 吃了AGI药丸  
**Figurative EN:** having adopted a strong belief in or perspective shaped by AGI concepts (from 'red-pilled')  
**Figurative CN:** 采纳了对AGI的强烈信念或被AGI概念塑造的视角（源自'红色药丸'）

**Original examples:**
- [52:14] I think it is very hard to be the right amount of **AGI-pilled**.  
  我认为很难把握对AGI理念的信任程度。

**Extra example:**
- He's so **AGI-pilled** that he thinks all jobs will disappear in five years.  
  他对AGI理念深信不疑，认为所有工作都会在五年内消失。

### golden path
**Type:** collocation

**EN:** the optimal or recommended way to accomplish something  
**CN:** 最佳路径，黄金路径

**Original examples:**
- [53:07] How do you help users get onto the **golden path**?  
  你如何帮助用户走上最佳路径？

**Extra example:**
- Our documentation shows the **golden path** for setting up authentication.  
  我们的文档展示了设置身份验证的最佳路径。

### jump into
**Type:** phrasal_verb

**EN:** to start doing something, often quickly or enthusiastically  
**CN:** 投入到，开始做

**Original examples:**
- [56:15] I personally **jump into** evals when there's a feature that I think needs a bit more product definition.  
  当有功能需要更多产品定义时，我个人会投入到评估工作中。

**Extra example:**
- Let's **jump into** the code review right away.  
  让我们马上开始代码审查吧。

### eat your harness for breakfast
**Type:** idiom

**EN:** to easily overcome or make obsolete the systems built around it  
**CN:** 轻易超越或淘汰围绕它构建的系统

**Literal:** 把你的马具当早餐吃掉  
**Figurative EN:** to render unnecessary the scaffolding or support structures through superior capability  
**Figurative CN:** 通过卓越的能力使支撑结构或脚手架变得不必要

**Original examples:**
- [01:02:44] I forget who said this on the podcast that the model will **eat your harness for breakfast**.  
  我忘了是谁在播客上说过，模型会轻易超越你构建的框架。

**Extra example:**
- The new AI model will **eat your harness for breakfast** - all those workarounds are now obsolete.  
  新的AI模型会轻易超越你的框架——所有那些变通方法现在都过时了。

### swap in
**Type:** phrasal_verb

**EN:** to replace something with something else  
**CN:** 替换，换入

**Original examples:**
- [01:05:12] with the newest model you can just **swap it in** to the prototype you've already made and see, okay, does this new model close that gap.  
  有了最新的模型，你可以直接把它换入你已经做好的原型中，看看这个新模型是否能弥补那个差距。

**Extra example:**
- We can **swap in** a faster processor to improve performance.  
  我们可以换入一个更快的处理器来提升性能。

### close the gap
**Type:** collocation

**EN:** to reduce or eliminate a difference or deficiency  
**CN:** 弥补差距，缩小差距

**Original examples:**
- [01:05:12] with the newest model you can just swap it in to the prototype you've already made and see, okay, does this new model **close that gap**.  
  有了最新的模型，你可以直接把它换入你已经做好的原型中，看看这个新模型是否能弥补那个差距。

**Extra example:**
- The new training program should help **close the gap** between junior and senior developers.  
  新的培训计划应该有助于缩小初级和高级开发人员之间的差距。

### hand off
**Type:** phrasal_verb

**EN:** to give responsibility or a task to someone else  
**CN:** 移交，转交

**Original examples:**
- [01:09:08] my push is to lean into these tools, **hand off** the work that you're not excited to do  
  我的建议是充分利用这些工具，把你不想做的工作移交出去

**Extra example:**
- I'll **hand off** this project to Sarah when I go on vacation.  
  我休假时会把这个项目移交给莎拉。

### elbow grease
**Type:** idiom

**EN:** hard physical work and effort  
**CN:** 苦干，费力气的工作

**Literal:** 肘部的油脂  
**Figurative EN:** hard work, physical effort, and determination put into a task  
**Figurative CN:** 投入到任务中的辛勤工作、体力付出和决心

**Original examples:**
- [01:10:18] Put in the **elbow grease** to teach it your preferences, to like give it feedback so that it can improve its skill  
  投入精力去教它你的偏好，给它反馈，这样它才能提升技能

**Extra example:**
- This old car needs some **elbow grease** to get it running again.  
  这辆旧车需要下点苦功夫才能重新运转起来。

### one-shot
**Type:** collocation

**EN:** to do something in a single attempt or instance  
**CN:** 一次性完成

**Original examples:**
- [01:12:38] when it's like, okay, I just **one-shotted** something. Oh, that's cool. And then you never come back to it.  
  就像是，好吧，我一次性做了个东西。哦，挺酷的。然后你就再也不回来看它了。

**Extra example:**
- He **one-shotted** the entire tutorial without any errors.  
  他一次性完成了整个教程，没有任何错误。

### eye-opening
**Type:** collocation

**EN:** surprising and revealing, providing new understanding or awareness  
**CN:** 令人大开眼界的，启发性的

**Original examples:**
- [01:15:10] And when people feel that, I think that's the **eye-opening** moment.  
  当人们感受到这一点时，我认为那就是令人大开眼界的时刻。

**Extra example:**
- The documentary was **eye-opening** - I had no idea about those issues.  
  这部纪录片令人大开眼界——我之前完全不知道那些问题。

### get used to
**Type:** phrasal_verb

**EN:** to become familiar with something through experience  
**CN:** 习惯于，适应

**Original examples:**
- [01:19:06] And then you **get used to** it.  
  然后你就习惯了。

**Extra example:**
- It took me a while to **get used to** working from home.  
  我花了一段时间才习惯在家工作。

### come back to
**Type:** phrasal_verb

**EN:** to return to a topic, idea, or place repeatedly  
**CN:** 回到（某话题或想法），反复想起

**Original examples:**
- [01:19:30] Do you have a favorite life motto that you often **come back to** in work or in life?  
  你有没有一个在工作或生活中经常回想起的人生格言？

**Extra example:**
- I keep **coming back to** the same question: what's the real problem here?  
  我一直回到同一个问题：真正的问题是什么？

### first principles thinking
**Type:** collocation

**EN:** reasoning from fundamental truths rather than by analogy  
**CN:** 第一性原理思维，从基本原理出发进行推理

**Original examples:**
- [01:19:38] I think there's a lot of value in **first principles thinking**.  
  我认为第一性原理思维非常有价值。

**Extra example:**
- Using **first principles thinking**, we broke down the problem to its core components.  
  运用第一性原理思维，我们把问题分解到核心要素。

### figure out
**Type:** phrasal_verb

**EN:** to understand or solve something through thinking  
**CN:** 弄清楚，想出办法

**Original examples:**
- [01:19:38] You can normally deduce what the right course of action is and be able to clearly articulate that to all the stakeholders and then you should just do it.  
  你通常可以推断出正确的行动方案，并能够向所有利益相关者清楚地表达，然后你就应该去做。
- [01:20:08] If you understand the constraints, you can **figure out** what you can do.  
  如果你理解了限制条件，你就能弄清楚自己能做什么。
- [01:21:04] And it was like I really appreciate Alex and the rest of the team for like empowering me and the rest of the team to just like **figure things out**.  
  我真的很感激Alex和团队其他成员赋予我和团队自主解决问题的权力。

**Extra example:**
- Give me some time to **figure out** the best approach.  
  给我点时间想出最好的方法。

### bias towards action
**Type:** collocation

**EN:** a preference for taking action rather than waiting or deliberating  
**CN:** 行动导向，倾向于采取行动

**Original examples:**
- [01:20:46] **Bias towards action**. All these ways of describing just like, don't wait for permission.  
  行动导向。所有这些说法都是在描述，不要等待许可。

**Extra example:**
- Our company culture emphasizes a **bias towards action** over endless planning.  
  我们的公司文化强调行动导向，而不是无休止的计划。

### at your disposal
**Type:** idiom

**EN:** available for you to use  
**CN:** 供你使用，任你支配

**Literal:** 在你的处置之下  
**Figurative EN:** available for your use whenever needed  
**Figurative CN:** 随时可供你使用

**Original examples:**
- [01:21:18] Just like you have all the tools **at your disposal**.  
  你拥有所有可用的工具。

**Extra example:**
- We have a full team of experts **at your disposal** for this project.  
  我们有一整个专家团队可供你在这个项目中使用。

### bring the world along
**Type:** idiom

**EN:** to help others adapt to or adopt new changes  
**CN:** 帮助世界跟上（变化），带领大家一起前进

**Literal:** 把世界带在一起  
**Figurative EN:** to help society or others keep pace with and adapt to changes  
**Figurative CN:** 帮助社会或他人跟上并适应变化

**Original examples:**
- [01:22:28] I think the immediate thing is actually just like helping **bring the world along**.  
  我认为当务之急实际上就是帮助世界跟上这个变化。

**Extra example:**
- As leaders in AI, we have a responsibility to **bring the world along** with these technological advances.  
  作为人工智能领域的领导者，我们有责任帮助世界跟上这些技术进步。

### thrive on
**Type:** phrasal_verb

**EN:** to grow or develop well because of something; to enjoy and benefit from  
**CN:** 因...而茁壮成长，从...中获益

**Original examples:**
- [01:23:49] But the things that we **thrive on** is edge cases, errors, like specific tasks that we can reproduce.  
  但我们真正受益的是边缘案例、错误，以及我们可以重现的具体任务。

**Extra example:**
- Some people **thrive on** pressure and perform better under tight deadlines.  
  有些人在压力下茁壮成长，在紧迫的截止日期下表现更好。

---

## Complex Sentences

### [00:27]
**Original:** You're interviewing hundreds of PMs and you just keep feeling like they're approaching it very incorrectly.

**Translation:** 你面试了数百名产品经理,却总是觉得他们的方法非常不对。

**Core structure:**
- You're interviewing PMs and you keep feeling they're approaching it incorrectly.  
  你在面试产品经理,并且总觉得他们方法不对。

**Structure tree:**
```
main clause 1: You're interviewing hundreds of PMs
coordinating conjunction: and
main clause 2: you keep feeling...
object clause: (that) they're approaching it very incorrectly
```

**Grammar points:**
- **并列复合句** - 两个独立主句用 and 连接,第二个主句包含宾语从句
- **keep + -ing** - 表示持续或反复发生的动作
- **省略 that 的宾语从句** - feeling 后省略了连接词 that

### [00:32]
**Original:** The thing that is extremely important for building AI native products is iterating so quickly, figuring out a way for you to actually launch features every single week.

**Translation:** 对于构建AI原生产品来说极其重要的事情是快速迭代,找到一种方法让你能够真正做到每周都发布新功能。

**Core structure:**
- The thing is iterating quickly and figuring out a way to launch features weekly.  
  重要的事情是快速迭代并找到每周发布功能的方法。

**Structure tree:**
```
subject: The thing (+ that clause modifier)
predicate: is
predicative (compound gerunds):
  - iterating so quickly
  - figuring out a way...
    └─ infinitive phrase: to actually launch features every single week
```

**Grammar points:**
- **定语从句修饰主语** - that 从句修饰 the thing
- **并列动名词作表语** - iterating 和 figuring out 并列,都作系动词 is 的表语
- **不定式作定语** - to launch 修饰 way,表示方法的具体内容

### [01:41]
**Original:** I want to start with giving people an understanding of your role alongside Boris, everybody knows Boris.

**Translation:** 我想先让大家了解一下你和Boris一起的角色定位,大家都认识Boris。

**Core structure:**
- I want to start with giving people an understanding of your role.  
  我想先让大家了解你的角色。

**Structure tree:**
```
main clause: I want to start with...
prepositional phrase: with giving people an understanding
  └─ of your role alongside Boris
independent clause (comma splice): everybody knows Boris
```

**Grammar points:**
- **start with + 动名词** - with 后接动名词短语作宾语
- **逗号连接独立句** - 口语中用逗号连接两个独立句子,书面语中应用分号或句号

### [02:37]
**Original:** And a lot of my role is figuring out okay what is the path from where we are today to like that vision 3 to 6 months from now.

**Translation:** 我的很大一部分工作是弄清楚,从我们今天所处的位置到3到6个月后的那个愿景,路径是什么。

**Core structure:**
- My role is figuring out what is the path from today to that vision.  
  我的工作是弄清楚从今天到那个愿景的路径是什么。

**Structure tree:**
```
subject: a lot of my role
predicate: is
predicative: figuring out...
  └─ embedded question: what is the path
      └─ from where we are today
      └─ to that vision 3 to 6 months from now
```

**Grammar points:**
- **动名词短语作表语** - figuring out 作系动词 is 的表语
- **嵌入式疑问句** - what is the path 作 figuring out 的宾语,保持疑问句语序
- **where 引导地点从句** - where we are today 作 from 的宾语

### [03:36]
**Original:** If you're building a product for the enterprise, you've felt the pain of integrating single sign-on, SCIM, RBAC, audit logs, and other features required by large companies.

**Translation:** 如果你在为企业构建产品,你一定感受过集成单点登录、SCIM、RBAC、审计日志以及大公司要求的其他功能所带来的痛苦。

**Core structure:**
- If you're building a product for enterprise, you've felt the pain of integrating features.  
  如果你在为企业构建产品,你感受过集成功能的痛苦。

**Structure tree:**
```
conditional clause: If you're building a product for the enterprise
main clause: you've felt the pain of integrating...
  └─ object: single sign-on, SCIM, RBAC, audit logs, and other features
      └─ past participle modifier: required by large companies
```

**Grammar points:**
- **条件状语从句** - if 引导条件从句,主句用现在完成时表示经验
- **过去分词作后置定语** - required by large companies 修饰 features
- **动名词复合结构** - integrating 后接多个并列宾语

### [05:18]
**Original:** The other partner teams to make sure that they're shipping features that unblock your features because code at that time was very expensive to make.

**Translation:** 其他合作团队要确保他们发布的功能能够解锁你的功能，因为那时候编写代码的成本非常高。

**Core structure:**
- Partner teams make sure that they're shipping features that unblock your features.  
  合作团队确保他们发布能解锁你功能的特性。

**Structure tree:**
```
main clause: teams make sure that...
object clause: that they're shipping features
relative clause: that unblock your features
causal clause: because code was expensive
```

**Grammar points:**
- **嵌套的 that 从句** - 第一个 that 引导宾语从句，第二个 that 引导定语从句修饰 features
- **unblock 作动词** - 技术术语，意为'解除阻碍、使能够进行'

### [05:26]
**Original:** I think now with AI and with how much that has accelerated engineering and with how quickly the model capabilities are improving, the timelines for a lot of our product features have gone down from 6 months to one month and sometimes to one week or even one day.

**Translation:** 我认为现在随着人工智能的发展，以及它在多大程度上加速了工程开发，以及模型能力提升的速度有多快，我们很多产品功能的时间线已经从6个月缩短到1个月，有时甚至缩短到1周或1天。

**Core structure:**
- The timelines have gone down from 6 months to one month.  
  时间线已从6个月缩短到1个月。

**Structure tree:**
```
main clause: the timelines have gone down
prepositional phrases: with AI / with how much... / with how quickly...
noun clauses: how much that has accelerated / how quickly capabilities are improving
```

**Grammar points:**
- **多重 with 介词短语** - 三个并列的 with 短语作状语，说明背景条件
- **how 引导名词性从句** - how much/how quickly 引导的从句作介词 with 的宾语

### [06:05]
**Original:** How can we figure out how to make a concept corner of our product suite where an engineer has an idea or a PM has an idea and by the end of the week we are able to get it into our users' hands.

**Translation:** 我们如何能够找到方法，在我们的产品套件中创建一个概念角落，在那里工程师或产品经理有了想法后，到周末时我们就能把它交到用户手中。

**Core structure:**
- How can we make a concept corner where we can get ideas into users' hands.  
  我们如何创建一个能快速将想法交付用户的概念角落。

**Structure tree:**
```
main question: How can we figure out...
embedded question: how to make a corner
relative clause: where an engineer has an idea
time clause: by the end of the week
result clause: we are able to get it...
```

**Grammar points:**
- **嵌套疑问句** - 外层 how 引导主句疑问，内层 how to 引导宾语从句
- **where 引导定语从句** - 修饰 concept corner，描述这个空间的特点

### [06:19]
**Original:** I think the PMs who do the best on AI native products are the ones who can figure out how can I shorten the time from having this idea to actually getting the product in the hands of users and help define what are the most important tasks that need to work out of the box for my product.

**Translation:** 我认为在AI原生产品上做得最好的产品经理，是那些能够找出如何缩短从有想法到真正将产品交到用户手中的时间，并帮助定义哪些是我的产品需要开箱即用的最重要任务的人。

**Core structure:**
- The PMs who do the best are the ones who can figure out how to shorten the time and define important tasks.  
  做得最好的产品经理是那些能缩短时间并定义重要任务的人。

**Structure tree:**
```
main clause: PMs are the ones
relative clause 1: who do the best (modifies PMs)
relative clause 2: who can figure out... (modifies ones)
embedded questions: how can I shorten / what are the tasks
relative clause 3: that need to work (modifies tasks)
```

**Grammar points:**
- **多层定语从句嵌套** - 三个 who/that 定语从句层层修饰，结构复杂
- **间接疑问句** - how can I 和 what are 作为宾语从句，保留疑问词但用陈述语序
- **out of the box** - 习语，意为'开箱即用、无需配置'

### [09:29]
**Original:** The second thing that we do is we have this list of team principles. And this includes who our key users are, why those are our key users. And the reason that we articulate all of this is so that everybody on the team feels like they understand how our business works.

**Translation:** 我们做的第二件事是我们有一份团队原则清单。这包括我们的核心用户是谁，为什么他们是我们的核心用户。我们阐明所有这些的原因是为了让团队中的每个人都觉得他们理解我们的业务是如何运作的。

**Core structure:**
- The reason we articulate this is so that everybody understands how our business works.  
  我们阐明这些是为了让每个人都理解我们的业务如何运作。

**Structure tree:**
```
main clause: The reason is so that...
relative clause: that we articulate (modifies reason)
purpose clause: so that everybody feels
object clause: like they understand
embedded question: how our business works
```

**Grammar points:**
- **so that 引导目的状语从句** - 表示'以便、为了'，说明阐明原则的目的
- **feel like + 从句** - 表示'感觉好像'，like 后接完整从句
- **间接疑问句作宾语** - how our business works 作 understand 的宾语

### [11:20]
**Original:** So we're very low on process. We want to remove every single barrier to shipping things. We want to make sure every single person on the team feels empowered to take their idea from just an idea to out in the world in less than a week, sometimes.

**Translation:** 所以我们的流程非常少。我们想要移除每一个阻碍发布产品的障碍。我们想要确保团队中的每一个人都感到有能力将他们的想法从仅仅是一个想法变成推向世界,有时在不到一周的时间内完成。

**Core structure:**
- We want to make sure every person feels empowered to take their idea to the world.  
  我们想确保每个人都感到有能力将想法推向世界。

**Structure tree:**
```
main clause: We want to make sure...
object clause: every person feels empowered
infinitive phrase: to take their idea from...to...
time adverbial: in less than a week
```

**Grammar points:**
- **make sure + 宾语从句** - 确保某事发生,从句用一般现在时表将来
- **feel empowered to do** - 感到有能力做某事,empowered 作主语补足语
- **from...to... 结构** - 表示从一个状态到另一个状态的转变

### [12:23]
**Original:** A human working with Claude to write a PR. This was just an update to how we release our packages and it actually went through two layers of human review.

**Translation:** 一个人与 Claude 合作编写 PR。这只是对我们如何发布软件包的一次更新,而且它实际上经过了两层人工审查。

**Core structure:**
- This was an update and it went through two layers of review.  
  这是一次更新,它经过了两层审查。

**Structure tree:**
```
main clause 1: This was an update to...
prepositional phrase: to how we release...
indirect question: how we release our packages
main clause 2: it went through two layers...
```

**Grammar points:**
- **间接疑问句** - how we release 作介词 to 的宾语,用陈述语序
- **go through** - 经历、通过(某个过程或阶段)

### [13:18]
**Original:** So, we've been seeing a lot of demand for Claude and we've been working very hard to both scale our infrastructure and also to make our models more token efficient so that you can get more usage out of it.

**Translation:** 所以,我们看到对 Claude 有很大的需求,我们一直在非常努力地扩展我们的基础设施,同时也让我们的模型在 token 使用上更高效,这样你就可以从中获得更多的使用量。

**Core structure:**
- We've been seeing demand and we've been working to scale infrastructure and make models efficient.  
  我们看到需求,一直在努力扩展基础设施并提高模型效率。

**Structure tree:**
```
main clause 1: we've been seeing demand
main clause 2: we've been working to...
parallel infinitives: to scale...and to make...
purpose clause: so that you can get...
```

**Grammar points:**
- **现在完成进行时** - 强调动作从过去持续到现在并可能继续
- **both...and... 并列结构** - 连接两个不定式短语,表示同时进行的动作
- **so that 目的状语从句** - 表示前面行为的目的或结果

### [14:31]
**Original:** So we have the research PM team who Diane leads, and this team is responsible for understanding all of the feedback from our customers for our models and then feeding that to the research team to act on it, and they also shepherd the model launch.

**Translation:** 所以我们有 Diane 领导的研究产品经理团队,这个团队负责理解来自客户对我们模型的所有反馈,然后将这些反馈提供给研究团队来采取行动,他们还负责引导模型的发布。

**Core structure:**
- We have the research PM team, and this team is responsible for understanding feedback and feeding that to the research team, and they shepherd the launch.  
  我们有研究产品经理团队,这个团队负责理解反馈并提供给研究团队,他们还引导发布。

**Structure tree:**
```
main clause: we have the team
relative clause: who Diane leads
coordinate clause: this team is responsible for...
parallel gerunds: understanding...and feeding...
coordinate clause: they shepherd...
```

**Grammar points:**
- **be responsible for + 动名词** - 对...负责,后接动名词或名词
- **并列动名词结构** - understanding 和 feeding 并列,共同作 for 的宾语

### [15:06]
**Original:** And then there's Claude that works on both Cloud Code and the Cowork core products. There's Enterprise that helps make Cloud Code and Cowork easier to adopt for all of our enterprise customers.

**Translation:** 然后还有 Claude 团队,负责 Cloud Code 和 Cowork 核心产品。还有企业团队,帮助让 Cloud Code 和 Cowork 对我们所有的企业客户来说更容易采用。

**Core structure:**
- There's Claude that works on products. There's Enterprise that helps make products easier to adopt.  
  有 Claude 团队负责产品。有企业团队帮助让产品更容易采用。

**Structure tree:**
```
existential clause 1: There's Claude
relative clause: that works on...
existential clause 2: There's Enterprise
relative clause: that helps make...
object complement: easier to adopt
```

**Grammar points:**
- **help + 动词原形** - help 后可接不带 to 的不定式
- **make + 宾语 + 形容词比较级** - 使某物变得更...,形容词作宾语补足语
- **easier to adopt** - 形容词 + 不定式结构,表示在某方面更容易

### [17:27]
**Original:** I was an engineer for many years. I was then a VC very briefly before joining Anthropic and actually almost all the PMs on our team have either been engineers or ship code here on Claude code and so that's one of the things that I think helps build trust with the team and also just enables us to move a lot faster and then actually our designers also have been front-end engineers before.

**Translation:** 我做了很多年工程师。然后在加入Anthropic之前我短暂地做过风投，实际上我们团队几乎所有的产品经理要么曾是工程师，要么在这里为Claude编写代码，所以这是我认为有助于与团队建立信任的事情之一，也能让我们行动更快，而且实际上我们的设计师之前也都做过前端工程师。

**Core structure:**
- Almost all the PMs have been engineers or ship code, and that helps build trust and enables us to move faster.  
  几乎所有产品经理都曾是工程师或编写代码，这有助于建立信任并让我们行动更快。

**Structure tree:**
```
compound sentence with multiple clauses:
- background: I was an engineer / I was a VC
- main: almost all PMs have been engineers or ship code
- result 1: that helps build trust
- result 2: (that) enables us to move faster
- addition: our designers have been front-end engineers
```

**Grammar points:**
- **连续并列结构** - 多个and连接的平行句子，需要追踪多层逻辑关系
- **either...or 结构** - 表示两种可能性之一
- **指示代词that的远距离指代** - that指代前面整个从句内容

### [18:28]
**Original:** We get tens of thousands of GitHub issues asking for every single thing under the sun, and it takes a lot of care and taste to figure out, okay, which of these is worth building and what is the right way to build it, and I think that that skill is what matters.

**Translation:** 我们收到成千上万个GitHub问题，要求实现世界上的各种功能，而要弄清楚其中哪些值得构建以及正确的构建方式是什么，需要大量的用心和品味，我认为这种技能才是最重要的。

**Core structure:**
- We get thousands of issues, and it takes care to figure out which is worth building, and that skill matters.  
  我们收到成千上万的问题，弄清楚哪些值得构建需要用心，这种技能很重要。

**Structure tree:**
```
three-part compound sentence:
1. We get issues asking for everything
2. it takes care to figure out:
   - which is worth building
   - what is the right way
3. that skill is what matters
```

**Grammar points:**
- **形式主语it** - it作形式主语，真正主语是to figure out...
- **间接疑问句** - which/what引导的疑问句作宾语
- **强调句型what matters** - what从句强调重要性

### [19:13]
**Original:** But if something is harder to build and you know that upfront, you know that, okay, this will just cost a lot more for our team to get this out the door.

**Translation:** 但如果某个东西更难构建，而你事先知道这一点，你就会知道，好吧，这会让我们团队付出更多代价才能把它推出去。

**Core structure:**
- If something is harder and you know that, you know this will cost more.  
  如果某个东西更难而你知道这点，你就知道这会花费更多。

**Structure tree:**
```
conditional sentence:
if-clause: something is harder + you know that
main clause: you know that this will cost more
  - nested clause: this will cost more to get out
```

**Grammar points:**
- **条件句中的双重you know** - 第一个know在条件句中，第二个know在主句中，容易混淆
- **不定式表目的** - to get this out表示目的

### [20:08]
**Original:** I think the most important thing is to be able to have this first principles thinking where you can figure out how the tech landscape is changing, what the team really needs from you, and to jump in and fix that hole, because I think the work is becoming more amorphous, which means that a great PM is able to understand what all the gaps are, to figure out what the highest priority ones are and then to just like figure out okay how do I learn that skill set or what is like the skill set that I have that I can like apply to this challenge.

**Translation:** 我认为最重要的是能够具备这种第一性原理思维，在这种思维下你能弄清楚技术格局如何变化、团队真正需要你做什么，然后跳进去填补那个空缺，因为我认为工作正变得更加模糊，这意味着一个优秀的产品经理能够理解所有的差距是什么，弄清楚哪些是最优先的，然后就是弄清楚我该如何学习那个技能组合，或者我拥有的哪个技能组合可以应用到这个挑战上。

**Core structure:**
- The most important thing is to have first principles thinking and to jump in and fix that hole.  
  最重要的是具备第一性原理思维并跳进去填补空缺。

**Structure tree:**
```
complex sentence with nested clauses:
main: the thing is to have thinking and to jump in
- where clause: you can figure out how/what
- because clause: work is becoming amorphous
  - which clause: a PM is able to understand/figure out
    - what clauses: what gaps are / what priority ones are
    - how/what clauses: how to learn / what to apply
```

**Grammar points:**
- **多层嵌套从句** - 主句包含where从句，后接because从句，其中又包含which从句和多个what/how从句
- **并列不定式结构** - to have...to jump in...to understand...to figure out多个不定式并列
- **口语化插入语like** - 多次使用like作为口语填充词，增加理解难度

### [20:51]
**Original:** So I think the current environment values people who are able to wear a lot of hats, are able to swap them, and are like very low ego about what work they do to help the team move faster.

**Translation:** 所以我认为当前的环境重视那些能够身兼多职、能够灵活切换角色、并且对于为了帮助团队更快行动而做什么工作都非常不在意自我的人。

**Core structure:**
- The environment values people who are able to wear hats, swap them, and are low ego about what work they do.  
  环境重视那些能够身兼多职、灵活切换、并且对工作内容不在意自我的人。

**Structure tree:**
```
main: environment values people
who-clause with three parallel parts:
- who are able to wear hats
- (who) are able to swap them
- (who) are low ego about what work they do
  - purpose: to help team move faster
```

**Grammar points:**
- **定语从句中的并列结构** - 三个are able to/are并列修饰people
- **what引导宾语从句** - what work they do作about的宾语
- **习语wear hats** - 比喻承担多种角色

### [21:59]
**Original:** I think the model doesn't always have a great sense of who all the stakeholders are, how they relate to each other, what their preferences are, what are the right venues to communicate with them to keep them on board.

**Translation:** 我认为模型并不总是能很好地理解所有利益相关者是谁、他们之间的关系、他们的偏好是什么、以及与他们沟通以保持他们支持的正确渠道是什么。

**Core structure:**
- The model doesn't have a sense of who stakeholders are, how they relate, what their preferences are, and what venues are right.  
  模型不理解利益相关者是谁、他们如何关联、他们的偏好以及正确的沟通渠道。

**Structure tree:**
```
main clause: the model doesn't have a sense of...
4 parallel noun clauses as objects of 'of':
- who all the stakeholders are
- how they relate to each other
- what their preferences are
- what are the right venues to communicate
```

**Grammar points:**
- **并列宾语从句** - 四个由 who/how/what 引导的名词性从句并列作 of 的宾语
- **不定式作定语** - to communicate 和 to keep 修饰 venues，表示目的

### [22:54]
**Original:** And so we really look for people who can kind of look at a challenge and be like, that's going to be hard, but I'm excited to tackle it and I'm going to do the best that I possibly can.

**Translation:** 所以我们真的在寻找那些能够面对挑战并说'这会很难，但我很兴奋去应对它，我会尽我所能做到最好'的人。

**Core structure:**
- We look for people who can look at a challenge and be excited to tackle it.  
  我们寻找能够面对挑战并兴奋地应对它的人。

**Structure tree:**
```
main clause: we look for people
relative clause: who can look at a challenge and be like...
quoted speech: that's going to be hard, but I'm excited...
relative clause: that I possibly can (modifying 'the best')
```

**Grammar points:**
- **定语从句嵌套** - who 引导定语从句修饰 people，内部又包含 that 引导的定语从句修饰 the best
- **be like 口语用法** - 表示'说/想'，引出直接引语，常见于非正式口语

### [23:40]
**Original:** But I think you just have to acknowledge that there's only so much that you can do, that you need to sleep well so that you can make good decisions the next day and just like brutally prioritize where you spend your time.

**Translation:** 但我认为你必须承认，你能做的只有这么多，你需要睡好觉以便第二天能做出好的决定，并且要残酷地优先考虑你把时间花在哪里。

**Core structure:**
- You have to acknowledge that you can only do so much, need to sleep well, and prioritize where you spend time.  
  你必须承认你只能做这么多，需要睡好，并优先考虑时间分配。

**Structure tree:**
```
main clause: you have to acknowledge that...
3 parallel clauses as objects of 'acknowledge':
- there's only so much that you can do
- you need to sleep well (with purpose clause: so that...)
- (you) brutally prioritize where you spend your time
```

**Grammar points:**
- **并列宾语从句** - 三个 that 从句并列作 acknowledge 的宾语，第二、三个 that 省略
- **so that 目的状语从句** - 表示'以便'，说明睡好觉的目的

### [25:42]
**Original:** Historically, when code was expensive to write, you would carefully plan out everything in your product suite, how every product relates to each other, what the use case for every single one is, how they integrate, and you would pretty much have one product for each use case.

**Translation:** 从历史上看，当代码编写成本很高时，你会仔细规划产品套件中的所有内容、每个产品如何相互关联、每个产品的用例是什么、它们如何集成，而且你基本上会为每个用例配备一个产品。

**Core structure:**
- When code was expensive, you would plan everything and have one product for each use case.  
  当代码昂贵时，你会规划一切并为每个用例配备一个产品。

**Structure tree:**
```
time clause: when code was expensive to write
main clause 1: you would plan out everything...
3 parallel noun clauses as objects of 'plan out':
- how every product relates
- what the use case is
- how they integrate
main clause 2: you would have one product...
```

**Grammar points:**
- **would 表过去习惯** - would 在此表示过去的习惯性动作，相当于 used to
- **并列宾语从句** - 三个由 how/what 引导的名词性从句并列作 plan out 的宾语

### [27:04]
**Original:** Usually in traditional PM you ship a feature every month or quarter, and so it's really easy for a user to understand, okay, I just need to check in on this once a month and I'll learn some new things, and if I ignore it for six months, it's fine.

**Translation:** 通常在传统的产品管理中，你每月或每季度发布一个功能，所以用户很容易理解，好的，我只需要每月查看一次就能学到一些新东西，如果我忽略它六个月，也没关系。

**Core structure:**
- In traditional PM you ship a feature monthly, so it's easy for a user to understand they need to check in once a month.  
  在传统产品管理中你每月发布功能，所以用户容易理解他们需要每月查看一次。

**Structure tree:**
```
condition: in traditional PM
main clause 1: you ship a feature...
main clause 2: it's easy for a user to understand...
informal quoted thought: okay, I just need to...
2 parallel clauses: I'll learn... and if I ignore..., it's fine
```

**Grammar points:**
- **it's + adj + for sb + to do** - it 作形式主语，真正主语是 to understand 及其后的内容
- **条件状语从句** - if I ignore it for six months 表示假设条件

### [29:29]
**Original:** We hire people who care most about bringing safe AGI to all of humanity, and this is actually something that we reference frequently in our decisions about what our entire product org should focus on shipping.

**Translation:** 我们雇佣最关心为全人类带来安全AGI的人,而这实际上是我们在决定整个产品组织应该专注于发布什么时经常参考的东西。

**Core structure:**
- We hire people, and this is something that we reference in our decisions.  
  我们雇佣人,而这是我们在决策中参考的东西。

**Structure tree:**
```
main clause 1: We hire people who...
relative clause: who care most about bringing...
main clause 2: this is something that...
relative clause: that we reference frequently in...
nested clause: what our entire product org should focus on
```

**Grammar points:**
- **嵌套定语从句** - 两个that/who引导的定语从句层层修饰
- **what引导的宾语从句** - what从句作about的宾语,表示'...的事情'

### [31:10]
**Original:** Mission means that teams are willing to make sacrifices that hurt their own goals and their own KPIs in service of Anthropic's goals and Anthropic's KPIs.

**Translation:** 使命意味着团队愿意做出牺牲,这些牺牲会损害他们自己的目标和KPI,以服务于Anthropic的目标和KPI。

**Core structure:**
- Mission means that teams are willing to make sacrifices.  
  使命意味着团队愿意做出牺牲。

**Structure tree:**
```
main clause: Mission means that...
that-clause: teams are willing to make sacrifices
relative clause: that hurt their own goals...
purpose phrase: in service of Anthropic's goals
```

**Grammar points:**
- **that引导宾语从句** - means后接that从句说明含义
- **定语从句修饰sacrifices** - that hurt...描述牺牲的性质
- **in service of** - 介词短语表示目的'为了服务于'

### [32:09]
**Original:** One of the ways that we're able to do this is with the Claude subscriptions with our first-party products, and so we just very much want to double down on that, but that does come at the expense of third-party products sometimes.

**Translation:** 我们能够做到这一点的方法之一是通过Claude订阅和我们的第一方产品,所以我们非常想加倍投入,但这有时确实是以牺牲第三方产品为代价的。

**Core structure:**
- One way is with Claude subscriptions, and we want to double down, but that comes at the expense of third-party products.  
  一种方法是通过Claude订阅,我们想加倍投入,但这以牺牲第三方产品为代价。

**Structure tree:**
```
main clause 1: One of the ways is...
relative clause: that we're able to do this
main clause 2: we want to double down on that
main clause 3: but that does come at the expense of...
```

**Grammar points:**
- **三个并列主句** - and和but连接三个独立观点
- **come at the expense of** - 固定搭配'以...为代价'

### [34:24]
**Original:** CLI and desktop both require you to be on your local laptop, and this is constraining because sometimes you're out and about, you're like touching grass, you're going on a walk and you don't have your laptop open.

**Translation:** CLI和桌面版都要求你在本地笔记本电脑上,这很受限制,因为有时你在外面,比如在户外活动,在散步,而你没有打开笔记本电脑。

**Core structure:**
- CLI and desktop require you to be on your laptop, and this is constraining because you're out and don't have your laptop.  
  CLI和桌面版要求你在笔记本电脑上,这很受限制,因为你在外面且没有笔记本电脑。

**Structure tree:**
```
main clause 1: CLI and desktop require you to be...
main clause 2: this is constraining
because-clause: because sometimes you're out...
parallel clauses: you're touching grass, you're going on a walk, you don't have...
```

**Grammar points:**
- **because引导原因状语从句** - 解释为什么受限制
- **多个并列分句** - you're...you're...you don't...描述多个同时发生的情况

### [35:13]
**Original:** So whether that's like getting to Slack zero or inbox zero, or whether that's creating a slide deck for some customer meeting that's coming up, or whether that's writing a quick doc on what the goals of a feature are or what the launch plan for a feature is.

**Translation:** 所以无论是清空Slack消息或收件箱,还是为即将到来的客户会议制作幻灯片,或者是写一份关于功能目标或功能发布计划的快速文档。

**Core structure:**
- Whether that's getting to zero, or creating a deck, or writing a doc.  
  无论是清空消息,制作幻灯片,还是写文档。

**Structure tree:**
```
parallel whether-clauses:
- whether that's getting to Slack zero...
- or whether that's creating a slide deck...
- or whether that's writing a doc on...
nested what-clauses: what the goals are / what the launch plan is
```

**Grammar points:**
- **whether...or whether并列结构** - 列举多个可能的情况
- **what引导的宾语从句** - what从句作on的宾语,表示'...的内容'

### [36:08]
**Original:** If you're getting started on Cowork, the first thing that you really need to do is connect all the data sources that are relevant to your role, because Cowork can only do a great job if it has access to all the context that it needs to be able to curate the output for you.

**Translation:** 如果你刚开始使用Cowork,你真正需要做的第一件事就是连接所有与你的角色相关的数据源,因为Cowork只有在能够访问它需要的所有上下文信息时才能做好工作,以便为你策划输出内容。

**Core structure:**
- The first thing you need to do is connect all the data sources because Cowork can only do a great job if it has access to all the context.  
  你需要做的第一件事是连接所有数据源,因为Cowork只有在能访问所有上下文时才能做好工作。

**Structure tree:**
```
conditional clause: If you're getting started...
main clause: the first thing is...
predicative clause: connect all the data sources
causal clause: because Cowork can only do...
nested conditional: if it has access to...
purpose clause: to be able to curate...
```

**Grammar points:**
- **多层嵌套从句** - 条件从句+表语从句+原因从句+条件从句+目的状语,层层嵌套
- **定语从句修饰** - that are relevant / that it needs 修饰前面的名词
- **be able to 结构** - 表示能力,此处强调Cowork需要具备的能力

### [36:23]
**Original:** So what that means for me is I connect it to my Google Calendar, I connect it to my Slack, to my Gmail, to my Google Drive, so that it just knows it has the flexibility to find relevant context, to ask questions, to pull in threads, and this substantially improves the quality of the result.

**Translation:** 所以这对我来说意味着,我把它连接到我的Google日历、Slack、Gmail和Google Drive,这样它就知道它有灵活性去找到相关的上下文、提出问题、提取线索,而这大大提高了结果的质量。

**Core structure:**
- What that means is I connect it to my tools so that it has the flexibility to find context, and this improves the quality.  
  这意味着我把它连接到我的工具,这样它就有灵活性找到上下文,这提高了质量。

**Structure tree:**
```
main clause: What that means is...
subject clause: What that means for me
predicative clause: I connect it to...
purpose clause: so that it knows...
object clause: it has the flexibility...
parallel infinitives: to find / to ask / to pull in
result clause: and this improves...
```

**Grammar points:**
- **What 引导主语从句** - What that means 作主语,表示'那意味着什么'
- **so that 目的状语从句** - 表示连接这些工具的目的
- **并列不定式结构** - to find / to ask / to pull in 三个并列的动作

### [36:42]
**Original:** The kinds of things I use it for are, like last night I was working where we have this Cowork Cloud conference coming up and there's a few talks that I'm giving there, and one of the talks that we're doing talks about the transition of Cowork from an assistant to like a full-on agent, and one of the things that I wanted to do in this talk was to showcase all of the products that we've been shipping that to enable this transition.

**Translation:** 我用它做的事情类型是,比如昨晚我在工作,我们即将举办Cowork Cloud会议,我要在那里做几个演讲,其中一个演讲是关于Cowork从助手到完整代理的转变,而我想在这个演讲中做的一件事就是展示我们一直在发布的所有产品,以实现这一转变。

**Core structure:**
- The things I use it for are: I was working on a conference talk about the transition of Cowork, and I wanted to showcase all the products we've been shipping.  
  我用它做的事情是:我在准备一个关于Cowork转变的会议演讲,我想展示我们一直在发布的所有产品。

**Structure tree:**
```
main clause: The kinds of things are...
relative clause: I use it for
example clause: like last night I was working...
relative clause: where we have this conference
relative clause: that I'm giving there
relative clause: that we're doing
relative clause: that I wanted to do
predicative clause: to showcase all the products
relative clause: that we've been shipping
```

**Grammar points:**
- **多重定语从句叠加** - 多个that/where引导的定语从句层层修饰,造成理解困难
- **口语化插入语** - like 作为口语标记词插入,打断句子流畅性
- **现在完成进行时** - we've been shipping 表示持续到现在的动作

### [40:02]
**Original:** And I think this is like an example of what the role of the PM still is today. It's like Claude is a great brainstorming partner. It's able to synthesize a massive amount of information really quickly and present all of the possibilities to you.

**Translation:** 我认为这就像是一个例子,说明了PM的角色在今天仍然是什么。就像Claude是一个很好的头脑风暴伙伴。它能够非常快速地综合大量信息,并向你展示所有的可能性。

**Core structure:**
- This is an example of what the role of the PM is. Claude is able to synthesize information and present possibilities.  
  这是PM角色的一个例子。Claude能够综合信息并展示可能性。

**Structure tree:**
```
main clause: this is an example
predicative clause: of what the role is
explanation: Claude is a partner
ability clause: It's able to synthesize...
parallel structure: synthesize... and present...
```

**Grammar points:**
- **what 引导宾语从句** - what the role of the PM still is 作介词of的宾语
- **be able to 表能力** - 强调Claude具备的能力

### [42:02]
**Original:** So my stack is pretty heavily Claude, Code, co-work. Anthropic largely runs on Slack. I feel like it's like the core OS of our company and day-to-day. A lot of—I would say maybe 30% of my time is pushing the boundaries of what co-work can do so that I have a very strong sense of what we're not good at.

**Translation:** 所以我的工具栈主要是Claude、Code和co-work。Anthropic主要在Slack上运行。我觉得它就像是我们公司日常工作的核心操作系统。很多——我会说大概30%的时间是在推动co-work能做什么的边界,这样我就能非常清楚地知道我们不擅长什么。

**Core structure:**
- 30% of my time is pushing the boundaries of what co-work can do so that I have a sense of what we're not good at.  
  30%的时间是在推动co-work的边界,这样我就知道我们不擅长什么。

**Structure tree:**
```
main clause: 30% of my time is pushing...
object: the boundaries of what co-work can do
purpose clause: so that I have a sense
object clause: of what we're not good at
```

**Grammar points:**
- **what 引导宾语从句** - what co-work can do / what we're not good at 两个宾语从句
- **so that 目的状语从句** - 表示推动边界的目的是为了了解不足

### [43:41]
**Original:** And so it'll pull out things like, okay, this customer is using like Bedrock or Claude for Enterprise or Console, which affects what features are available to them.

**Translation:** 所以它会提取出这样的信息,比如,这个客户正在使用Bedrock或Claude企业版或控制台,这会影响他们可以使用哪些功能。

**Core structure:**
- It'll pull out things which affects what features are available.  
  它会提取出影响可用功能的信息。

**Structure tree:**
```
main clause: it'll pull out things
inserted example: okay, this customer is using...
relative clause: which affects...
noun clause: what features are available to them
```

**Grammar points:**
- **非限制性定语从句** - which指代前面整个情况,补充说明影响
- **what引导宾语从句** - what在从句中作主语,表示'什么样的功能'
- **口语化插入语** - okay和like作为填充词,增加理解难度

### [44:11]
**Original:** And then for example, if this is a customer that's on Vertex or Bedrock and doesn't want to use Claude for Enterprise, then we'll just take out some of the slides that are Claude for Enterprise only features.

**Translation:** 然后举个例子,如果这是一个使用Vertex或Bedrock并且不想使用Claude企业版的客户,那么我们就会删除一些仅限Claude企业版功能的幻灯片。

**Core structure:**
- If this is a customer, then we'll take out some slides.  
  如果这是某类客户,那么我们会删除一些幻灯片。

**Structure tree:**
```
conditional: if this is a customer...
relative clause 1: that's on Vertex or Bedrock
relative clause 2: and doesn't want to use...
main clause: we'll take out slides
relative clause 3: that are Claude for Enterprise only features
```

**Grammar points:**
- **条件状语从句** - if...then结构,then可省略
- **多重定语从句嵌套** - customer和slides都有that引导的定语从句修饰
- **并列谓语** - is on...and doesn't want共同修饰customer

### [44:27]
**Original:** And so normally this is like manual work that could take 20 to 30 minutes, and so people either like spend that time doing it or they'll just decide not to do it and use the general deck.

**Translation:** 所以通常这是需要20到30分钟的手动工作,因此人们要么花时间去做,要么就决定不做而使用通用版本。

**Core structure:**
- This is manual work, and people either spend time doing it or decide not to do it.  
  这是手动工作,人们要么花时间做,要么决定不做。

**Structure tree:**
```
main clause 1: this is manual work
relative clause: that could take 20 to 30 minutes
main clause 2: people either spend time... or decide not to...
parallel structure: spend time doing it / decide not to do it and use the general deck
```

**Grammar points:**
- **either...or并列结构** - 连接两个选择,注意or后还有and连接的动作
- **定语从句** - that修饰manual work,说明工作耗时

### [48:36]
**Original:** And so what they often use Cowork to do is the night before they'll ask it to summarize, okay, what are all my customer meetings that are coming up the next day?

**Translation:** 所以他们经常用Cowork做的事情是,在前一天晚上他们会让它总结,好的,我第二天即将进行的所有客户会议有哪些?

**Core structure:**
- What they use Cowork to do is they'll ask it to summarize meetings.  
  他们用Cowork做的是让它总结会议。

**Structure tree:**
```
subject clause: what they often use Cowork to do
main clause: is the night before they'll ask it to summarize
time adverbial: the night before
indirect question: what are all my customer meetings...
relative clause: that are coming up the next day
```

**Grammar points:**
- **what引导主语从句** - what从句作主语,表示'他们做的事情'
- **间接疑问句** - summarize后接疑问句作宾语,保持疑问词顺序
- **时间状语位置** - the night before插入在系动词is之后,增加理解难度

### [49:04]
**Original:** Customer asked, okay, when is feature X going to launch? CoWork can help the PM person research through Slack to get the latest ETA, add that to the notes so that during the customer call the PM person has the absolute latest.

**Translation:** 客户问道,好的,功能X什么时候会发布?CoWork可以帮助产品经理通过Slack进行研究以获取最新的预计时间,将其添加到笔记中,这样在客户电话会议期间,产品经理就能掌握最新信息。

**Core structure:**
- CoWork can help research, add that to notes so that the PM has the latest.  
  CoWork可以帮助研究,添加到笔记中,以便产品经理掌握最新信息。

**Structure tree:**
```
main clause: CoWork can help the PM person research...
parallel verbs: research / add that to the notes
purpose clause: so that during the call the PM has the latest
time adverbial: during the customer call
```

**Grammar points:**
- **并列动词结构** - help后接research和add两个并列动作
- **so that目的状语从句** - 表示目的,从句中插入时间状语during the call

### [50:50]
**Original:** I think we also believe a lot in empowering our internal teams to build as fast as possible, and we also trust that everyone understands how much capacity that serving these models truly cost, and we trust our team to use the tokens responsibly.

**Translation:** 我认为我们也非常相信赋能我们的内部团队尽可能快地构建，我们也相信每个人都理解服务这些模型真正需要多少容量成本，我们相信我们的团队会负责任地使用这些tokens。

**Core structure:**
- We believe in empowering teams, trust that everyone understands the cost, and trust our team to use tokens responsibly.  
  我们相信赋能团队，相信每个人理解成本，并相信我们的团队会负责任地使用tokens。

**Structure tree:**
```
main: We believe... and trust... and trust...
clause 1: believe in empowering teams to build
clause 2: trust that everyone understands how much...
  embedded question: how much capacity serving models cost
clause 3: trust team to use tokens responsibly
```

**Grammar points:**
- **并列结构** - 三个并列的动词短语(believe, trust, trust)，用and连接
- **宾语从句 + 间接疑问句** - trust that... 后接宾语从句，其中包含how much引导的间接疑问句

### [52:04]
**Original:** Can set a direction and can steadily execute towards it and change the path if the model capabilities are much better than or worse than what they had originally expected.

**Translation:** 能够设定方向并稳步朝着它执行，如果模型能力比他们最初预期的要好得多或差得多，就改变路径。

**Core structure:**
- PMs can set direction, execute towards it, and change the path if needed.  
  产品经理能设定方向，朝它执行，并在需要时改变路径。

**Structure tree:**
```
main: can set... and execute... and change...
condition: if capabilities are better/worse than expected
comparison: better than or worse than what they expected
```

**Grammar points:**
- **省略主语的并列结构** - 三个can并列，主语(best PMs)被省略
- **比较级 + what从句** - better/worse than后接what引导的宾语从句

### [52:31]
**Original:** You can actually just have a text box again where you tell the model what you want, and it's so smart that it can add any tool or add any integration that it needs to get the job done.

**Translation:** 你实际上可以再次只有一个文本框，在那里你告诉模型你想要什么，它非常聪明，可以添加任何它需要的工具或集成来完成工作。

**Core structure:**
- You can have a text box where you tell the model what you want, and it's smart enough to add tools.  
  你可以有一个文本框告诉模型你想要什么，它足够聪明可以添加工具。

**Structure tree:**
```
main 1: You can have a text box
  where clause: where you tell the model what you want
main 2: it's so smart that...
  result clause: it can add any tool/integration
    purpose: to get the job done
```

**Grammar points:**
- **so...that结构** - 表示'如此...以至于'，引出结果状语从句
- **定语从句 + 宾语从句嵌套** - where引导定语从句，其中包含what引导的宾语从句

### [53:32]
**Original:** One of the things I really like to do is to ask the model to introspect on its own behaviors, so sometimes when I notice that the model does something unexpected, like for example, there are situations where the model will make a front-end change and run tests but not actually use the UI, it's actually pretty useful to ask the model to reflect on why it did this.

**Translation:** 我真正喜欢做的事情之一是要求模型反思自己的行为，所以有时当我注意到模型做了一些意外的事情，比如说，有些情况下模型会做前端更改并运行测试但实际上不使用UI，要求模型反思为什么这样做实际上是非常有用的。

**Core structure:**
- One thing I like to do is ask the model to introspect, so when I notice unexpected behavior, it's useful to ask why.  
  我喜欢做的一件事是让模型反思，所以当我注意到意外行为时，询问原因很有用。

**Structure tree:**
```
main: One thing I like is to ask...
so clause: when I notice unexpected behavior
  example insertion: like for example, situations where...
conclusion: it's useful to ask the model to reflect
```

**Grammar points:**
- **主语从句 + 表语从句** - One of the things作主语，is后接不定式作表语
- **插入语** - like for example插入句中，打断主要结构，增加理解难度
- **时间状语从句嵌套** - when从句中包含that引导的宾语从句

### [54:42]
**Original:** Usually there's like a handful of people who are much better than others at articulating what makes a specific model or model harness combination good.

**Translation:** 通常有少数几个人在阐述是什么使特定模型或模型工具组合变好这方面比其他人要好得多。

**Core structure:**
- There are people who are better at articulating what makes a model good.  
  有些人更擅长阐述是什么使模型变好。

**Structure tree:**
```
main: there are people
who clause: who are better at articulating
  what clause: what makes combination good
    subject: what (充当makes的主语)
    object: combination
    complement: good
```

**Grammar points:**
- **定语从句 + 宾语从句嵌套** - who引导定语从句，其中articulating后接what引导的宾语从句
- **what作主语的从句** - what makes...good中，what既引导从句又作makes的主语

### [58:24]
**Original:** So that informs what data we look at to verify, okay, is this a larger pattern.

**Translation:** 所以这会告诉我们应该查看哪些数据来验证，好吧，这是否是一个更大的模式。

**Core structure:**
- That informs what data we look at.  
  这告诉我们应该查看哪些数据。

**Structure tree:**
```
main clause: That informs what data we look at
object clause: what data we look at
purpose: to verify...
indirect question: is this a larger pattern
```

**Grammar points:**
- **what 引导宾语从句** - what data 作 look at 的宾语
- **不定式表目的** - to verify 说明查看数据的目的
- **间接疑问句** - verify 后接 whether/if 类疑问句

### [01:00:25]
**Original:** I think part of what makes a great coworker is this positivity, this like bias towards action, this ability to give you like earnest feedback, not just agreeing with every single thing that you say.

**Translation:** 我认为成为一个优秀同事的部分原因是这种积极性，这种倾向于行动的偏好，这种给你真诚反馈的能力，而不仅仅是同意你说的每一件事。

**Core structure:**
- Part of what makes a great coworker is this positivity and ability.  
  成为优秀同事的部分原因是这种积极性和能力。

**Structure tree:**
```
main clause: Part is this positivity...
subject: part of what makes a great coworker
what clause: what makes a great coworker
predicative: positivity, bias, ability
participle phrase: not just agreeing...
```

**Grammar points:**
- **what 引导的名词性从句** - what makes... 作 part of 的宾语
- **并列结构** - 三个 this 引导的并列表语
- **现在分词短语作状语** - not just agreeing 补充说明能力的内涵

### [01:01:10]
**Original:** So a lot of times we add features to the product as a crutch for the model because it's not naturally doing it itself.

**Translation:** 所以很多时候我们会给产品添加功能，作为模型的拐杖，因为它不会自然地自己做这件事。

**Core structure:**
- We add features as a crutch because the model is not doing it.  
  我们添加功能作为拐杖，因为模型不会做这件事。

**Structure tree:**
```
main clause: we add features to the product
purpose: as a crutch for the model
reason clause: because it's not naturally doing it itself
```

**Grammar points:**
- **as 表示作为/当作** - as a crutch 说明添加功能的角色
- **because 引导原因状语从句** - 解释为什么需要添加功能

### [01:01:45]
**Original:** Similar to how in VS Code you would look up all the call sites and it would be a list on the left side and you would go through them one by one and replace all.

**Translation:** 类似于在 VS Code 中你会查找所有调用位置，它会在左侧显示一个列表，然后你会逐个检查它们并全部替换。

**Core structure:**
- Similar to how you would look up call sites and go through them.  
  类似于你会查找调用位置并逐个检查它们。

**Structure tree:**
```
prepositional phrase: Similar to how...
how clause: how you would look up...
parallel structure: would look up... and it would be... and you would go through... and replace
```

**Grammar points:**
- **how 引导方式从句** - how 从句描述具体操作方式
- **并列谓语结构** - 四个 would 动词短语并列
- **one by one 副词短语** - 修饰 go through，表示逐个进行

### [01:03:04]
**Original:** We can remove a lot of prompting interventions every time the model gets smarter. And we actually do this every time we launch a model. We read through the entire system prompt and we reflect on, okay, for each of these sections, does the model really need this reminder anymore?

**Translation:** 每次模型变得更智能时，我们都可以移除很多提示干预。而且我们实际上每次发布模型时都会这样做。我们会通读整个系统提示，然后反思，好吧，对于这些部分中的每一个，模型真的还需要这个提醒吗？

**Core structure:**
- We read through the prompt and reflect on whether the model needs this reminder.  
  我们通读提示并反思模型是否需要这个提醒。

**Structure tree:**
```
parallel clauses: We read... and we reflect on...
reflect on: indirect question
indirect question: does the model really need this reminder
prepositional phrase: for each of these sections
```

**Grammar points:**
- **reflect on 后接间接疑问句** - 疑问句作 reflect on 的宾语
- **插入语 okay** - 口语化表达，引出思考内容
- **anymore 表示不再** - 用于否定或疑问句，表示时间变化

### [01:03:36]
**Original:** We tried to build a code review product a few times and we've launched like simpler versions of code review which is the slash code review command in the past and it was only with the most recent models that we felt like okay this code review is so good that our engineering team relies on this code review to pass before we merge PRs and we found that this was we've always dreamed of Quad being able to be a reliable code reviewer that can actually that we can like confidently feel catches the majority of bugs.

**Translation:** 我们尝试过几次构建代码审查产品，我们过去推出过更简单的代码审查版本，也就是斜杠代码审查命令，只有在使用最新模型时我们才觉得，好的，这个代码审查非常好，我们的工程团队在合并PR之前依赖这个代码审查来通过，我们发现这就是我们一直梦想的——Quad能够成为一个可靠的代码审查员，能够让我们有信心地感觉它能捕获大多数错误。

**Core structure:**
- We tried to build a product and it was only with recent models that we felt this was good.  
  我们尝试构建产品，只有在使用最新模型时我们才觉得这很好。

**Structure tree:**
```
compound sentence with multiple coordinated clauses:
- we tried... and we've launched...
- and it was only with... that we felt...
- nested: that our team relies on...
- nested: that we've always dreamed of...
- nested: that can actually... catches...
```

**Grammar points:**
- **强调句型 it was only... that** - 强调时间状语 'with the most recent models'
- **多重嵌套从句** - 包含多个 that 引导的从句，层层嵌套，增加理解难度
- **口语化插入语** - 'like' 和 'okay' 等口语标记打断句子流畅性

### [01:04:52]
**Original:** It's pretty important to build products that don't necessarily work yet so that you know what is missing for this product to work, and then with the newest model you can just swap it in to the prototype you've already made and see, okay, does this new model close that gap.

**Translation:** 构建那些目前还不一定能工作的产品是非常重要的，这样你就知道这个产品要工作还缺少什么，然后有了最新模型，你就可以把它替换到你已经做好的原型中，看看，好的，这个新模型是否能填补那个差距。

**Core structure:**
- It's important to build products so that you know what is missing, and then you can swap in the newest model.  
  构建产品很重要，这样你就知道缺少什么，然后你可以替换最新模型。

**Structure tree:**
```
main: It's important to build products...
purpose clause: so that you know...
object clause: what is missing...
coordinated: and then... you can swap it in...
parenthetical question: does this model close that gap
```

**Grammar points:**
- **so that 引导目的状语从句** - 表示构建产品的目的
- **what 引导宾语从句** - what is missing 作 know 的宾语

### [01:06:46]
**Original:** How do we build the interface so that you as a human know which tasks you need to look into?

**Translation:** 我们如何构建界面，以便你作为人类知道需要查看哪些任务？

**Core structure:**
- How do we build the interface so that you know which tasks to look into?  
  我们如何构建界面以便你知道要查看哪些任务？

**Structure tree:**
```
main question: How do we build...
purpose clause: so that you know...
object clause: which tasks you need to look into
appositive: you as a human
```

**Grammar points:**
- **疑问词 + 不定式结构** - which tasks you need to look into 表示间接疑问
- **同位语插入** - 'as a human' 作为 'you' 的同位语，强调人类角色

### [01:07:05]
**Original:** How do we make sure that this process is self-improving so that when you do see a task that isn't done to your liking, you can give it feedback and the model will know for every future run to incorporate that feedback so it never makes that mistake again.

**Translation:** 我们如何确保这个过程是自我改进的，这样当你确实看到一个没有按你的喜好完成的任务时，你可以给它反馈，模型会知道在未来的每次运行中都要整合那个反馈，这样它就不会再犯那个错误了。

**Core structure:**
- How do we make sure the process is self-improving so that you can give feedback and the model will know to incorporate it.  
  我们如何确保过程是自我改进的，这样你可以给反馈，模型会知道整合它。

**Structure tree:**
```
main: How do we make sure...
that-clause: that this process is self-improving
purpose: so that when you see...
time clause: when you do see a task...
relative: that isn't done to your liking
result: the model will know... to incorporate...
purpose: so it never makes...
```

**Grammar points:**
- **多层嵌套的目的状语从句** - 两个 'so that' 从句表示递进的目的关系
- **时间状语从句中的强调** - 'do see' 使用助动词 do 强调动作
- **know + 不定式** - 'know to incorporate' 表示知道要做某事

### [01:09:45]
**Original:** I would also push listeners towards focusing on bringing your automations from 'okay, this is a cool concept' to like 'hey, this actually works 100% of the time.'

**Translation:** 我还会推动听众专注于将你的自动化从'好的，这是个很酷的概念'提升到'嘿，这实际上100%的时间都有效'。

**Core structure:**
- I would push listeners towards bringing automations from concept to working 100% of the time.  
  我会推动听众将自动化从概念提升到100%有效。

**Structure tree:**
```
main: I would push listeners towards...
gerund phrase: focusing on bringing...
from-to structure: from 'concept' to 'works 100%'
quoted speech: direct quotations as endpoints
```

**Grammar points:**
- **push sb towards doing sth** - 推动某人做某事的固定搭配
- **bring sth from A to B** - 将某物从A状态转变到B状态

### [01:11:34]
**Original:** A lot easier because right now I think you have to know too many concepts. You have to know to define a skill. You have to know to use this skill and give it feedback. And then you have to know to tell Cowork to update the skill based on all the feedback that you gave. And then you also have to know where to read the skill to make sure that the feedback was incorporated the way that you want.

**Translation:** 容易得多，因为现在我认为你必须了解太多概念。你必须知道如何定义一个技能。你必须知道如何使用这个技能并给它反馈。然后你还必须知道如何告诉Cowork根据你给出的所有反馈来更新技能。然后你还必须知道在哪里阅读技能以确保反馈按照你想要的方式被整合进去。

**Core structure:**
- It's easier because you have to know too many concepts.  
  这更容易，因为你必须了解太多概念。

**Structure tree:**
```
main: A lot easier because...
reason clause: because you have to know...
parallel structures: You have to know to [verb]... (repeated 5 times)
relative clause: that you gave
purpose clause: to make sure that...
embedded clause: the way that you want
```

**Grammar points:**
- **并列重复结构** - 五个平行的 'You have to know to...' 结构，强调复杂性
- **嵌套从句** - 多层从句嵌套：based on... that you gave / to make sure that... the way that...
- **know to do** - know + to不定式，表示'知道如何做某事'

### [01:13:07]
**Original:** MCPs and these like workflow improvements, and I think sometimes that can even distract from your core goal of like launching some product or building some feature.

**Translation:** MCP和这些工作流程改进，我认为有时这甚至会分散你的核心目标，比如推出某个产品或构建某个功能。

**Core structure:**
- That can distract from your core goal.  
  这会分散你的核心目标。

**Structure tree:**
```
main: that can distract from your core goal
prepositional phrase: of launching... or building...
modal verb: can (possibility)
gerund phrases: launching / building (parallel)
```

**Grammar points:**
- **distract from** - 固定搭配，表示'使分心、转移注意力'
- **介词of + 动名词** - of后接并列的动名词短语，说明goal的具体内容

### [01:13:59]
**Original:** There's this Karpathy tweet that just came out yesterday where he talked about this divide that's interesting between people that tried ChatGPT and Claude back in the day. It was like okay and they're like nah this is terrible and they kind of gave up on like what AI could do for them and they're just like so cynical of like no way it's not actually that big of a deal and then there's people that are using it to code essentially who see the full intense power of it and how good it is and people on both sides don't understand the other side and why they like how much they see the world and so your advice is really good here just like actually use it for real things and see how good it actually has gotten.

**Translation:** 有一条Karpathy昨天刚发的推文，他在推文中谈到了一个有趣的分歧，即那些在早期尝试过ChatGPT和Claude的人之间的分歧。他们觉得还行，然后又觉得不行这太糟糕了，他们有点放弃了AI能为他们做什么，他们就是非常愤世嫉俗地认为不可能它实际上没那么了不起，然后还有一些人本质上在用它来编程，他们看到了它全部的强大力量以及它有多好，而双方的人都不理解另一方以及他们为什么如此看待世界，所以你的建议在这里真的很好，就是真正地用它来做实际的事情，看看它实际上变得有多好。

**Core structure:**
- There's a tweet where he talked about a divide between people, and then there's people who see the power, and people don't understand the other side.  
  有一条推文谈到了人们之间的分歧，然后有些人看到了力量，而人们不理解另一方。

**Structure tree:**
```
main 1: There's this tweet that... where he talked about...
main 2: then there's people that... who see...
main 3: and people don't understand...
conclusion: so your advice is good... just use it... and see...
multiple relative clauses: that/who/where
embedded clauses: what AI could do / how good it is / how much they see
```

**Grammar points:**
- **口语化run-on sentence** - 多个独立句子用and连接，缺少标点，模仿口语流
- **多重定语从句嵌套** - tweet that... where... / people that... who... / divide that's... between people that...
- **like作填充词** - 口语中的like用作犹豫标记，无实际意义

### [01:14:38]
**Original:** Yeah I think the big shift is that the 2024 generation of products were chat-based and the current code generation of products is action-based.

**Translation:** 是的，我认为最大的转变是2024年这一代产品是基于聊天的，而当前这一代代码生成产品是基于行动的。

**Core structure:**
- The shift is that products were chat-based and products are action-based.  
  转变是产品曾是基于聊天的，现在是基于行动的。

**Structure tree:**
```
main: the shift is that...
predicative clause: that products were... and products is...
contrast: 2024 generation (past) vs current generation (present)
compound predicate: were chat-based and is action-based
```

**Grammar points:**
- **表语从句** - that从句作is的表语，说明shift的内容
- **对比结构** - 用and连接两个时态不同的并列句，强调过去与现在的对比
- **-based复合形容词** - chat-based/action-based表示'以...为基础的'

### [01:15:51]
**Original:** The other book that I'm really into is The Technology Trap. So this is actually about the past few technology revolutions, so the Industrial Revolution and the computer revolution, and how this has affected workers.

**Translation:** 我真正喜欢的另一本书是《技术陷阱》。所以这本书实际上是关于过去几次技术革命的，比如工业革命和计算机革命，以及这如何影响了工人。

**Core structure:**
- The book is The Technology Trap. This is about revolutions and how this affected workers.  
  这本书是《技术陷阱》。这是关于革命以及这如何影响工人的。

**Structure tree:**
```
main 1: The book that I'm into is...
main 2: This is about revolutions... and how...
relative clause: that I'm really into
appositive: the Industrial Revolution and the computer revolution
noun clause: how this has affected workers
```

**Grammar points:**
- **be into** - 口语表达，表示'对...感兴趣、喜欢'
- **how引导名词性从句** - how从句作about的宾语，说明影响方式

### [01:16:58]
**Original:** And I think similarly, it's just such a pure achievement to be able to climb this extremely challenging, dangerous route and to be able to have the mental focus to do it knowing that if you make a single mistake, you die.

**Translation:** 我认为同样地,能够攀登这条极具挑战性、危险的路线,并且能够拥有精神专注力去完成它——明知道如果你犯一个错误就会死——这是一项如此纯粹的成就。

**Core structure:**
- It's such a pure achievement to be able to climb this route and to have the mental focus to do it.  
  能够攀登这条路线并拥有精神专注力去完成它是一项纯粹的成就。

**Structure tree:**
```
main clause: it's such a pure achievement
to-infinitive 1: to be able to climb this route
to-infinitive 2: to be able to have the mental focus to do it
participle clause: knowing that if you make a mistake, you die
conditional clause: if you make a single mistake
```

**Grammar points:**
- **形式主语 it** - 真正主语是后面的 to-infinitive 结构
- **现在分词作状语** - knowing 引导伴随状语,表示在攀登时的心理状态
- **嵌套从句** - knowing that 后接宾语从句,其中又包含条件从句 if you make a mistake

### [01:17:31]
**Original:** It's one of the rare movies where like the more you know about it, the more you're blown away by how insane this is.

**Translation:** 这是少数几部你对它了解越多,就越会被它的疯狂程度震撼的电影之一。

**Core structure:**
- It's one of the rare movies where you're blown away.  
  这是少数几部让你震撼的电影之一。

**Structure tree:**
```
main clause: It's one of the rare movies
relative clause: where the more you know, the more you're blown away
comparative structure: the more...the more...
object clause: how insane this is
```

**Grammar points:**
- **the more...the more 结构** - 比较级递进结构,表示两个变化同步增长
- **where 引导定语从句** - 修饰 movies,在从句中作地点状语

### [01:18:35]
**Original:** But one thing I really appreciate about the Waymo is I can call into a work call. I'm not worried about someone overhearing me. I'm not worried about, hey, is this like rude? Am I talking too loud? Do I need to ask someone to like change the music?

**Translation:** 但我真正欣赏 Waymo 的一点是,我可以打工作电话。我不担心有人会偷听我。我不担心,嘿,这样是不是很粗鲁?我说话声音太大了吗?我需要让别人换音乐吗?

**Core structure:**
- One thing I appreciate is I can call into a work call.  
  我欣赏的一点是我可以打工作电话。

**Structure tree:**
```
main clause: one thing I appreciate is...
relative clause: (that) I really appreciate about the Waymo
predicative clause: I can call into a work call
parallel clauses: I'm not worried about... (multiple)
```

**Grammar points:**
- **省略 that 的定语从句** - I really appreciate 修饰 one thing,关系代词 that 被省略
- **表语从句** - is 后接完整句子作表语
- **间接疑问句** - worried about 后接多个疑问句,保持疑问语序

### [01:19:38]
**Original:** I think there's a lot of value in first principles thinking and if you know what you're optimizing for and you have strong first principles, then you can normally deduce what the right course of action is and be able to clearly articulate that to all the stakeholders and then you should just do it.

**Translation:** 我认为第一性原理思维有很大价值,如果你知道你在优化什么,并且你有强大的第一性原则,那么你通常可以推断出正确的行动方案是什么,并能够向所有利益相关者清楚地阐述,然后你就应该去做。

**Core structure:**
- There's value in first principles thinking and if you know what you're optimizing for, then you can deduce what the right course is and you should do it.  
  第一性原理思维有价值,如果你知道你在优化什么,那么你可以推断出正确方案并去做。

**Structure tree:**
```
main clause: there's a lot of value in first principles thinking
conditional: if you know...and you have...
object clause 1: what you're optimizing for
object clause 2: what the right course of action is
result clause: then you can deduce...and you should do it
```

**Grammar points:**
- **if...then 条件句** - 复合条件(两个并列条件)引导结果
- **what 引导宾语从句** - what 在从句中作宾语或主语,表示'...的事情'
- **并列谓语** - can deduce, be able to articulate, should do 三个动作并列

### [01:21:04]
**Original:** And it was like I really appreciate Alex and the rest of the team for like empowering me and the rest of the team to just like figure things out without any boundaries for what sales is supposed to do, what ops is supposed to do, what engineer is supposed to do.

**Translation:** 我真的很感激 Alex 和团队其他成员赋予我和团队权力,让我们在没有任何界限的情况下解决问题——不管销售应该做什么、运营应该做什么、工程师应该做什么。

**Core structure:**
- I appreciate Alex for empowering me to figure things out without boundaries.  
  我感激 Alex 赋予我权力在没有界限的情况下解决问题。

**Structure tree:**
```
main clause: I appreciate Alex and the team
prepositional phrase: for empowering me and the team
to-infinitive: to figure things out
prepositional phrase: without any boundaries for...
parallel clauses: what sales/ops/engineer is supposed to do
```

**Grammar points:**
- **appreciate sb for doing sth** - 感激某人做某事的固定搭配
- **what 引导的并列宾语从句** - 三个 what 从句并列,作 boundaries for 的宾语

### [01:23:49]
**Original:** And then the thing that is most helpful is tell us where Claude Code and Cowork aren't working well for you.

**Translation:** 然后最有帮助的事情是告诉我们 Claude Code 和 Cowork 在哪些方面对你来说运行得不好。

**Core structure:**
- The thing is tell us where they aren't working well.  
  最有帮助的事情是告诉我们它们在哪里运行得不好。

**Structure tree:**
```
main clause: The thing is [to] tell us...
subject: the thing that is most helpful
predicative: tell us where...
embedded question: where Claude Code and Cowork aren't working well
```

**Grammar points:**
- **定语从句修饰主语** - that is most helpful 修饰 the thing
- **间接疑问句作宾语** - where 引导的从句作 tell 的宾语
- **省略 to 的不定式** - 系动词 is 后省略了 to，完整形式为 is to tell

### [01:24:13]
**Original:** Because if you're able to share that with us and we're able to reproduce it, then this is something that we're able to actively improve for our next generations of models and for our next harnesses.

**Translation:** 因为如果你能够与我们分享这些问题，而我们能够重现它，那么这就是我们能够为下一代模型和下一代工具积极改进的东西。

**Core structure:**
- If you share it and we reproduce it, then this is something we can improve.  
  如果你分享它，我们重现它，那么这就是我们能改进的东西。

**Structure tree:**
```
conditional: if you're able to share... and we're able to reproduce...
main clause: then this is something that...
relative clause: that we're able to actively improve
prepositional phrases: for our next generations... and for our next harnesses
```

**Grammar points:**
- **复合条件句** - if 从句中用 and 连接两个并列条件
- **定语从句** - that 从句修饰 something
- **并列介词短语** - for... and for... 表示改进的两个方面

### [01:24:44]
**Original:** We have this channel of like user love, and so whenever you guys share a success story, we post it there, and whenever you guys share like issues with our product, we put it into our feedback channel.

**Translation:** 我们有一个用户喜爱频道，所以每当你们分享成功故事时，我们就把它发布在那里，每当你们分享产品问题时，我们就把它放入反馈频道。

**Core structure:**
- We have a channel, and whenever you share stories, we post them, and whenever you share issues, we put them in the feedback channel.  
  我们有一个频道，每当你们分享故事时我们发布它们，每当你们分享问题时我们把它们放入反馈频道。

**Structure tree:**
```
main clause 1: We have this channel
main clause 2: whenever you share..., we post it
main clause 3: whenever you share..., we put it
temporal clauses: whenever you guys share (×2)
```

**Grammar points:**
- **并列复合句** - 三个主句用 and 连接
- **whenever 引导时间状语从句** - 表示每次发生某事时的动作，出现两次形成平行结构