How Anthropic's product team moves faster than anyone else

[0:00]I think it is very hard to be the right amount of AGI build. 我认为要对 AGI 保持恰到好处的信念是非常困难的。

[0:03]It's very easy to build the product for the super AGI strong model. 为超级 AGI 强模型构建产品其实很简单。

[0:07]The hard thing is figuring out for the current model, 难的是搞清楚对于当前的模型，

[0:10]how do you elicit the maximum capability? 如何激发出最大的能力？

[0:13]I've never seen anything like the pace you folks at Anthropic are shipping at. 我从来没见过像你们 Anthropic 这样的发货速度。

[0:17]We want to remove every single barrier to shipping things. 我们想消除所有阻碍发货的障碍。

[0:20]The timelines for a lot of our product features 我们很多产品功能的开发周期

[0:22]have gone down from six months to one month and sometimes to even one day. 从六个月缩短到了一个月，有时候甚至一天。

[0:26]You're interviewing hundreds of PMs 你在面试几百个 PM

[0:28]and you just keep feeling like they're approaching it very incorrectly. 然后你一直觉得他们的方法完全不对。

[0:31]The PM role is changing a lot. PM 的角色正在发生巨大变化。

[0:34]It's changing really quickly. 变化非常快。

[0:35]The thing that is extremely important for building AI native products 构建 AI 原生产品最关键的一点

[0:38]is iterating so quickly, 就是快速迭代，

[0:40]figuring out a way for you to actually launch features every single week. 想办法真正做到每周都发布新功能。

[0:44]What do you think are the emerging skills PMs need to develop? 你觉得 PM 需要培养哪些新兴技能？

[0:47]It comes back to product taste. 归根结底还是产品品味。

[0:49]As code becomes much cheaper to write, 随着写代码变得越来越廉价，

[0:52]the thing that becomes more valuable is deciding what to write. 更有价值的是决定写什么代码。

[0:56]Today, my guest is Kat Wu, 今天我的嘉宾是 Cat Wu，

[0:58]head of product for Cloud Code and co-work at Anthropic. Anthropic Cloud Code 和 CoWork 的产品负责人。

[1:01]Kat is at the center of everything that is changing in AI and product and building, Cat 处于 AI、产品和构建领域所有变革的核心，

[1:06]and she and her team are building the product 她和她的团队正在打造的产品

[1:08]that is most changing the way that we all build our products. 正在改变我们所有人构建产品的方式。

[1:12]She is so full of insights and wisdom and lessons. 她充满了洞察力和智慧与经验。

[1:16]This is an episode you cannot miss. 这是一集你不能错过的节目。

[1:18]Before we get into it, don't forget to check out Lenny's Product Pass dot com 在我们开始之前，别忘了去 Lenny's Product Pass dot com 看看

[1:21]for an insane set of deals available exclusively to Lenny's newsletter 为 Lenny's newsletter 订阅者

[1:26]and our subscribers. With that, I bring you Kat Wu. 独家提供的超值优惠。接下来，有请 Cat Wu。

[1:31]Kat, welcome to the podcast. Cat，欢迎来到播客。

[1:35]Thanks for having me. 谢谢邀请我。

[1:36]I have so many questions. 我有好多问题想问。

[1:38]I'm so excited to have you on this podcast. 很高兴你能来参加这个播客。

[1:40]I want to start with giving people 我想先让大家了解一下

[1:42]an understanding of your role alongside Boris. 你和 Boris 的角色分工。

[1:46]Everybody knows Boris. 大家都认识 Boris。

[1:47]This is his episode is the number one most popular episode on this podcast. 他那期节目是这个播客上最受欢迎的节目。

[1:51]No pressure. 没有压力。

[1:52]He created Cloud Code. 他创造了 Cloud Code。

[1:54]He leads the team, ships, 他带领团队，发货，

[1:56]a bazillion PRs a day from his phone, just like, I don't even know what the number is 每天从手机上提交无数的 PR，我甚至都不知道数字是多少了。

[1:59]anymore. I think people don't give you enough credit for the success that Cloud 我觉得大家没有给你足够的认可，Cloud Code

[2:04]Code has had and co-work and all the things you all are building. 以及 CoWork 和你们正在构建的所有东西取得的成功。

[2:07]Help us understand your role on the team, how you work with Boris, 帮我们了解一下你在团队中的角色，你是怎么和 Boris 合作的，

[2:11]how you split responsibilities, just like what does the PM role look like on 你们怎么分工的，Cloud Code

[2:15]the Cloud Code team? I feel very lucky to work with Boris. 团队的 PM 角色是什么样的？

[2:17]He's been an amazing thought partner. 我觉得很幸运能和 Boris 一起工作。

[2:19]He's our tech lead. 他一直是一个很好的思考伙伴。

[2:20]He's very much the product visionary, and he is great at setting, 他是我们的技术负责人。

[2:25]like this is what the product needs to be in like three months, six months from now. 他在很大程度上是产品远见者，他非常擅长设定

[2:30]This is like what the AGI-pilled version of the product is. 产品在三个月、六个月后应该是什么样的。

[2:33]And a lot of my role is figuring out, OK, what is the path from where we are today 这就是产品在高度 AGI 化之后的样子。

[2:37]to like that vision three to six months from now? 我的角色很大程度上是弄清楚，好吧，从我们现在所处的位置

[2:41]And I spend more of my time on the cross-functional. 到三到六个月后的愿景，路径是什么？

[2:45]So making sure that our marketing team, sales team, finance capacity, 我把更多时间花在跨职能协调上。

[2:50]et cetera, are like bought in on the plan and that we're all rowing the same direction. 确保我们的市场团队、销售团队、财务产能

[2:55]And that once the feature is ready, that there aren't any blockers to shipping it. 等等，都认同这个计划，并且我们都在朝同一个方向努力。

[2:59]I think in many ways it works well because 而且一旦功能准备好了，没有任何障碍可以阻止发货。

[3:02]we kind of like mind meld, but it is actually like remarkably blurry of a line. 我觉得在很多方面这种合作效果很好，因为

[3:06]Like, I think we're like 80 percent mind meld. 我们基本上心意相通，但这条界限实际上非常模糊。

[3:09]And then there's like this 20 percent of things that like maybe I care a lot more 我觉得我们大概 80% 是心有灵犀的。

[3:13]about than Boris, so like I'll drive those and like 20 percent where he cares a lot 然后有大约 20% 的东西可能我比 Boris 更在意，

[3:16]more than me and he just like drives those. 所以我会去推动那些，还有 20% 是他比我更在意的，

[3:19]This episode is brought to you by our season's presenting sponsor, WorkOS. 他就直接推动那些。

[3:24]What do OpenAI 本期节目由我们本季的首席赞助商 WorkOS 带来。

[3:25]Anthropic, Cursor, Vercel, Replit, Sierra, Clay and hundreds of other winning OpenAI、

[3:30]companies all have in common, they are all powered by WorkOS. Anthropic、Cursor、Vercel、Replit、Sierra、Clay 以及数百家其他成功

[3:34]If you're building a product for the enterprise, you've felt the pain 公司有什么共同点？它们都由 WorkOS 驱动。

[3:37]of integrating single sign on, skim, RBAC, audit logs and other features required 如果你在为企业构建产品，你一定感受过那种痛苦

[3:43]by large companies. WorkOS turns those deal blockers into drop in APIs ——集成单点登录、SCIM、RBAC、审计日志以及其他大公司需要的功能。

[3:47]with a modern developer platform built specifically for B2B SaaS. WorkOS 将这些阻碍交易的功能变成了即插即用的 API，

[3:51]Literally every startup that I'm an investor in that starts to expand 这是一个专为 B2B SaaS 构建的现代开发者平台。

[3:55]up market ends up working with WorkOS and that's because they are the best. 我投资的几乎所有初创公司在开始向

[3:59]Whether you are a seed stage startup trying 高端市场扩张时最终都会选择 WorkOS，因为他们是最好的。

[4:01]to land your first enterprise customer or a unicorn expanding globally. 无论你是一家试图

[4:05]WorkOS is the fastest path to becoming enterprise ready and unblocking growth. 拿下第一个企业客户的种子期初创公司，还是一家在全球扩张的独角兽。

[4:09]It's essentially Stripe for enterprise features. WorkOS 是成为企业级就绪和释放增长的最快途径。

[4:12]Visit WorkOS.com to get started or just hit up their Slack where they have actual 它本质上就是企业功能版的 Stripe。

[4:17]engineers waiting to answer your questions. 访问 WorkOS.com 开始使用，或者直接联系他们的 Slack，那里有真正的

[4:19]WorkOS allows you to build faster with delightful APIs, comprehensive docs and a 工程师等着回答你的问题。

[4:24]smooth developer experience. WorkOS 让你能用令人愉悦的 API、全面的文档和

[4:25]Go to WorkOS.com to make your app enterprise ready today. 流畅的开发者体验更快地构建产品。

[4:29]Something that you shared actually before we started recording is the fact that 前往 WorkOS.com，今天就让你的应用为企业级做好准备。

[4:34]you're interviewing hundreds of PMs all the time. 你在录制前分享的一件事是

[4:36]Like if I had a nickel every time someone asked me for an intro to someone at 你一直在面试大量的 PM。

[4:39]Anthropic to go work at Anthropic as a PM, I'd have 30 billion in ARR. 如果每次有人找我要介绍去

[4:44]It's just like the number one place people want to go work at. Anthropic 做 PM，我就能赚 300 亿美元的 ARR。

[4:47]So I can only imagine how many PMs you're interviewing. 那简直就是大家最想去工作的地方。

[4:49]You told me that you're just seeing people doing it wrong the way they're 所以我都能想象你在面试多少 PM。

[4:55]researching what they think it takes to be a successful AI PM. 你告诉我你看到人们做得不对，他们

[4:58]Talk about what you're seeing and what 研究成为成功 AI PM 的方式是错误的。

[4:59]people need to understand about what it is, what it takes to be successful these days. 说说你看到了什么，以及

[5:03]I think before AI, technology shifts were a lot slower. 人们需要理解什么才能在当下取得成功。

[5:07]So you could plan on the six to 12 month time horizons. 我认为在 AI 之前，技术变革的速度慢得多。

[5:11]And because you were shipping features at a bit of a slower rate, there was a lot 所以你可以在六到十二个月的时间跨度上做规划。

[5:16]more emphasis on coordinating with all the other partner teams to make sure that 而且因为你发布功能的速度相对较慢，所以有更多

[5:20]they're shipping features that unblock your features because code at that time was 精力花在与所有合作伙伴团队协调上，确保

[5:25]very expensive to make. 他们发布的功能能解锁你的功能，因为那个时候代码

[5:27]I think now with AI and with how much that has accelerated engineering and with how 非常昂贵。

[5:32]quickly the model capabilities are improving, the timelines for a lot of our 我认为现在有了 AI，加上工程效率的大幅提升，以及模型

[5:37]product features have gone down from six months to one month and sometimes to one 能力提升的速度如此之快，我们很多

[5:42]week or even one day. And with that, we actually need to make 产品功能的开发周期从六个月缩短到了一个月，有时候甚至缩短到一

[5:46]sure that products ship quite quickly. 周甚至一天。这意味着我们实际上必须确保

[5:49]And what that means is as a PM, there should be less emphasis on 产品能非常快地发布。

[5:55]making sure that you're aligning your multi-quarter roadmaps with your partner 这意味着作为 PM，不应该那么强调

[6:00]teams and more emphasis on, okay, how can we figure out the fastest way to get 确保你的多季度路线图与合作伙伴

[6:05]something out the door? How can we figure out how to make a concept corner of our 团队对齐，而应该更多强调，好吧，我们怎么找到最快的途径

[6:10]product suite where we can just... An engineer has an idea or a PM has an idea. 把东西推出去？我们怎么打造一个概念试验田，让

[6:15]And by the end of the week, we are able to get into our users' hands. 工程师有一个想法或者 PM 有一个想法，

[6:18]I think the PMs who do the best on AI native products are the ones 到周末就能交到用户手中。

[6:25]who can figure out how can I shorten the time from having this idea to actually 我认为在 AI 原生产品上做得最好的 PM

[6:29]getting the product in the hands of users and help define what are the most important 是那些能想出如何缩短从有想法到实际

[6:33]tasks that need to work out of the box for my product. 把产品交到用户手中的时间的人，并且能定义哪些是最重要的、

[6:37]So what I love about this is what you're saying is just like people 需要开箱即用的功能。

[6:40]haven't grasped how fast they need to move and how much 我喜欢的是你说的，人们

[6:44]of the job now is just moving, is helping the team move fast. 还没有意识到他们需要多快地行动，以及现在

[6:48]What helps do that? 工作中有多大一部分就是推动行动，就是帮助团队快速前进。

[6:50]What do you do? 什么能帮助做到这一点？

[6:51]What does your PM team do to help them move this fast? 你做了什么？

[6:54]Other than have access to the most advanced models? 你的 PM 团队做了什么来帮助这么快地推进？

[6:57]I think the first thing is to set clear 除了能使用最先进的模型之外？

[7:00]goals because LLMs are so general that actually creates a lot of ambiguity in 我认为首先是设定清晰的

[7:05]who we're building for, what problems we're trying to solve, what the top use cases are. 目标，因为 LLM 太通用了，这实际上在

[7:08]And so I think a great PM is able to say, OK, our key user is professional developers. 我们为谁构建、解决什么问题、核心用例是什么方面造成了很大的模糊性。

[7:15]The main problem that we want to solve for 所以我认为一个优秀的 PM 能说，好吧，我们的核心用户是专业开发者。

[7:18]this feature is maybe there's like too many permission prompts and people are feeling fatigue and like the 我们想为这个功能解决的

[7:24]use case is we want professional developers at enterprises to safely get to zero permission prompts. 主要问题可能是权限提示太多，人们感到疲劳，而

[7:31]And that actually sets a pretty clear goal because it rules out a lot of potential 用例是我们希望企业中的专业开发者安全地实现零权限提示。

[7:36]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，因为它排除了很多潜在的

[7:42]And then I think the second thing that's very important is figuring out some 减少权限提示的方法，让人们可以通过一个提示完成更多事情。

[7:47]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某种

[7:50]So for Cloud Code, what we do is we actually ship almost all 可重复的流程来发布这些功能。

[7:54]of our features in Research Preview. 所以对于 Cloud Code，我们几乎以 Research Preview 的形式

[7:56]We clearly brand this when we ship something so that users 发布所有功能。

[8:00]know that this is an early product, this is just an idea. 我们在发布时会明确标注，这样用户

[8:03]This is just something that we're trying 就知道这是一个早期产品，这只是一个想法。

[8:05]to get feedback on and iterating on and that this might not be supported forever. 这只是我们正在尝试获取

[8:08]And what this does is it reduces our commitment for shipping something. 反馈并不断迭代的东西，这个功能可能不会永远存在。

[8:13]We can just get something out in a week or two. 这样做的好处是降低了我们发布东西的承诺。

[8:16]And the third thing that a PM should do is 我们可以在一两周内就推出一些东西。

[8:19]help create the framework for the team so that they know when to pull in cross-functional partners. PM 应该做的第三件事是

[8:24]And what those cross-functional partners expectations are. 帮助团队建立框架，让他们知道什么时候需要拉入跨职能伙伴。

[8:26]So, for example, we have a really tight process between engineering, marketing and docs. 以及这些跨职能伙伴的期望是什么。

[8:32]So when engineers have a feature that they feel is ready and that we've dog fooded 比如说，我们在工程、市场和文档之间有一个非常紧密的流程。

[8:37]internally, they post it in our evergreen launch room. 所以当工程师觉得某个功能已经准备好并且我们已经内部

[8:40]And then Sarah, who leads our docs, and Alex, who leads PMM and Tarek and Lydia 吃自己狗粮了，他们会把它发到我们永久的发布房间里。

[8:46]on DevRel, just like jump in and can turn around the marketing announcement for it 然后负责文档的 Sarah、负责 PMM 的 Alex 以及 DevRel 的 Tarek 和 Lydia

[8:51]the very next day. And because we have this really tight process, 就会跳进来，第二天就能把市场公告发出来。

[8:54]it lowers the friction for any engineer to ship something. 因为我们有这个非常紧密的流程，

[8:56]And PM is the role that should be setting this up. 任何工程师发布东西的摩擦都降低了。

[8:59]How do PRDs fit into this? 而 PM 就是应该搭建这个的角色。

[9:01]The fact that you said that goals are a really important part, just like being PRD 在这里面是怎么 fits 的？

[9:04]aligned on what does success look like, who is this for, who is this not for? 你说目标是非常重要的部分，就像对齐

[9:07]Are you writing PRDs? Is it just like a couple of bullet points? 成功是什么样的，这是为谁的，不是为谁的？

[9:09]How does how's that evolved in the world of a PM? 你们写 PRD 吗？还是就是几个要点？

[9:11]So there's two things that we do. 这个在 PM 的世界里是怎么演变的？

[9:14]One is we have very rigorous metrics and we do metrics readouts with the entire team every week. 所以我们做两件事。

[9:20]The goal of this is to make sure that everyone deeply understands all 一是我们有非常严格的指标，我们每周都会和整个团队一起做指标回顾。

[9:24]the facets of our business, what our key goals are, how they're trending and what 这样做的目的是确保每个人都深入理解

[9:27]drives them. The second thing that we do is we have this list of team principles 我们业务的所有方面，我们的核心目标是什么，它们的趋势如何，以及

[9:32]and this includes who our key users are, why those are our key users. 什么在驱动它们。第二件事是我们有一份团队原则清单

[9:38]And the reason that we articulate all of ，包括我们的核心用户是谁，为什么这些是我们的核心用户。

[9:40]this is so that everybody on the team feels like they understand how our business works. 我们阐述所有这些

[9:45]They understand what's important to us and what we're willing to trade off. 是为了让团队中的每个人都觉得自己理解我们的业务是如何运作的。

[9:48]And it lets people make decisions by themselves without feeling like they're blocked on PM or 他们理解什么对我们重要，以及我们愿意做什么取舍。

[9:54]any other stakeholder. 这让人们可以自己做决定，而不会觉得被 PM 或

[9:55]I love how so much of this is like, OK, we still need PMs in the future. 任何其他利益相关者阻碍。

[9:58]And there's so much talk of like, why do we need PMs? 我喜欢这里面大部分内容都是，好吧，我们未来仍然需要 PM。

[10:00]We're just going to ship and build. 而网上有那么多讨论说，我们为什么还需要 PM？

[10:01]We need engineers. 我们只需要发布和构建就行了。

[10:02]Oh, we actually do PRD sometimes. 我们需要工程师。

[10:04]So I think for features that are like particularly ambiguous, 哦，我们实际上还是会写 PRD 的。

[10:08]it does help to write out just a one pager on what the goals are, 所以我认为对于那些特别模糊的功能，

[10:12]what the delightful use cases are, what the failure modes currently are that we need to fix. 确实有助于写一页纸来说明目标是什么，

[10:18]And there are occasionally some projects, especially things that require heavy infrastructure. 令人愉悦的用例是什么，目前需要修复的失败模式是什么。

[10:24]That do take many months. 偶尔也会有一些项目，特别是需要大量基础设施的项目。

[10:26]And for those situations, we do write PRD still. 确实需要好几个月。

[10:28]I want to drill a little bit further into just how you're able to move so fast. 对于这些情况，我们确实还是会写 PRD。

[10:34]I've never seen anything like the pace folks at Anthropic are shipping at. 我想进一步深入了解一下你们怎么能这么快。

[10:38]Like someone made this calendar of launches across Anthropic. 我从来没见过 Anthropic 这样的发货速度。

[10:43]And it was literally every day there was like a major feature or product. 有人做了一个 Anthropic 各项发布的日历。

[10:47]So one question people had online is you guys just launched this, not launched, but built this 几乎每天都有重大的功能或产品发布。

[10:53]incredible model, Mythos, that is still in preview because it's so powerful. 所以网上有人问的是你们刚刚——不是发布，而是构建了这个

[10:57]People are a little afraid of what it can do. 不可思议的模型 Mythos，它目前还在预览阶段，因为它太强大了。

[10:59]Have you guys been using this? 人们有点害怕它能做什么。

[11:00]Is this part of the reason you've been able to move so fast? 你们有用过这个吗？

[11:03]We've been moving pretty fast for several quarters now. 这是你们能这么快的原因之一吗？

[11:07]So I think it's not fully Mythos. 我们已经连续好几个季度都在快速推进了。

[11:11]Mythos is an incredibly powerful model. 所以我认为这不完全是 Mythos 的功劳。

[11:13]We do use the models internally. Mythos 是一个非常强大的模型。

[11:15]And I think this has increased our rate 我们确实在内部使用这些模型。

[11:18]of shipping a little bit, but I don't think it explains the bulk of the increase. 我认为这在一定程度上提高了我们的

[11:21]I think a lot of it is the 发布速度，但我不认为这是速度提升的主要原因。

[11:23]process and the expectation on the team. 我认为很大程度上是

[11:25]So we're very low on process. 流程和团队的期望。

[11:27]We want to remove every single barrier to shipping things. 所以我们的流程非常少。

[11:30]We want to make sure every single person on the team feels empowered to take their 我们想消除每一个阻碍发布的东西。

[11:35]idea from just an idea to like out in the world in less than a week, sometimes even in a day. 我们要确保团队里的每一个人都觉得有权力把自己的

[11:41]Cool. Oh, man. 想法从只是一个想法变成一周内发布到世界上，有时候甚至是一天内。

[11:42]What a what an advantage to have the best model and also be building product. 太酷了。天哪。

[11:46]That's so cool. 拥有最好的模型同时又做产品，这是多大的优势啊。

[11:46]We are very lucky to be able to work with the Frontier models. 太酷了。

[11:49]Oh, my God. 我们很幸运能和前沿模型一起工作。

[11:51]What an awesome advantage, just like build a thing and then use it and 我的天。

[11:53]accelerate faster. It's so interesting. 多么棒的优势啊，就是构建一个东西然后使用它，

[11:55]There's a couple of like these other side 然后加速得更快。太有趣了。

[11:56]things I want to just kind of go on these like side quests on this conversation. 有几个其他的

[11:59]There's so much happening with Anthropic and I just I'm so curious to get your 我想在这段对话中走一些支线话题。

[12:02]insight. One is a week ago or so, the whole source code of cloud code leaked. Anthropic 发生了太多事情，我太好奇你的

[12:08]Somebody got it out there. 见解了。一个是大约一周前，Cloud Code 的整个源代码泄露了。

[12:09]I think it was a mistake someone made. 有人把它弄出去了。

[12:11]Is there anything you comment there? 我觉得是有人犯了个错误。

[12:12]Just like what happened? What went wrong? 你有什么可以说的吗？

[12:13]What should people know? 就是发生了什么？出了什么问题？

[12:14]So we immediately looked into this when we saw it. 大家应该知道什么？

[12:19]We realized that this was the result of human error. 我们看到后立即进行了调查。

[12:22]There is a human 我们发现这是人为错误导致的。

[12:23]working with cloud to write PR. 有一个人在

[12:25]This was just an update to how we release our packages. 用 Cloud 写 PR。

[12:29]And it actually went through two layers of human review. 这只是对我们发布包方式的一个更新。

[12:32]And so this was a result of human error. 而且它实际上经过了两个人工审核。

[12:35]And we've hardened our processes to make sure that it doesn't happen in the future. 所以这是人为错误的结果。

[12:39]This person's still at Anthropic. Are they doing all right? 我们已经加固了流程，确保以后不会再发生。

[12:42]Yes, yes. It's it's a process failure. 这个人还在 Anthropic 吗？他还好吗？

[12:44]And the most important thing is to just like learn from it and to add more 是的，是的。这是一个流程失误。

[12:48]safeguards so that doesn't happen again. 最重要的是从中学习，并添加更多

[12:50]And so that's that's what we've been focused on and most of those 保障措施，确保不会再发生。

[12:53]things have shipped. 所以这是我们一直在关注的，而且大部分的

[12:54]OK, another question I had is open claw. 措施已经实施了。

[12:58]So recently there's been this move to keep people from using cloud subscription 好的，我另一个问题是关于 OpenClaw。

[13:04]with their open clause, people got really upset. 最近有这个动向，阻止人们使用 Cloud 订阅

[13:08]They're confused why this is happening. 来配合他们的 OpenClaw，人们很不满。

[13:09]It feels like you're there's like, you know, harm cost to the open source community. 他们搞不清楚为什么会这样。

[13:14]What what are people what do people need 感觉像是你们在——你知道，对开源社区造成了伤害。

[13:16]to understand about kind of what went into this decision? 大家需要理解

[13:18]So we've been seeing a lot of demand for cloud and we've been working very hard 这个决定背后的考量是什么？

[13:23]to both scale our infrastructure and also to make our harness more token efficient 我们看到了对 Cloud 的大量需求，我们一直在非常努力地

[13:27]so that you can get more usage out of it. 扩展基础设施，同时让我们的系统更加节省 token，

[13:29]It wasn't designed for third party products, 这样你能获得更多的使用量。

[13:32]which have different usage patterns than our first party ones. 它不是为第三方产品设计的，

[13:36]We spent a bunch of time. 第三方产品与我们第一方产品的使用模式不同。

[13:39]Trying to figure out what is the most seamless transition that we can offer. 我们花了不少时间。

[13:43]And so I was very happy to be able to say 试图找出我们能提供的最无缝的过渡方案。

[13:47]that everyone gets some credits alongside their subscription. 所以我非常高兴能宣布

[13:50]But yeah, we did have to make the hard decision that 每个订阅用户都会获得一些 API 额度。

[13:53]we needed to prioritize our first products and our API. 但是是的，我们确实不得不做出艰难的决定，

[13:57]And so this is the this is the decision that resulted from that. 我们需要优先考虑我们的第一方产品和 API。

[14:00]Yeah, like to me, it makes so much sense. 所以这就是那个决定的结果。

[14:03]Like you guys are subsidizing this usage at like 200 bucks a month. 是的，对我来说，这太合理了。

[14:07]There's like it's like basically unlimited use of this. 你们基本上是在以每月 200 美元的价格补贴这些使用量。

[14:10]And like I think people don't understand this is they're trying to make money. 基本上就是无限使用这个。

[14:13]We're trying to be profitable. 我觉得大家不理解的是——他们是在赚钱的。

[14:15]We can't just like give away compute when it's so in demand. 我们在努力实现盈利。

[14:18]So I get it. 当计算资源如此紧缺的时候，我们不能就这么送出去。

[14:19]Coming back to the PM team, what is just like the PM team? 所以我理解。

[14:23]Like at Anthropic, how many PMs are there? 回到 PM 团队，PM 团队是什么样的？

[14:25]How are they kind of organized? 在 Anthropic，有多少 PM？

[14:26]Yeah, so we have a few PM teams. 他们是怎么组织的？

[14:29]I think we're maybe around 30 or 40 PMs right now. 是的，我们有几个 PM 团队。

[14:32]So we have the research PM team 我想我们目前大概有 30 到 40 个 PM。

[14:35]who Diane leads. 我们有研究 PM 团队，

[14:36]And this team is responsible for understanding all of the feedback from our 由 Diane 带领。

[14:41]customers for our models and then feeding that to the best research team to act on 这个团队负责理解来自我们

[14:45]it. And they also shepherd the model launch. 客户的所有关于模型的反馈，然后把反馈传递给最合适的研究团队去执行。

[14:48]There is the cloud developer platform team that maintains the APIs that 他们还负责管理模型发布。

[14:53]CloudCode is built on top of, and they also release things like managed 还有 Cloud 开发者平台团队，负责维护那些 API，

[14:58]agents, which is a way for you to build your agents and we can host it on your behalf. Cloud Code 就是建立在那些 API 之上的，他们还发布了类似 managed

[15:03]And then there's CloudCode that works on both CloudCode and the Cowork core products. agents 这样的功能，让你可以构建自己的 agent，我们可以代为托管。

[15:08]There's Enterprise that helps make CloudCode 然后是 Cloud Code 团队，同时负责 Cloud Code 和 CoWork 核心产品。

[15:11]and Cowork easier to adopt for all of our Enterprise customers. 还有企业级团队，帮助让 Cloud Code

[15:15]And so this is everything from like cost 和 CoWork 更容易被所有企业客户采用。

[15:17]controls, RBAC, security controls, and just making sure that these enterprises feel 这包括从成本

[15:23]very confident and comfortable using our tools. 控制、RBAC、安全控制，一直到确保这些企业对使用我们的工具感到

[15:27]And then we also have our growth team 非常放心和舒适。

[15:29]that is responsible for growing across our entire product suite. 然后我们还有增长团队，

[15:32]So we work very closely with them on CloudCode and Cowork growth. 负责推动我们整个产品组合的增长。

[15:36]And I know they also work with our other teams on CDP growth. 我们和他们密切合作，推动 Cloud Code 和 CoWork 的增长。

[15:40]So growth of people who use the Cloud API. 我知道他们还和其他团队合作推动 CDP 的增长。

[15:43]So speaking of growth, so Amol was just on the podcast. 也就是使用 Cloud API 的用户的增长。

[15:46]He had this really interesting insight that most people haven't been sharing. 说到增长，Amol 刚刚上了播客。

[15:49]There's always the sense that we need fewer PMs in the future. 他有一个很有趣的洞察，大多数人还没分享过。

[15:53]Why do we need PMs? Engineers can just ship. 总有一种感觉说我们未来需要更少的 PM。

[15:55]His take is that because engineers are moving so fast, 我们为什么需要 PM？工程师直接发布就行了。

[15:58]PMs and designers are squeezed, there's less time to stay on top of everything 他的观点是因为工程师推进得如此之快，

[16:02]that is happening, there's a feature shipping every day. PM 和设计师被压缩了，没有那么多时间跟上所有

[16:05]So his take is he needs more PMs because it's hard to keep up. 正在发生的事情，每天都有新功能发布。

[16:08]What's your take there? 所以他的观点是他需要更多的 PM，因为很难跟上节奏。

[16:09]Do you feel like there will be an increase in hiring of PMs? 你怎么看？

[16:12]What do you think is going on with the PM profession long term? 你觉得 PM 的招聘会增加吗？

[16:15]I think all of the roles are emerging. 你觉得 PM 这个职业长期来看会怎样？

[16:17]PMs are doing some engineering work, engineers are doing PM work, designers 我觉得所有角色都在演变融合。

[16:22]are doing PMing and also landing code. PM 在做一些工程工作，工程师在做 PM 工作，设计师

[16:25]You can either hire a lot more engineers 在做 PM 工作，同时也在提交代码。

[16:28]who have great product taste or you can keep your engineering hiring the same 你可以选择雇佣更多

[16:33]and hire a lot more PMs to help guide some of their work. 具有出色产品品味的工程师，或者保持工程师招聘不变，

[16:38]On our team, we're pretty focused on hiring engineers with great product taste. 然后雇佣更多 PM 来帮助指导他们的部分工作。

[16:43]This way we can reduce the amount of overhead for shipping any product. 在我们团队，我们很专注于招聘具有出色产品品味的工程师。

[16:48]Like there are many engineers on our team who are fully able 这样可以减少发布任何产品的开销。

[16:52]to end-to-end go from see user feedback on Twitter through to like ship a product 我们团队中有很多工程师完全能够

[16:57]at the end of the week with almost no product involvement. 端到端地完成从在 Twitter 上看到用户反馈到周末

[17:00]And this, I think, is actually like the most efficient way to ship something. 发布产品，几乎不需要 PM 参与。

[17:05]So I think like 我认为这实际上是最高效的发货方式。

[17:07]engineer and PM are kind of overlapping 所以我觉得

[17:10]and you will get a lot of benefit from having more of either. 工程师和 PM 的角色在某种程度上是重叠的，

[17:14]I think product taste is still a very rare skill to have and we'll pretty much hire anyone who we feel 增加任何一个都会带来很多好处。

[17:22]has demonstrated this strongly. 我认为产品品味仍然是一种非常稀缺的技能，我们基本上会雇佣任何我们认为

[17:25]And your background was in engineering, right? 强烈展现出这种能力的人。

[17:27]Yeah, I was an engineer for many years. 你的背景是工程对吧？

[17:29]I was then a VC very briefly before joining Anthropic. 是的，我做了很多年工程师。

[17:34]And actually, almost all the PMs on our team have either been engineers or ship 然后我很短暂地做过 VC，然后加入了 Anthropic。

[17:40]code here on Cloud Code, and so that's one of the things that I think helps build 实际上，我们团队上几乎所有的 PM 要么曾经是工程师，要么在

[17:45]trust with the team and also just enables us to move a lot faster. Cloud Code 上提交过代码，所以这是我认为有助于建立

[17:50]And then actually our designers also 团队信任的一件事，也让我们能够快得多。

[17:52]have been front end engineers before. 然后实际上我们的设计师

[17:54]Wow, because that's that's the big question. 之前也做过前端工程师。

[17:56]Like, there's definitely this merging that's happening. 哇，因为这是个大问题。

[17:58]The Venn diagrams are combining. 确实有这种融合在发生。

[18:00]I think the big question for a lot of people is if you're coming from engineering 维恩图在合并。

[18:03]or product or design, which of those core skills is going to be most valuable? 我觉得很多人关心的大问题是，如果你来自工程

[18:07]I could see at Anthropic and on Cloud Code, engineering is very valuable. 或者产品或设计，哪些核心技能会是最有价值的？

[18:10]I'm curious if other companies, 我在 Anthropic 和 Cloud Code 看到，工程背景非常宝贵。

[18:11]if you have a design background, becoming a PM is more valuable or just a PM PM. 我好奇在其他公司，

[18:15]I still think it comes back to product taste. 如果你有设计背景，做 PM 是不是更有价值，或者就是传统的 PM。

[18:19]Like as code becomes much cheaper to write, 我还是觉得归根结底是产品品味。

[18:22]the thing that becomes more valuable is deciding what to write. 随着写代码变得越来越廉价，

[18:25]Like, what is the right UX for this feature? 更有价值的是决定写什么代码。

[18:28]What is the most delightful way that a user can experience it? 比如，这个功能的正确 UX 是什么？

[18:31]What like we get tens of thousands 用户体验它的最令人愉悦的方式是什么？

[18:34]of GitHub issues asking for every single thing under the sun. 我们收到几万条

[18:39]And it takes a lot of. GitHub issue，什么要求都有。

[18:42]Care and taste to figure out, OK, 这需要大量的

[18:45]which of these is worth building and what is the right way to build it? 心思和品味来判断，好吧，

[18:48]And I think that that skill set can come from any background, but I think that's 哪些值得构建，正确的构建方式是什么？

[18:52]the most important thing. I think the reason why an engineering 我认为这个技能集可以来自任何背景，但我觉得这是

[18:55]background is particularly useful, at least for the next 最重要的东西。我认为工程背景之所以

[18:59]few months, is if you have an engineering background, 特别有用，至少在未来

[19:03]you have a better sense for how hard something should be. 几个月内，是因为如果你有工程背景，

[19:06]And that's often a factor in what you choose to build. 你对某件事应该有多难有更好的感觉。

[19:08]So like if something is very easy 这通常是决定你选择构建什么的因素。

[19:10]to build, then maybe instead of debating it, you just spend an hour doing it. 所以如果某件事非常容易

[19:14]But if something is harder to build and 构建，那与其争论，不如花一个小时就做了。

[19:16]you know that upfront, then you know that, OK, this will just cost a lot more. 但如果某件事更难构建，

[19:22]For our team to get this out the door. 而且你一开始就知道，那你就会知道，好吧，这会花费更多。

[19:24]So it helps a bit with the prioritization. 让我们的团队把这个推出来。

[19:27]You said in the next for the next few months. 所以在优先级排序上有一些帮助。

[19:30]Is that just like because the models will get so good potentially in the next few 你说在未来几个月内。

[19:34]months, you may not even need to know that as much. 这是不是因为模型在未来几

[19:37]I think the value skill sets does change quite frequently. 个月内可能会变得如此之好，你可能都不需要了解那么多工程知识了。

[19:41]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很频繁。

[19:45]So it's less a commentary on what shifts I think will happen and more of a commentary that I think large 所以很难预测超过几个月之后的事情。

[19:52]steps will happen. 所以这不太是我认为会发生什么变化的评论，更多的是我认为大的

[19:53]So you're not saying that's when Muthos comes out and will change everything. 变化会发生。

[19:56]And we don't need to know anything about engineering. 所以你不是说那就是 Mythos 出来的时候，会改变一切。

[19:59]No, I'm just saying that every every few 然后我们就不需要了解任何工程知识了。

[20:01]months, it seems like there's a yeah, there's a large increase in coding 不，我只是在说每隔

[20:06]capability, which then changes what other roles are valuable. 几个月，似乎都会有一次大的编码能力的

[20:10]I think 提升，然后其他角色的价值就会随之改变。

[20:11]the most important thing is to be able to. 我认为

[20:16]To to have this like first principles thinking where you can figure out how the 最重要的是能够

[20:22]tech landscape is changing, what the team really needs from you and to like jump in 拥有这种第一性原理思维，这样你就能弄清楚技术

[20:29]and fix that hole, because I think the work is becoming more amorphous, 格局是如何变化的，团队真正需要你做什么，然后

[20:35]which means that a great PM is able to understand what all the gaps are to figure 跳进去填补那个缺口，因为我认为工作变得越来越模糊，

[20:42]out what the highest priority ones are and then to just like figure out, OK, how do I 这意味着一个优秀的 PM 能够理解所有的差距，

[20:46]learn that skill set or what is like the skill set that I have that I can like apply to this challenge? 找出哪些是最高优先级的，然后想办法，好吧，我怎么

[20:52]So I think the current environment 学习那个技能，或者我已有的什么技能可以

[20:55]values people who are who are able to wear a lot of hats, are able to swap them 应用到这个挑战上？

[21:01]and are like very low ego about what work they do to help the team move faster. 所以我认为当前环境

[21:06]I love this answer. 重视那些能戴很多帽子的人，能随时切换帽子，

[21:07]There's this question I've been asking people in your in your shoes, folks that 而且对自己做什么工作来帮助团队加速非常低姿态。

[21:11]are kind of at the bleeding edge of what is capable of and building with the latest 我喜欢这个回答。

[21:14]tools, which is just like where will human brains continue to be useful and necessary 有一个问题我一直在问像你这样处于

[21:18]for a while until we get to super intelligence? 前沿的人，那些在用最新

[21:22]What I'm hearing here is essentially picking the things to work on, 工具构建的人，就是在达到超级智能之前，

[21:26]knowing where the market's going and figuring out where what to prioritize essentially. 人类的大脑在哪里会继续有用和必要？

[21:30]And then it's knowing if the thing you've built is good and right and getting it out 我听到的是本质上就是选择要做什么，

[21:35]there in some early version, at least. Does that sound right? 知道市场的走向，弄清楚优先做什么。

[21:37]Is there anything else of just like where 然后就是知道你构建的东西是不是好的、对的，并以某种早期版本

[21:39]human brains will continue to be useful for at least the next few months? 发布出去。这听起来对吗？

[21:43]I think humans still provide a level of common sense that the models don't. 还有什么人类大脑至少在未来几个月内

[21:49]And there's like a thousand moving 会继续有用的地方吗？

[21:52]pieces to any product launch. 我认为人类仍然提供了一种模型所不具备的常识水平。

[21:53]Some of them are very small, but there's always a lot that could potentially go wrong. 而且任何产品发布都有上千个

[21:59]I think the model doesn't always have a great sense of who all the stakeholders 移动的部件。

[22:04]are, how they relate to each other, what their preferences are, what are the right 有些非常小，但总是有很多可能出错的地方。

[22:07]venues to communicate with them, to keep them on board. 我认为模型不一定对所有利益相关者是谁、

[22:10]I think a lot of this like 他们之间怎么关联、他们的偏好是什么、什么是与他们沟通、

[22:12]more tacit, common sense, like EQ kind of knowledge is still very valuable. 让他们保持配合的正确渠道有很好的理解。

[22:18]Of course, we want the models to get better at this, and I think they will be. 我认为很多这种更加

[22:22]But right now, I think there's still gaps. 隐性的、常识性的、偏情商类的知识仍然非常有价值。

[22:24]How do you just kind of deal as a human 当然，我们希望模型在这些方面变得更好，我认为它们会的。

[22:26]going through so much constant change, just like just being on the inside 但目前，我认为仍然存在差距。

[22:30]of the tornado, maybe it's calm there, but just like how do you how do you stay 作为一个人，你怎么应对

[22:34]on top of what's going on, how you stay sane through all this craziness that we're 如此多的持续变化，就像在龙卷风

[22:38]moving through? I think our team is full of people who lean into the chaos. 的中心——也许里面是平静的，但就是你怎么跟上

[22:42]So we try to face every challenge with a smile because there's always so much going 正在发生的一切，你怎么在这种疯狂中保持理智？

[22:47]on, there's always so many risks and tricky situations. 我认为我们团队充满了拥抱混乱的人。

[22:51]That, you know, if you get too stressed about anything, you'll burn out. 所以我们试着笑着面对每一个挑战，因为总是有这么多事情

[22:55]And so we really look for people who 在发生，总是有这么多风险和棘手的情况。

[22:58]can kind of like look at a challenge, be like, oh, that's going to be hard, but I'm 如果你对任何事情太紧张的话，你会崩溃的。

[23:03]excited to tackle it and I'm going to do the best that I possibly can. 所以我们真的在寻找那些

[23:06]And I know I won't be perfect, but 能看着一个挑战说，哦，这会很难，但我很

[23:09]I'll be able to sleep at night knowing that I did my best. 兴奋能去应对，而且我会尽我所能做到最好。

[23:11]That's an interesting answer to just like what skills will be important in this 我知道我不会完美的，但

[23:16]future, because it's I forget who said this, maybe Ben Mann, 我晚上能睡得着觉，因为我知道我尽了最大努力。

[23:18]that this is the most normal this is the world will ever be. 这是一个有趣的关于未来需要什么技能的回答，因为——我忘了谁说的，可能是 Ben Mann，

[23:21]Yeah, it definitely gets harder. 说这是世界有史以来最正常的时候了。

[23:24]Like, I feel like there are a lot of weeks where maybe Sunday night there's some like 是的，肯定会变得更难。

[23:29]P0 and then by Monday there's like a P00 and by Monday afternoon there's a P000. 就是，我感觉有很多个星期，也许周日晚上有什么

[23:35]And you're like, wow, I can't believe I was so worried about that P0 from Sunday. P0 的事情，然后到了周一就变成了 P00，到了周一下午就变成了 P000。

[23:39]But I think you just have to acknowledge 你就会想，哇，我真不敢相信我周日还在为那个 P0 担心。

[23:42]that there's only so much that you can do that you need to sleep well so that you 但我觉得你只需要承认

[23:45]can make good decisions next day and just like brutally prioritize where you spend 你能做的只有那么多，你需要好好睡觉，这样

[23:50]your time, what's the most important 第二天才能做出好的决定，然后就是无情地优先排序你的

[23:51]thing to get right and be OK letting things go. 时间花在哪里，什么是最重要的

[23:54]Like there's there's products that we 事情要做对，并且接受放手一些东西。

[23:56]ship that aren't as polished as I wish they were. 就是，有些产品我们

[24:00]But. 发布的时候没有我希望的那么完善。

[24:01]You know, our top goal is to help empower professional developers. 但是。

[24:06]And if a product isn't successful, as long as it's not blocking the core use case, 你知道，我们的首要目标是帮助赋能专业开发者。

[24:12]it's OK because we'll hear the feedback and we'll fix it in the next release. 如果一个产品不成功，只要它不阻碍核心用例，

[24:17]Launching a feature that is buggy is the kind of thing that would have kept me up at night. 那就没关系，因为我们会听到反馈，然后在下一个版本中修复它。

[24:21]But it is something that I am now able to, like, live with knowing that, 发布一个有 bug 的功能，这种事情以前会让我彻夜难眠。

[24:27]OK, we're going to get that quick feedback and we're going to fix it in the next release. 但这现在是我能够接受的事情，因为我知道，

[24:31]What I'm imagining is there's that gif. 好吧，我们会得到快速的反馈，然后在下一个版本中修复它。

[24:33]I think it's maybe from Pirates of the 我脑海中浮现出那个 gif 图。

[24:34]Caribbean where it's this guy walking down a pair of stairs on a ship and the whole 我觉得可能是《加勒比海盗》里的

[24:38]ship is just being demolished around him and he's so chill just strolling down the 那个在一艘船的楼梯上走下来的人，整艘

[24:42]staircase as everything's falling apart. 船都在他周围被摧毁，但他却非常淡定地走下

[24:44]And that's interesting because everyone I've met from Anthropic is just so chill 楼梯，一切都在他周围崩塌。

[24:48]and just so like optimistic. 这很有趣，因为我遇到的每一个 Anthropic 的人都特别淡定

[24:50]Yeah. 而且特别乐观。

[24:51]I think that's a really interesting insight is just like having this calmness 是的。

[24:55]and optimism versus just like, oh, my God, everything's crazy and going nuts. 我觉得这是一个很有趣的洞察，就是拥有这种平静

[24:59]Yeah, I think if you don't have it, you'll get pretty burnt out. 和乐观，而不是——我的天啊，一切都疯了。

[25:03]I think we also tend to hire people who 是的，我觉得如果你没有这种心态，你会很快崩溃的。

[25:06]have been in the industry for a while and have experienced lots of ups and downs and 我觉得我们招聘的也往往是那些

[25:11]have a good sense for what gives them energy and how to 在行业里待了很长时间、经历过很多起起落落的人，

[25:16]maintain their energy over time, and I think that's helped us a lot. 他们对什么给自己能量有很好的感觉，

[25:20]So interesting. 知道怎么长期维持自己的能量，我觉得这对我们帮助很大。

[25:21]Something that I wanted to ask about is, 太有趣了。

[25:22]so there's these roles blurring, engineers are becoming PMs, everyone's dogs or cats, 有一个我想问的事情是，

[25:27]everyone's everyone. What do we lose in that world? 这些角色在模糊化，工程师变成了 PM，每个人又当爹又当妈，

[25:30]Do we lose like career ladders and clear career paths? 每个人都变成了所有人。我们会在这个世界中失去什么？

[25:33]Do we lose design consistency, code quality? 我们会失去职业阶梯和清晰的职业路径吗？

[25:36]You know, there's probably some downsides. 我们会失去设计一致性、代码质量吗？

[25:38]What are some things you find are just 可能有一些缺点。

[25:39]like, OK, that's something we're sacrificing for the greater good? 你觉得哪些事情是

[25:42]We're sacrificing product consistency. 好吧，那是我们为了更大的利益而牺牲的东西？

[25:45]Historically, when code was expensive to write, you would carefully plan out everything your products 我们在牺牲产品一致性。

[25:51]were going to use, the product suite, how every product relates to each other, 过去，当写代码很昂贵的时候，你会仔细规划好你的产品

[25:55]what the use case for every single one is, how they integrate. 要用的所有东西、产品组合、每个产品之间如何关联、

[25:59]And you would pretty much have one product for each use case. 每个产品的用例是什么、它们怎么集成。

[26:03]And now with AI moving so quickly and with so many ideas that we need to test out, 你基本上每个用例只有一个产品。

[26:10]we do sometimes have features that overlap with each other. 现在随着 AI 推进得如此之快，加上我们需要测试的想法如此之多，

[26:15]A lot of the times it's because there's two form factors that we love internally, 我们确实有时候会有功能互相重叠。

[26:19]and we want to, 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形式都很喜欢，

[26:21]What that means for someone 我们想让外部用户告诉我们哪个更好。

[26:24]who's a new user, though, is a new user might not know, OK, 这对新用户来说意味着，

[26:29]what is the best path to accomplish X? 新用户可能不知道，好吧，

[26:33]There is more education we need to do to help people understand what the core 完成某件事的最佳路径是什么？

[26:38]features are and what the best practices are for using them. 我们需要做更多的教育工作，帮助人们理解核心

[26:41]I think this is the this is the cost of launching a lot of features. 功能是什么，以及使用它们的最佳实践是什么。

[26:47]I think users also feel like it's hard to keep up with 我认为这就是发布大量功能的代价。

[26:51]the latest. 我觉得用户也觉得很难跟上

[26:52]Usually in traditional PM, you ship a feature every month or quarter, 最新动态。

[26:58]and so it's really easy for a user to to understand, OK, I just need to check in 通常在传统 PM 中，你每月或每季度发布一个功能，

[27:03]on this once a month and I'll learn some new things. 所以用户很容易理解，好吧，我只需要每月

[27:05]And if I ignore it for six months, it's fine. 来看一次，就能学到一些新东西。

[27:09]I don't feel like I'm missing out. 如果我忽略六个月，也没关系。

[27:10]I think with these agentic tools, not just called code and co-work, 不会觉得自己错过了什么。

[27:14]but like across the whole ecosystem, people feel this need to check Twitter every single day to see 我觉得对于这些智能体工具，不仅仅是 Cloud Code 和 CoWork，

[27:21]what the absolute latest thing is. 而是整个生态系统，人们觉得需要每天刷 Twitter 来看

[27:23]And I think there's more we can do to help people feel less like they're on this 绝对最新的东西是什么。

[27:30]ever increasingly fast treadmill and that they feel like I would love people to feel 我觉得我们还能做更多来帮助人们减少这种

[27:37]like they can just open these tools, the tools will educate them 越来越快的跑步机上的感觉，我希望人们能感觉到

[27:41]or like teach them what they want to know and that they can just feel more bought along. 他们可以直接打开这些工具，工具会教育他们，

[27:47]Yeah, I saw you launch this really interesting feature the other day. 或者教他们想知道的东西，他们能感觉到被带着一起走。

[27:50]I think it's slash. 是的，我看到你们前几天发布了一个很有趣的功能。

[27:51]Power up where it basically walks you through all the cool ways and basically 我觉得是 slash

[27:54]all the best practices to use cloud code is that kind of all in these lines? power up，它基本上带你了解所有很酷的用法，基本上

[27:57]Yeah, exactly. 就是使用 Cloud Code 的所有最佳实践，是这个方向的东西吗？

[27:58]So in the past, we didn't actually want to do something like power up because we 是的，没错。

[28:02]felt like the product should be intuitive enough that you can 过去，我们其实不想做像 power up 这样的功能，因为我们

[28:06]that you don't actually need to go through any tutorial. 觉得产品应该足够直观，你

[28:09]And over time, we've just realized that there's just so many features and there's 实际上不需要任何教程。

[28:14]so much demand for a built in onboarding experience that we we diverged a bit from 随着时间的推移，我们意识到有太多功能了，而且

[28:21]what was possible saying no, no onboarding flow. 对内置引导体验的需求如此之大，所以我们就偏离了

[28:23]And I did this because there's just so 之前说不做引导流程的理念。

[28:26]many users who wanted to know there's 100 features. 我这样做是因为有太多

[28:29]What are the 10 that I absolutely need to use? 用户想知道，有 100 个功能，

[28:31]And so we put that together. 哪 10 个是我必须用的？

[28:32]Yeah, it's such a bizarre world. 所以我们就把这个整合出来了。

[28:33]So Anthropic has been really successful with B2B enterprises where traditionally 是的，这是一个很奇异的世界。

[28:37]you don't launch a bunch of stuff, you just kind of have a quarterly release, Anthropic 在 B2B 企业市场非常成功，而传统上

[28:40]maybe, and it's like the opposite of every day we got some new. 你不会发布一堆东西，你基本上是每季度发布一次，

[28:43]So just maybe following that thread, 也许吧，而这恰恰相反，每天我们都有新东西。

[28:45]the run Anthropic has been on is just otherworldly. 所以顺着这个思路，

[28:49]Anthropic is way behind. Anthropic 的这波表现简直就是超凡脱俗的。

[28:50]And when it started, it was a mole share. Anthropic 曾经远远落后。

[28:53]This just like one of the least funded companies didn't have distribution. 刚开始的时候根本微不足道。

[28:56]Was it the first to go open? 就是那种最不受资金青睐的公司之一，没有分发渠道。

[28:58]I was way ahead. 它是第一个开放的吗？

[28:58]It was just like, no way. 远远领先。

[29:00]Anthropic has any chance to compete significantly long term. 就是觉得不可能。

[29:03]Now it's just killing it, just beating the biggest companies teams so much. Anthropic 长期来看有任何机会进行有力竞争。

[29:09]Just like the growth is just like $11 billion in ARR in one month. 现在它就是大杀特杀，击败了最大的公司和团队。

[29:15]Perps and growth. 增长简直了，一个月就 110 亿美元的 ARR。

[29:16]By the time this comes out, it'll probably be even higher. 利润和增长。

[29:20]I think on the inside, what what are some ingredients that have allowed 等到这期节目出来的时候，可能还会更高。

[29:24]Anthropic to be this successful and kind of come from behind and do this well? 我觉得从内部来看，哪些因素让

[29:29]The two most important things are one, this unifying mission. Anthropic 能够如此成功，从落后追赶到做到这么好？

[29:33]It's hard to state how important this is. 最重要的两件事，一是这个统一的使命。

[29:37]We hire people who care most about bringing safe AGI to all of humanity. 很难说这有多重要。

[29:44]And this is actually something that we reference frequently in our decisions 我们雇佣最关心为全人类带来安全 AGI 的人。

[29:50]that our entire product org should focus on shipping. 而且这实际上是我们在决策中经常参考的东西，

[29:54]And because we put this mission above any individual product line, we're able to And because we put this mission above any individual product line, we're able to

[30:00]very fast decisions that cut across the entire org and like execute on them in a unified way. 快速做出贯穿整个组织的决策，然后统一执行。

[30:06]So I think this is like something that I've never seen at a company of our scale. 所以我觉得这在同等规模的公司里是从未见过的。

[30:11]And so just to make sure that's clear. So essentially having the number one mission 所以为了确保这一点清楚明白。基本上，把第一使命

[30:15]is safety, alignment, making sure AI is good for the world. And you're saying just having that as 定为安全、对齐、确保 AI 对世界有益。你是说仅仅把它作为

[30:20]a clear mission makes decisions a lot easier to make. If there's two competing priorities, 一个清晰的使命就能让决策变得容易得多。如果有两个互相竞争的优先事项，

[30:26]we'll talk about which one is more important for Anthropic's mission. 我们会讨论哪个对 Anthropic 的使命更重要。

[30:29]And it makes it a lot easier to decide which of the two we prioritize. And then everyone 这样就更容易决定我们优先做哪个。然后所有人

[30:35]will stand behind the one that we decide. And so sometimes that means that like, hey, 都会支持我们做出的决定。所以有时候这意味着，比如说，

[30:41]we want to ship something on cloud code, but this other thing is more important. And so we 我们想在 Claude Code 上发布某个功能，但另一件事更重要。所以我们就

[30:44]deprioritize shipping this and we just wait until later. What's really interesting about that is 降低这个功能的优先级，等到以后再说。真正有趣的是

[30:48]that explains, I think, versus another company, maybe rhymes with Bopen AI, did a lot of different 这解释了，我认为，相比于另一家可能跟 Bopen AI 押韵的公司，做了很多不同的

[30:54]things. And I think that's a really interesting thing. And I think that's a really interesting 事情。我觉得这是非常有趣的一点。而且我认为这是一个非常有趣的

[30:56]What I'm hearing here essentially is like, okay, we're not going to launch a social network. We're 我在这里听到的基本上就是，好吧，我们不会去做社交网络。我们

[30:59]not going to launch a feed of interesting information because it's not aligned to this 不会去做什么有趣的信息流，因为这不符合我们的

[31:04]mission. And that has kept Anthropic focused, which just seems to be a core ingredient to the 使命。这让 Anthropic 保持了专注，而这似乎是

[31:09]success. Well, when I think about mission, I think about putting Anthropic's goals ahead of 成功的核心要素。说到使命，我觉得就是把 Anthropic 的目标放在

[31:15]any individual org or any individual product. And so for me, I think the second thing that we're 任何单个团队或单个产品之上。所以对我来说，我觉得我们要谈的第二件事

[31:26]going to talk about is mission. To me, it's slightly different. Mission means that teams 是使命。对我来说，使命的含义稍有不同。使命意味着各个团队

[31:31]are willing to make sacrifices that hurt their own goals and their own KRs in service of Anthropic's 愿意做出牺牲，哪怕损害自己的目标和 KR，也要服务于 Anthropic 的

[31:37]goals and Anthropic's KRs. And people are very happy to make those trade-offs. So like, 目标和 KR。而且大家非常乐意做这些权衡。比如说，

[31:44]an extreme example is if cloud code failed, but Anthropic succeeded, I would be extremely happy. 一个极端的例子是，如果 Claude Code 失败了，但 Anthropic 成功了，我会非常开心。

[31:51]And like, we're like, the whole team is very willing to make decisions, 整个团队都非常愿意做出这样的决定，

[31:56]that follow that chain of thought. I don't know if you can talk about this 遵循这样的思路。我不知道你是否能深入谈谈这个

[32:00]in depth, but do you feel like the open cloud decision is a part of this? Just like, okay, 但你觉得 Open Cloud 的决定是不是也是这个原因？就是觉得，

[32:04]this is not furthering the mission of Anthropic. We need to stop this because it's not working 这没有推进 Anthropic 的使命。我们需要停下来，因为它的效果

[32:09]in the way we want it to work. I think one of the most important things for Anthropic is to 没有达到我们的期望。我认为对 Anthropic 来说最重要的事情之一就是

[32:14]grow the number of users that we're able to reach. One of the ways that we're able to do this is with 扩大我们能触达的用户数量。实现这一目标的方式之一就是

[32:19]the cloud subscriptions with our first-party products. And so we just very much want to 通过我们第一方产品的 Claude 订阅。所以我们非常希望

[32:26]come at the expense of third-party products sometimes. So we've been talking about cloud 有时候以第三方产品为代价。所以我们一直在聊 Claude

[32:30]co-work, all these things, something that I want to make sure people get. And I'm curious just how Cowork 这些东西，我想确保大家能理解。我也很好奇

[32:34]you use these tools. So there's cloud code, there's cloud desktop/web, there's co-work. 你是怎么用这些工具的。有 Claude Code，有 Claude 桌面版/网页版，还有 Cowork。

[32:38]What's the best way to understand when to use which? When do you use each of these three? 最好的理解方式是什么？什么时候该用哪个？你分别什么时候用这三个？

[32:44]So I tend to use cloud code in the terminal when 所以我通常会在终端里使用 Claude Code，当我

[32:48]I'm just kicking off like a one-off coding task, and I want all of the latest features. The CLI is 只是启动一个一次性编码任务，而且想要所有最新功能的时候。CLI 是

[32:56]our initial product surface, and it's also the one where our features often land first. 我们最初的产品界面，也是功能通常最先落地的地方。

[33:01]And so it's the most powerful of all the tools. So that's what I tend to use when I'm just like 所以它是所有工具中最强大的。这就是我通常用的，当我只是想

[33:07]trying to kick off one or like maybe like a handful of tasks at a time. I think desktop 启动一两个或者几个任务的时候。我觉得桌面版

[33:13]really shines when you're doing something that requires front-end work. And so one thing that I 在做前端工作的时候特别出色。我

[33:18]love to do is to use our preview feature. So if I'm building a web app, I'll often use cloud code in desktop. 特别喜欢用的一个功能就是预览。如果我在做一个网页应用，我通常会使用桌面版的 Claude Code。

[33:26]I'll have the preview pane open on the right-hand side so that I can actually see the web app that 我会把预览面板打开在右边，这样我就能实时看到

[33:30]I'm making in real time as I'm chatting with cloud. It's also really great for people who 我正在和 Claude 聊天的同时，正在构建的网页应用。它也非常适合那些

[33:34]want something a bit more graphical. A terminal can feel very unfamiliar to someone who's non-technical. 想要更图形化界面的人。终端对非技术人员来说可能感觉很陌生。

[33:40]You get a bunch of these like scary pop-ups on your machine, and you can't click around 你的电脑上会出现一堆看起来很吓人的弹窗，而且你没法像在

[33:46]the way that you're used to in pretty much every other product that you use. So there's a lot of 其他几乎所有产品里那样随意点击。所以有很多

[33:50]people who just like don't feel comfortable in the terminal. And if that's you, I would highly 人就是在终端里感觉不舒服。如果你也是这样，我强烈

[33:55]recommend checking out cloud code on desktop. Desktop is also great for getting an at-a-glance 建议试试桌面版的 Claude Code。桌面版也非常适合快速一览

[34:01]view of everything that's happening. So you can see your CLI terminal sessions in desktop. You can 所有正在进行的任务。你可以在桌面版里看到你的 CLI 终端会话。你可以

[34:06]see your other desktop sessions. You can see your sessions that you kicked off on web and mobile. 看到你的其他桌面会话。你可以看到你在网页版和移动端启动的会话。

[34:11]So it's a one-stop control plane where you can see all of your tasks. I think the benefit of web and 所以它是一个一站式的控制台，你可以看到所有任务。我觉得网页版和

[34:18]mobile is that it's really great for kicking things off on the go. So CLI and desktop both 移动端的好处是它非常适合在外面随时启动任务。CLI 和桌面版都

[34:23]require you to be on your local laptop. And this is constraining because sometimes you're out and 需要你在本地笔记本电脑上操作。这就有局限，因为有时候你在外面

[34:29]about, you're like touching grass, you're going on a walk, and you don't have your laptop open. 比如说出去走走、散个步什么的，没带着笔记本电脑。

[34:33]I can't count the number of people who I've seen like holding their laptop open, 我数不清见过多少人，

[34:38]like tethered to their phone while they're outside. And this just means that we're 在外面的时候笔记本电脑开着，手机热点连着。这就说明我们

[34:42]missing a product that solves that need. And so for me, what mobile lets you do 缺少一个满足这个需求的产品。所以对我来说，移动端能让你

[34:48]is kick off these tasks on the go so that you don't need to bring your laptop everywhere 在外面也能启动这些任务，这样你就不需要到处带着笔记本电脑

[34:53]and make sure that your laptop's open wherever you are. 还得确保随时随地都开着笔记本电脑。

[34:56]I love that. I've seen people on plane, like it's just like such a meme now, 太有意思了。我见过有人在飞机上，这现在都成了一个梗了，

[35:00]just I need to finish, let this agent finish. I can't shut this down. 就是"我需要等它完成，让这个 agent 跑完。我不能关掉它。"

[35:02]Exactly. And then I think for cowork, the role that this fills is 确实如此。然后我觉得 Cowork 填补的角色是

[35:07]there's a lot of work that everyone does where the output isn't code. 每个人都有很多工作，产出不是代码。

[35:11]So whether that's like getting to Slack zero or inbox zero, or whether that's creating a slide 比如说把 Slack 消息处理完，或者把收件箱清空，或者做一叠

[35:17]deck for some customer meeting that's coming up, or whether that's writing a quick doc on what the 为即将到来的客户会议准备演示文稿，或者写一个简短的文档来说明

[35:23]goals of a feature are, or what the launch plan for a feature is, all these tasks produce outputs 某个功能的目标是什么，或者某个功能的上线计划，所有这些任务的产出

[35:29]that are non-code and cowork is best positioned for that. So the way that I split the products 都不是代码，而 Cowork 最适合做这些。所以我对这些产品的划分方式是

[35:35]in my mind is if I'm building something where the output is code, I'll use cloud code or desktop or 如果我构建的东西产出是代码，我就用 Claude Code 或桌面版，或者

[35:42]cloud code on mobile. And if the output is anything that's not code, I'll use cowork for 移动端的 Claude Code。如果产出不是代码，我就用 Cowork。

[35:47]it. People are just like 大家真的

[35:49]sleeping on the success that cowork is having. It's just like growing incredibly fast. And I 低估了 Cowork 取得的成功。它的增长速度惊人。我

[35:55]think people still don't understand maybe what it's for. And so what if you give us a couple of 觉得大家可能还是不太了解它是做什么的。所以你能不能给我们举几个

[36:00]use cases just in your work as a PM, what are some like really interesting, maybe unexpected ways to 你作为 PM 工作中的用例，有哪些非常有趣、可能出人意料的

[36:05]use cowork to save you time, get more work done? If you're getting started on cowork, the first 使用 Cowork 来节省时间、完成更多工作的方式？如果你刚开始使用 Cowork，第一件

[36:11]thing that you really need to do is connect all the data sources that are relevant to your role. 你真正需要做的事情就是连接所有与你角色相关的数据源。

[36:17]You can only do a great job if it has access to all the context that it needs to be able to curate 只有当它能获取到所有需要的上下文信息时，才能做好

[36:23]the output for you. So what that means for me is I connect it to my Google calendar, I connect it to 为你整理输出的工作。所以对我来说，就是把它连接到 Google 日历，连接到

[36:28]my Slack, to my Gmail, to my Google Drive, so that it just knows, it has the flexibility to Slack、Gmail、Google Drive，这样它就知道了，它可以灵活地

[36:34]find relevant context, to ask questions, to pull in threads, 找到相关的上下文，提出问题，拉取讨论线程，

[36:39]and this like substantially improves the quality of the result. The kinds of things I use it for are 这大大提升了输出结果的质量。我用它做的事情包括

[36:47]I use it for when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[36:51]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[36:55]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[36:59]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:03]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:06]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:09]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:12]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:17]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:22]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:26]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:31]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:35]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:40]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:42]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:46]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:50]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[37:54]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[37:59]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:03]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:08]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:12]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:15]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:18]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:21]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:23]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:25]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:28]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:32]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:35]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:39]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:41]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:43]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:44]working on a project, I use it when I'm working on a project, I use it when I'm working on a project, [重复内容]

[38:47]I use it when I'm working on a project, I use it when I'm working on a project, I use it when I'm [重复内容]

[38:49]- So slow. - 太慢了。

[38:50]- And I love people will see this deck - 而且我很喜欢，人们会看到这叠演示文稿

[38:53]whenever you present this, this will be out in the world. 不管你什么时候展示，它都会公之于众。

[38:55]Like obviously it's not the one-shotted version, 虽然显然不是一版定稿的，

[38:57]but you've iterated on it. 但你已经反复迭代过了。

[38:59]So just to help people try this for themselves. 所以就是为了帮助大家自己尝试一下。

[39:02]So step one is connect their, what did you say, Slack? 所以第一步是连接他们的，你怎么说的，Slack？

[39:05]What else do you suggest they connect? 你还建议连接什么？

[39:07]- Slack, Google Calendar, Gmail, G Drive. - Slack、Google 日历、Gmail、Google Drive。

[39:11]You should connect your communications tools 你应该连接你的通讯工具

[39:14]and where you store your source of truth data 还有你存储团队核心数据的地方，

[39:16]for what your team cares about, 你们团队关注的，

[39:19]what you care about and what you're working on. 你关注的，以及你正在做的事情。

[39:21]- Okay, and then what was the prompt roughly - 好的，那你大概输入了什么提示词

[39:23]that you put in there to generate this deck? 来生成这叠演示文稿？

[39:25]- So I just wrote, make me a slide deck - 我就写了，给我做一叠演示文稿

[39:28]for the Code with Claude conference. 用于 Code with Claude 大会。

[39:29]This is what our PMM suggested it should cover. 这是我们的 PMM 建议应该涵盖的内容。

[39:32]This is the current draft that I made that I don't like. 这是目前我自己做的草稿，我不太满意。

[39:35]This is one that I made manually that I don't like, 这是我自己手动做的一版，不太喜欢，

[39:37]but I linked it. 但我附上了链接。

[39:38]Can you start by creating a proposed outline with details? 你能先创建一个建议的大纲和详细内容吗？

[39:41]Also make sure it doesn't overlap too much 还要确保不要和

[39:43]with a keynote talk, which is more important. 主题演讲有太多重叠，因为那个更重要。

[39:45]And then Claude read a bunch of the links that I sent 然后 Claude 读了我发给它的

[39:48]to it and created a proposed outline. 一堆链接，创建了一个建议大纲。

[39:51]So then I read through its proposal 然后我仔细看了它的提案，

[39:54]and all the different ideas that it had generated 还有它生成的所有

[39:57]for what we could cover. 关于我们可以涵盖什么内容的想法。

[39:59]And I just made a decision on what I wanted to actually be 然后我就决定了哪些内容我想要

[40:02]in the final deck. 放进最终的演示文稿里。

[40:03]And I think this is like an example of what the role 我觉得这就是一个很好的例子，说明了

[40:05]of the PM still is today. PM 如今的角色是什么。

[40:07]It's like, Claude is a great brainstorming partner. 就是说，Claude 是一个很好的头脑风暴伙伴。

[40:11]It's able to synthesize a massive amount of information 它能非常快速地

[40:14]really quickly and present all of the possibilities 综合大量信息，

[40:18]to you. 把所有可能性

[40:19]But the role of the PM is still to make the end decision of, 呈现给你。

[40:22]OK, what should belong in the final product? 但 PM 的角色依然是做出最终决定：

[40:25]So for this, what I ended up deciding 好吧，最终产品里应该包含什么？

[40:27]was that I wanted the talk to cover the progression 所以对于这个，我最终决定

[40:32]from making local tasks successful 让这场演讲涵盖从

[40:34]to making every PR green to helping engineers 让本地任务成功，

[40:37]land more PRs. 到让每个 PR 都变绿，再到帮助工程师

[40:39]And for each of these, which demo 提交更多 PR 的发展过程。

[40:40]would be the most compelling? 对于每个阶段，哪个演示

[40:42]And then after this decision about the outline, 然后在决定了大纲之后，

[40:46]co-work just went off for a few hours. Cowork 就自己跑了几个小时。

[40:48]And built the whole side deck. 然后把整叠演示文稿都做好了。

[40:50]This is so awesome. 太棒了。

[40:50]What an awesome part of the job to not have to do anymore. 不用再做这部分工作真是太好了。

[40:55]And it feels like you're talking to essentially a deck designer 感觉就像你在跟一个演示文稿设计师说话，

[40:59]that also has actual knowledge about what you've worked on 而且他还对你做过的项目有实际的了解，

[41:02]and can make it actually the content which you want it to be, 能把内容做成你想要的样子，

[41:07]not just make it look really nice. 不只是让它看起来好看而已。

[41:10]How did you do the design system piece? 你是怎么处理设计系统那部分的？

[41:12]How does that work? 那是怎么运作的？

[41:12]How does it know the design system of Anthropic? 它怎么知道 Anthropic 的设计系统的？

[41:15]So what I did for this is we actually already 所以我的做法是，我们其实已经有了一套标准化的演示文稿模板，

[41:18]have a standardized deck that we use across all 在我们所有的外部活动中使用。

[41:22]of our external engagements. 所以我就把这个模板给了 Claude。

[41:23]And so I just gave Cloud access to that. 然后它就能看到我们用什么颜色、什么字体，

[41:26]And so it's able to see what colors we use, the fonts 以及各种不同的——

[41:28]we use, the different kinds of-- 怎么说来着？

[41:31]what's it called? 可用的幻灯片格式。

[41:33]Slide formats that are possible. 它有 20 张这样的示例幻灯片。

[41:35]And so it has 20 of these example slides. 举个例子吧。

[41:37]So give an example. 明白了。

[41:37]Got it. 所以你上传了——这是我们的模板，从这个开始做。

[41:38]So you upload, here's our template, work from this. 对。

[41:40]Yeah. 你也可以连接到你的 Figma MCP，

[41:41]You can also connect to your Figma MCP 如果你的演示文稿格式保存在那里的话，它可以拉取过来。

[41:43]if you have your side format saved there, and it can pull that. 你可以把那个导入进来。

[41:47]You can pull that in. 说到这个，我一直很好奇的是

[41:48]Along those lines, something I'm always curious about 你作为 Anthropic 的 PM，你的工具栈都有什么？

[41:51]is what's in your stack of tools as a PM in Anthropic? 当然有 Claude Code 和 Cowork 以及所有 Anthropic 的工具。

[41:55]Obviously, Cloud Code and Cowork and all the Anthropic tools. 还有什么？你还用什么？

[41:58]What else are you using? 其他的——Slack，你提到了。

[41:59]What other-- Slack, you mentioned. 还有别的吗？

[42:01]Is there anything else? 所以我的工具栈主要就是 Claude Code、Cowork 和 Slack。

[42:02]So my stack is pretty heavily Cloud Code, Cowork, and Slack. Anthropic 基本上是在 Slack 上运转的。

[42:07]Anthropic largely runs on Slack. 我觉得它是我们公司的核心操作系统。

[42:10]I feel like it's the core OS of our company. 日常来说，很多——

[42:13]And day to day, a lot of-- 我大概有 30% 的时间是在探索

[42:17]I would say maybe 30% of my time is pushing Cowork 和 Claude Code 的能力边界，

[42:23]the boundaries of what Cowork and Cloud Code can do 这样我就非常清楚我们哪些地方做得不够好。

[42:26]so that I have a very strong sense of what we're not good at. 而且我花很多时间和模型对话，

[42:32]And I spend a lot of time talking with the model 来理解为什么它会犯这些错误。

[42:36]to understand why it makes mistakes that it does. 我们其实做了很多内部工具。

[42:39]We actually have a lot of internal tools that we make. 我觉得 Claude Code 真正为我们整个公司

[42:43]I think one of the things that Cloud Code has really unlocked for our entire 解锁的一点是，它大大降低了

[42:47]company is it really lowers the barrier to making any custom app that you want. 制作任何你想要的自定义应用的门槛。

[42:53]And so we've seen this surge in personalized work software 所以我们看到个性化工作软件的激增，

[42:58]that people are building for custom use cases instead of using tools 大家都在为特定的使用场景构建定制工具，

[43:04]that don't perfectly fit the use case. 而不是使用那些不能完美匹配需求的现成工具。

[43:06]I got to hear more. 我得听听更多。

[43:08]What are some examples? 有什么例子？

[43:09]What are things you've built, other people have built, 你自己或者其他人做了什么，

[43:10]that are really popular and useful? 特别受欢迎和有用的？

[43:12]One of the sales folks on Cloud Code, he realized he Claude Code 的一个销售同事，他发现自己

[43:17]was making these repetitive decks over and over and over again. 在反复制作这些相同的演示文稿。

[43:20]And so he actually has this web app that he 所以他做了一个网页应用，

[43:23]built with the examples of the core Cloud Code decks 里面包含了 Claude Code 核心演示文稿的模板，

[43:27]that we know work well, so like a 101, 201, and Mastering Cloud Code. 我们知道效果很好的那些，比如 101、201 和精通 Claude Code 系列。

[43:32]And then he has a way to input specific customer contexts that 然后他有一个方式可以输入特定的客户上下文信息，

[43:35]pulls from Salesforce, that pulls from Gong, 从 Salesforce 拉取，从 Gong 拉取，

[43:37]that pulls from other nodes so that we can customize 从其他来源拉取，这样我们就能为特定客户

[43:40]the decks for specific customers. 定制演示文稿。

[43:42]And so we'll pull out things like, OK, this customer is using 所以它会提取信息，比如，好的，这个客户在使用

[43:46]like Bedrock or called for enterprise or console, Bedrock 或者 Claude for Enterprise 或者控制台，

[43:50]which affects what features are available to them. 这会影响他们能使用哪些功能。

[43:53]It will pull out things like, OK, this customer 它会提取信息，比如，好的，这个客户

[43:55]is concerned about like the code review stage of the SCLC. 关注 SDLC 中的代码审查阶段。

[43:59]And so we'll add a slide about our code review features there. 所以我们就会在那里加一页关于代码审查功能的幻灯片。

[44:04]It'll pull out things like, OK, this customer 它会提取信息，比如，好的，这个客户

[44:06]needs to be like HIPAA compliant or needs XYZ security controls. 需要符合 HIPAA 合规，或者需要 XYZ 安全控制。

[44:10]And so we'll make sure to add a slide or two 所以我们就会确保在他们的演示文稿里

[44:12]in their deck about that. 加一两页关于这个的内容。

[44:14]And then, for example, if-- 然后，例如如果——

[44:16]this is a customer that's on Vertex or Bedrock 这是一个使用 Vertex 或 Bedrock 的客户，

[44:20]and doesn't want to use Cloud for Enterprise, 不想使用 Claude for Enterprise，

[44:22]then we'll just take out some of the slides that are 那我们就会把一些只有 Claude for Enterprise

[44:25]Cloud for Enterprise only features. 才有功能的幻灯片拿掉。

[44:26]And so normally, this is like manual work 通常来说，这是需要手动做的

[44:28]that could take 20, 30 minutes. 可能要花 20、30 分钟的工作。

[44:32]And so people either like spend that time doing it, 所以人们要么花时间去做，

[44:35]or they'll just decide not to do it and use the general deck. 要么就直接决定不做了，用通用版本的演示文稿。

[44:38]With this, it takes like a few seconds, 用了这个，只需要几秒钟，

[44:40]and you get a tailored deck. 你就能得到一份定制的演示文稿。

[44:41]MARK MANDEL: What's interesting about this, like Slack MARK MANDEL：有趣的是，像 Slack 这样的工具

[44:43]is like the tool that nobody's-- it's 没有人——就是

[44:46]just like nobody's trying to create their own. 没有人想去做一个自己的替代品。

[44:48]Slack just continues to win. Slack 持续在赢。

[44:50]And it's just like the way you describe it 你描述它的方式

[44:51]is kind of the OS of so many companies. 就是很多公司的操作系统。

[44:53]It's so interesting. 太有意思了。

[44:54]Like people talk about Salesforce as just like SaaS. 大家谈论 Salesforce 就像是 SaaS 的代表，

[44:56]We don't need SaaS software anymore. 但我们不再需要 SaaS 软件了。

[44:58]We're going to build our own. 我们要自己做。

[44:59]It's like Slack is a durable tool 但 Slack 是一个持久的工具，

[45:01]that nobody wants to try to compete with and build 没有人想跟它竞争或者做一个

[45:03]a better version. 更好的版本。

[45:04]LILY FIERRO: I think it's pretty important communications LILY FIERRO：我觉得它是非常重要的通讯

[45:06]infrastructure. 基础设施。

[45:07]And I think they do the core task 而且我觉得它把核心任务做得非常好，

[45:09]of helping everyone get real-time updates incredibly 帮每个人获取实时更新这一点做得

[45:12]well. 极其出色。

[45:13]MARK MANDEL: Yeah, like people hate on Slack, MARK MANDEL：对，大家虽然吐槽 Slack，

[45:14]but it's really great at what it's 但它在它要做的事情上

[45:16]going to do. 确实做得很好。

[45:16]And like the most cutting edge teams are hooked on it. 而且最前沿的团队都离不开它。

[45:20]So interesting. 真有意思。

[45:21]LILY FIERRO: Yeah, and I also love LILY FIERRO：是的，我也很喜欢

[45:23]how easy they've made to customize it. 他们把定制化做得这么简单。

[45:25]And so we love making Slack bots. 所以我们很喜欢做 Slack 机器人。

[45:29]And this kind of like hackability 而且这种可 hack 的特性

[45:33]means that we're able to integrate 意味着我们可以按照自己想要的方式

[45:34]with Slack the way that we want to. 跟 Slack 集成。

[45:35]So really appreciate Slack's work on that. 真的非常感谢 Slack 在这方面的努力。

[45:38]MARK MANDEL: Time to buy some CRM stock. MARK MANDEL：是时候买点 CRM 股票了。

[45:40]I am so excited to tell you about this season's 我非常激动地向大家介绍本季的

[45:42]supporting sponsor, Vanta. 赞助商，Vanta。

[45:44]Vanta helps over-- Vanta 帮助超过——

[45:45]LILY FIERRO: --15,000 companies like Cursor, Ramp, Duolingo, 15,000 家公司，包括 Cursor、Ramp、Duolingo、

[45:50]Snowflake, and Atlassian earn and prove trust Snowflake 和 Atlassian，赢得并向客户

[45:53]with their customers. 证明信任。

[45:54]Teams are building and shipping products faster than ever 团队们正在以史无前例的速度

[45:57]thanks to AI. 构建和发布产品，这要归功于 AI。

[45:58]But as a result, the amount of risk 但结果是，引入到你的产品和业务中的

[46:00]being introduced into your product and your business 风险量

[46:02]is higher than it's ever been. 比以往任何时候都高。

[46:04]Every security leader that I talk to 我交谈过的每一位安全负责人

[46:06]is feeling the increasing weight of protecting 都感受到了保护其组织、业务

[46:09]their organization, their business, 日益加重的压力，

[46:11]and not to mention their customer data. 更不用说他们的客户数据了。

[46:13]Because things are moving so fast, 因为事情发展得太快了，

[46:15]they are constantly reacting, having to guess at priorities, 他们一直在被动应对，不得不猜测优先级，

[46:19]and having to make do with outdated solutions. 不得不用过时的解决方案将就。

[46:21]Vanta automates compliance and risk management Vanta 通过超过 35 个安全和隐私框架

[46:24]with over 35 security and privacy frameworks, 自动化合规和风险管理，

[46:27]including SOC 2, ISO 27001, and HIPAA. 包括 SOC 2、ISO 27001 和 HIPAA。

[46:31]This helps companies get compliant fast 这帮助公司快速获得合规认证

[46:33]and stay compliant. 并保持合规。

[46:35]More than ever before, trust has the power 信任比以往任何时候都更有力量

[46:37]to make or break your business. 决定你企业的成败。

[46:39]Learn more at vanta.com/lenny. 了解更多请访问 vanta.com/lenny。

[46:42]And as a listener of this podcast, you get $1,000 off Vanta. 作为本播客的听众，你可以享受 Vanta 1,000 美元的优惠。

[46:45]That's vanta.com/lenny. 那就是 vanta.com/lenny。

[46:48]OK, so you talked about all these different teams 好的，你谈到了所有这些不同的团队

[46:51]and how they use Cloud Code and Cowork to operate. 以及他们如何使用 Claude Code 和 Cowork 来工作。

[46:54]Which teams do you find other than engineering? 除了工程团队之外，还有哪些团队？

[46:56]I imagine engineering is the biggest token spender. 我想工程团队应该是最大的 token 消耗者。

[46:59]But if not, that'd be really interesting. 如果不是的话，那就很有意思了。

[47:00]What's the second place function right now for tokens? 目前 token 使用量第二的是哪个职能？

[47:04]Oh, Applied AI is amazing at pushing 哦，Applied AI 团队在探索

[47:07]the boundaries of what Cloud Code and Cowork can do. Claude Code 和 Cowork 的能力边界方面做得非常棒。

[47:11]A lot of our Applied AI team spends time with our customers. 我们 Applied AI 团队很多人花时间跟客户在一起。

[47:14]Helping them adopt our API. 帮助他们使用我们的 API。

[47:17]And so sometimes our Applied AI team will, for example, 所以有时候我们的 Applied AI 团队会，例如，

[47:19]make prototypes on behalf of these customers, 代表这些客户做原型，

[47:22]which Cloud Code makes so much faster than it used to be. 而 Claude Code 让这比以前快了很多。

[47:26]They also have the dual goal of needing to manage 他们还有另一个目标，需要管理

[47:30]a lot of customer comms, a lot of customer inbound, 大量的客户沟通、大量的客户反馈，

[47:34]and historical contacts, call notes. 以及历史联系记录、通话记录。

[47:37]And so they're both extremely heavy on Cowork and on Cloud 所以他们既大量使用 Cowork，也大量使用 Claude Code。

[47:40]Code. 代码。

[47:41]And just to understand Applied AI, does that work? 那 Applied AI 具体是做什么的？

[47:44]Does that forward deploy engineering sort of role? 那是不是类似前线的部署工程那种角色？

[47:48]How would most people describe what 大多数人会怎么描述

[47:50]the Applied AI team is doing? Applied AI 团队在做什么？

[47:51]Yeah. 是的。

[47:52]It's helping our customers adopt the latest API and model 就是帮助我们的客户在公司内部采用最新的 API 和模型功能，

[47:57]features across their company, both for powering 既用于驱动

[48:01]their company's products and also for internal acceleration. 他们公司的产品，也用于内部效率提升。

[48:05]Got it. 明白了。

[48:05]So it's like customer success, go-to-market-y, 所以有点像客户成功、市场推广，

[48:08]kind of like forward deploy engineering sort of thing. 类似于前线部署工程那种角色。

[48:10]Exactly. 没错。

[48:10]It's like a very technical go-to-market person. 就像一个非常技术型的市场人员。

[48:12]Got it. 明白了。

[48:13]OK, awesome. 好的，太棒了。

[48:14]So you're saying that might be the second org that 所以你是说他们可能是

[48:17]uses the most tokens. token 使用量第二的团队。

[48:19]Yeah. 对。

[48:20]And then we also see them pushing the boundaries 而且我们也看到他们在不断推进

[48:22]of what Cowork can do. Cowork 能做到的极限。

[48:23]So for example, a lot of these folks cover multiple customers. 比如，这些人很多都同时负责多个客户。

[48:29]And in any given day, can have like 5 to 10 customer 忙的时候一天可能有 5 到 10 个

[48:34]engagements on a high day. 客户会议。

[48:36]And so what they often use Cowork to do 所以他们经常用 Cowork 做的是，

[48:39]is the night before, they'll ask it to summarize, 前一天晚上，会让它总结一下，

[48:42]OK, what are all my customer meetings that are coming up? 好的，我明天有哪些客户会议？

[48:44]The next day? 后天呢？

[48:46]What are all the things that this customer has asked me for? 这个客户之前都跟我提过什么需求？

[48:50]What's top of mind for them? 他们最关心什么？

[48:51]What are the action items from the past meetings? 之前会议的行动项是什么？

[48:53]And Cowork will just put together this dossier, 然后 Cowork 就会把一份简报整理出来，

[48:57]this brief of what they should be aware of going 一份关于他们在进入下次会议前

[49:00]into the next meeting. 应该了解的信息汇总。

[49:01]And Cowork can also research answers. 而且 Cowork 还能帮忙研究答案。

[49:03]So if a customer asked, OK, when is feature X going to launch? 如果客户问了，好的，功能 X 什么时候上线？

[49:08]Cowork can help the PyDI person research through Slack Cowork 可以帮助这位 PyDI 成员在 Slack 里搜索

[49:11]to get the latest ETA, add that to the-- 获取最新的预计时间，添加到——

[49:13]add that to the notes so that during the customer call, 添加到笔记里，这样在客户通话的时候，

[49:17]the PyDI person has the absolute latest. 这位 PyDI 成员就有了绝对最新的信息。

[49:20]And these are just workflows that people 这些都是大家

[49:21]are building for themselves and sharing 自己构建的工作流，然后分享

[49:23]with other people on their team. 给团队里的其他人。

[49:25]MARK MANDEL: So cool. MARK MANDEL：太酷了。

[49:26]Something that-- kind of this question, this trend-- 有一个——这个问题，这个趋势——

[49:29]I don't know, question topic comes up a lot recently, 我不知道，这个话题最近经常被提到，

[49:31]which is tokens spend exceeding people's salary, 就是 token 花费超过了人们的工资，

[49:36]where people just use AI and it costs more 大家用 AI，结果费用比他们的工资

[49:39]than how much they're making. 还高。

[49:40]Are there any numbers floating around on topic of just how much 有没有什么数据关于这个话题，比如到底

[49:43]tokens spend, say, engineers spend, I don't know, 工程师花多少 token，比方说，

[49:47]a month, a day, or PMs, anything like that? 一个月、一天，或者 PM，之类的？

[49:50]JENNY GUY: It is clear to us that as the models get better, JENNY GUY：我们很清楚，随着模型变得更好，

[49:53]people delegate far more tasks to it, 人们会把更多的任务交给它，

[49:56]and they spend a lot more hours in tools 他们在 Claude Code 和 Cowork 这类工具上

[49:58]like Cloud Code and Cowork. 花的时间也越来越多。

[50:00]And so we do see the token cost per engineer 所以我们确实看到每个工程师

[50:04]or per any knowledge worker increase every time 或每个知识工作者的 token 成本

[50:08]that there is a model jump or a substantial product 在每次模型升级或重大产品改进后

[50:11]improvement. 都有所增加。

[50:12]JENNY GUY: I think it's-- JENNY GUY：我觉得——

[50:13]it's still much lower than what the average engineer salary is, 它仍然比工程师的平均工资低很多，

[50:18]but we see the percentage increasing over time. 但我们看到这个比例在随时间增长。

[50:21]MARK MANDEL: It's such an interesting-- MARK MANDEL：这真是太有趣了——

[50:23]we talked about how you have access 我们聊到了你们如何能使用

[50:24]to the most cutting edge models and other advantage 最前沿的模型，这是在 Anthropic 工作的另一个优势。

[50:26]of working in Anthropic. 在 Anthropic 工作的那种方式。

[50:28]I believe you guys have basically unlimited tokens. 我相信你们基本上有无限的 token 可以用。

[50:30]You don't-- you can use as much as you want, is that right? 你们——想用多少就用多少，对吧？

[50:32]JENNY GUY: We can use a lot of tokens. JENNY GUY：我们可以用很多 token。

[50:34]Some people do run into limits, so-- 有些人确实会遇到限额，所以——

[50:36]MARK MANDEL: OK, there's a limit. MARK MANDEL：好吧，原来有限额。

[50:37]OK. 好吧。

[50:38]Boris, shut it down. Boris，关掉它。

[50:41]OK, it's so interesting how many advantages 好的，能使用最先进的模型有这么多优势，

[50:43]come from having the most advanced model. 真是太有意思了。

[50:45]It's such an interesting, like, flywheel 这就形成了一个非常有趣的

[50:47]that starts to kick in. 飞轮效应。

[50:49]I think we also believe a lot in empowering 我们也非常相信要赋予

[50:52]our internal teams to build as fast as possible. 我们的内部团队尽可能快的构建能力。

[50:56]And we also trust that everyone understands 而且我们相信每个人都理解

[50:59]how much capacity that serving these models truly costs. 运行这些模型到底需要多少成本。

[51:04]And we trust our team to use the tokens responsibly. 而且我们相信团队会负责任地使用 token。

[51:07]So it's very frowned upon to waste tokens, 所以浪费 token 是很不受待见的，

[51:11]but we do trust individuals to make that change. 但我们确实信任每个人做出这个判断。

[51:12]And we trust our team to make that judgment call. 我们相信团队可以做出这个判断。

[51:15]MARK MANDEL: Awesome. MARK MANDEL：太棒了。

[51:16]Coming back to the PM role, we talked a little bit about this, 回到 PM 角色的话题，我们之前聊过一些，

[51:19]but I think this will be really interesting for people to hear. 但我觉得这对听众来说会非常有趣。

[51:23]Just what I want to understand is, what do you think 我想了解的是，你认为

[51:25]are the kind of the emerging skills PM 需要培养哪些新兴技能，

[51:27]that PMs need to develop slash you most look for, 或者说你最看重什么，

[51:32]AI companies most look for when they're hiring PMs these days? AI 公司如今在招聘 PM 时最看重什么？

[51:36]JENNY GUY: I think the hardest skill is being able to define JENNY GUY：我觉得最难的技能是能够定义

[51:42]what the product should look like a month from now. 产品一个月后应该是什么样子。

[51:45]I think there's a lot of ambiguity in what models are capable of 我觉得在那个时间范围内模型能做到什么，

[51:48]in that timeline and how user behavior will change. 以及用户行为会怎么变化，都有很多不确定性。

[51:52]But I think there are patterns that the best PMs can see 但我认为最优秀的 PM 能看到一些模式，

[51:56]based on how users are abusing the limits of the existing product. 基于用户如何"滥用"现有产品边界的模式。

[52:01]And the best PMs can sense that, can set a direction, 最好的 PM 能感知到这一点，能设定方向，

[52:05]and can steadily execute towards it and change 能稳步朝目标执行，如果模型能力比

[52:09]the path if the model capabilities are much better than or 预期好得多或差得多，还能调整路径。

[52:12]worse than what they had originally expected. 比他们最初预期的还要糟。

[52:15]I think it is very hard to be the right amount of AGI pilled. 我觉得很难把握好"AGI 信仰"的程度。

[52:19]So I think everyone can see this future where the models are 所以我觉得每个人都能看到那个模型极其聪明、

[52:25]extremely smart and can do almost everything, in which case 几乎什么都能做的未来，在这种情况下

[52:29]you actually don't need that complicated a product. 你其实不需要那么复杂的产品。

[52:31]You can actually just have a text box again 你其实只需要一个文本框，

[52:33]where you tell the model what you want. 告诉模型你想要什么。

[52:35]And it's so smart that it can add any tool 它聪明到可以自动添加任何工具

[52:39]or add any integration that it needs to get the job done. 或任何它需要的集成来完成任务。

[52:42]It knows when it's uncertain. 它知道什么时候不确定。

[52:44]It can ask clarifying questions. 它可以问澄清问题。

[52:45]It's very easy to build a product for the super AGI strong model. 为超级 AGI 强模型构建产品其实很容易。

[52:53]I think the hard thing is figuring out, for the current model, 我觉得难的是，针对当前的模型，

[52:58]how do you elicit the maximum capability? 你如何激发它的最大能力？

[53:01]How do you help users get onto the golden path? 你如何帮用户找到最佳路径？

[53:08]How do you guide users to interact with the model 你如何引导用户与模型的优势互动，

[53:11]strengths and patch its weaknesses? 同时弥补它的弱点？

[53:15]This skill is pretty rare. 这种技能非常稀缺。

[53:17]And how do you build that skill? 那你怎么培养这种技能？

[53:20]Is it just basically understanding the limits of each model? 基本上就是理解每个模型的限制吗？

[53:24]Are you talking about taste? 你说的是品味吗？

[53:26]Understanding, having taste into what the model maybe is capable of, 理解，对模型可能的能力有品味，

[53:29]what it's great and not great at, where it's changed? 知道它擅长什么、不擅长什么，哪里有变化？

[53:31]I think it's spending a ton of time talking and using the model. 我觉得就是花大量时间跟模型对话和使用模型。

[53:36]One of the things I really like to do is to ask the model to introspect on its own behaviors. 我特别喜欢做的一件事就是让模型反思自己的行为。

[53:41]So, sometimes when I notice that the model does something unexpected, 所以，有时候当我发现模型做了意想不到的事情，

[53:46]like, for example, there's situations where the model will make a front-end change 比如，有些情况下模型会做前端修改

[53:52]and run tests but not actually use the UI, 然后跑测试，但实际上并没有使用 UI，

[53:55]it's actually pretty useful to ask the model to reflect on why it did this. 这时候让模型反思一下为什么这么做其实很有用。

[54:01]And sometimes they'll say that, "Hey, there was something confusing in the system prompt," 有时候它们会说，"嘿，系统提示里有一些令人困惑的内容，"

[54:06]or, "I didn't realize that the front-end verification was part of this task," 或者，"我没意识到前端验证是这项任务的一部分，"

[54:11]or, "Hey, I delegated the verification to this sub-agent, 或者，"嘿，我把验证委托给了子 agent，

[54:14]and the sub-agent didn't do the test, and I didn't check its work." 但子 agent 没有做测试，我也没有检查它的工作。"

[54:19]A lot of times, just being very curious about why the model made the decision that it did 很多时候，对模型为什么做出那个决定保持好奇心，

[54:23]will show you what misled it so that you can fix the harness in order to close this gap. 就能让你看到是什么误导了它，这样你就可以修复引导机制来弥补这个差距。

[54:31]The other thing that helps is to figure out who are the users who you trust the most to give you accurate feedback 另一件有帮助的事是找出你最信任的那些用户，

[54:41]about the model. 让他们给你关于模型的准确反馈。

[54:42]Usually, there's a handful of people who are much better than others at articulating 通常有那么几个人比其他人更善于表达

[54:47]what makes a specific model or model-harness combination good. 是什么让某个特定的模型或模型+工具组合好用。

[54:51]And there's a lot of people who will give you feedback, but not everyone's feedback is as qualified. 很多人会给你反馈，但不是每个人的反馈都同样有价值。

[54:57]And so, finding a group of those five people you trust is really important for getting very fast feedback. 所以，找到你信任的那五个人，对于获取快速反馈非常重要。

[55:04]I think the third thing that is useful but not everyone loves doing is building evals. 我觉得第三件有用但不是所有人都喜欢做的事是构建 eval。

[55:10]You don't need to build hundreds of evals for them to be useful. 你不需要构建几百个 eval 才有用。

[55:15]Just building 10 great evals is important for helping the team quantify what the goal is, 只要构建 10 个好的 eval，就能帮助团队量化目标是什么，

[55:23]and what their progress towards it is, and what they're missing. 他们离目标有多远，以及还缺什么。

[55:26]And so, I think evals is this underappreciated thing that more PMs, more engineers should be working on. 所以我觉得 eval 是一种被低估的东西，更多的 PM 和工程师应该去做。

[55:33]We've covered evals a bunch. 我们之前聊过很多关于 eval 的话题。

[55:35]There's this trend of just like, "That is the future of product management is writing evals," because essentially, 有一种趋势就是，"产品管理的未来就是写 eval，"

[55:40]it makes us look like, "Okay, cool. 因为本质上，

[55:41]Let me actually concretely define it, and then we'll know." 它让我们看起来就像，"好的，很酷。

[55:44]How much of your time are you spending writing evals, would you say? 让我具体定义一下，然后我们就知道了。"

[55:46]I think the importance of evals varies a bit based on the feature that you're working on or what the problem you're trying to solve is. 你觉得你花多少时间在写 eval 上？

[55:55]So, there are a lot of folks on our team who do spend a lot of time working on evals. 我觉得 eval 的重要性取决于你正在做的功能

[56:00]We have a small pod of folks who collaborate very closely with research to more precisely understand our cod code behaviors 或者你试图解决的问题是什么。

[56:10]and what the largest areas of improvement are, and trying to measure those pretty concretely. 所以，我们团队有很多人确实花了很多时间在 eval 上。

[56:16]I personally jump into evals when there's a feature that I think needs a bit more product definition. 我们有一个小组跟研究团队紧密合作，更精确地了解我们 Claude Code 的行为，

[56:23]And often, the output of this is, "Okay, here are like five evals that I made. 以及最大的改进空间在哪里，并且努力非常具体地衡量这些。

[56:30]This is how you run them. 我个人会在某个功能需要更多产品定义的时候跳进来做 eval。

[56:31]These are the ones that succeed, and these are the ones that don't. 通常的产出是，"好的，这是我做的五个 eval。

[56:34]And this is like the prompt that I've used to increase the success rate." 这是运行方法。

[56:38]It varies a lot, though. 不过差异很大。

[56:40]Based on the exact feature. 取决于具体的功能。

[56:42]Not every feature needs it, but I think features such as memory benefit a lot from it. 不是每个功能都需要，但我觉得像记忆这样的功能从中获益很大。

[56:46]This point you made about people being very good at evaluating models is so interesting. 你说的关于人们非常擅长评估模型这一点太有趣了。

[56:50]It's almost like a human eval of just like, "Okay, they understand where it's spiking or it's maybe lacking." 几乎就像一个人肉 eval，就是"好的，他们知道哪里表现出色，哪里可能不足。"

[56:57]Is there anyone specific that you want to shout out that's very good at this? 有没有什么特定的人你想提一下，特别擅长这个的？

[57:00]Two people who I think are incredible at this are, one, Amanda, who molds Claude's character. 我觉得在这方面非常厉害的两个人，一个是 Amanda，她塑造了 Claude 的性格。

[57:09]It's just like such a hard role because the task is so ambiguous. 这个角色真的很难，因为任务太模糊了。

[57:16]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. 即使是编码都更容易，因为你可以验证成功与否，而塑造性格需要对 Claude 应该成为什么样的人有非常强的信念。

[57:28]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 successful. 我觉得她不仅有能力塑造性格，还能非常清晰地表达目标是什么，性格应该是什么样的，什么是成功的，什么是不成功的。

[57:35]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 successful. 我觉得她不仅有能力塑造性格，还能非常清晰地表达目标是什么，性格应该是什么样的，什么是成功的，什么是不成功的。

[57:36]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 successful. 我觉得她不仅有能力塑造性格，还能非常清晰地表达目标是什么，性格应该是什么样的，什么是成功的，什么是不成功的。

[57:37]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 successful. 我觉得她不仅有能力塑造性格，还能非常清晰地表达目标是什么，性格应该是什么样的，什么是成功的，什么是不成功的。

[57:38]And I think she has like an incredible ability to not only mold the character, but also to like articulate what the character, what's successful, and what's not successful. 我觉得她不仅有能力塑造性格，还能非常清晰地表达性格应该是什么样的，什么是成功的，什么是不成功的。

[57:40]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。

[57:46]So we often have team lunches and whenever there's a new model we're testing, 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?" 所以我们经常有团队午餐，每当有新模型在测试时，我们获取反馈最快的方式之一就是在这些午餐上，走到每个人面前问，"嘿，你对这个模型感觉怎么样？"

[57:58]And oftentimes we'll get feedback like, "Okay, this model is like not fully explaining its thinking. 通常我们会得到这样的反馈，"好的，这个模型好像没有充分解释它的思考过程。

[58:06]It's like too abrupt." 太唐突了。"

[58:07]Or like, "Hey, this model is like, just like loves writing a ton of memories, but like we're not sure if the memories are high quality or not." 或者，"嘿，这个模型就是喜欢写一堆记忆，但我们不确定这些记忆质量高不高。"

[58:16]Or like some people will notice that, okay, this model loves to test itself, which is great. 或者有人会注意到，好的，这个模型喜欢自己测试自己，这很好。

[58:22]Or like this model isn't testing itself enough. 或者这个模型自我测试不够。

[58:24]So that informs what data we look at to verify, okay, is this a larger pattern? 这告诉我们应该去看什么数据来验证，好的，这是不是一个更大的模式？

[58:29]So we have a ton of data, but it is very hard to extract insights. 我们有大量数据，但很难从中提取洞察。

[58:34]And so the feedback from this group, it's like, "Hey, this model loves to write a ton of memories, but like we're not sure if the memories are high quality or not." 所以这个群体的反馈就是，"嘿，这个模型喜欢写一堆记忆，但我们不确定这些记忆质量高不高。"

[58:35]So we have a ton of data, but it is very hard to extract insights. 我们有大量数据，但很难从中提取洞察。

[58:36]And so the feedback from this group is like, "Hey, this model loves to write a ton of memories, but like we're not sure if the memories are high quality or not." 所以这个群体的反馈就像，"嘿，这个模型喜欢写一堆记忆，但我们不确定这些记忆质量高不高。"

[58:37]So we have a ton of data, but it is very hard to extract insights. 我们有大量数据，但很难从中提取洞察。

[58:38]And so the feedback from this group helps us inform, okay, what are the hypotheses we want to test? 所以这个群体的反馈帮助我们明确，好的，我们要测试哪些假设？

[58:39]And then we're able to extract data to test that. 然后我们就能提取数据来验证。

[58:40]And then we're able to extract data to test that. [重复内容]

[58:41]And then we're able to extract data to test that. [重复内容]

[58:42]And then we're able to extract data to test that. [重复内容]

[58:43]And then we're able to extract data to test that. [重复内容]

[58:44]And then we're able to extract data to test that. [重复内容]

[58:45]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. 你提到的关于 Claude 性格这一点，我之前请过联合创始人 Ben Mann 来做播客，他谈到过这个，就是 Claude 的性格和特质是 Claude 非常重要的一部分。

[58:54]And then we're able to extract data to test that. [重复内容]

[58:55]And then we're able to extract data to test that. [重复内容]

[58:56]And then we're able to extract data to test that. [重复内容]

[58:57]And then we're able to extract data to test that. [重复内容]

[58:58]And then we're able to extract data to test that. [重复内容]

[58:59]And then we're able to extract data to test that. [重复内容]

[59:00]And then we're able to extract data to test that. [重复内容]

[59:01]And then we're able to extract data to test that. [重复内容]

[59:02]And then we're able to extract data to test that. [重复内容]

[59:03]And then we're able to extract data to test that. [重复内容]

[59:04]And then we're able to extract data to test that. [重复内容]

[59:05]And then we're able to extract data to test that. [重复内容]

[59:06]And then we're able to extract data to test that. [重复内容]

[59:07]And then we're able to extract data to test that. [重复内容]

[59:08]And then we're able to extract data to test that. [重复内容]

[59:09]And then we're able to extract data to test that. [重复内容]

[59:10]And then we're able to extract data to test that. [重复内容]

[59:11]And then we're able to extract data to test that. [重复内容]

[59:12]And then we're able to extract data to test that. [重复内容]

[59:13]And then we're able to extract data to test that. [重复内容]

[59:14]And then we're able to extract data to test that. [重复内容]

[59:15]And then we're able to extract data to test that. [重复内容]

[59:16]And then we're able to extract data to test that. [重复内容]

[59:17]And then we're able to extract data to test that. [重复内容]

[59:18]And then we're able to extract data to test that. [重复内容]

[59:19]And then we're able to extract data to test that. [重复内容]

[59:20]And then we're able to extract data to test that. [重复内容]

[59:21]And then we're able to extract data to test that. [重复内容]

[59:22]And then we're able to extract data to test that. [重复内容]

[59:23]And then we're able to extract data to test that. [重复内容]

[59:24]And then we're able to extract data to test that. [重复内容]

[59:25]And then we're able to extract data to test that. [重复内容]

[59:26]And then we're able to extract data to test that. [重复内容]

[59:27]And then we're able to extract data to test that. [重复内容]

[59:28]And then we're able to extract data to test that. [重复内容]

[59:29]And then we're able to extract data to test that. [重复内容]

[59:30]And then we're able to extract data to test that. [重复内容]

[59:31]And then we're able to extract data to test that. [重复内容]

[59:32]And then we're able to extract data to test that. [重复内容]

[59:33]And then we're able to extract data to test that. [重复内容]

[59:34]And then we're able to extract data to test that. [重复内容]

[59:35]And then we're able to extract data to test that. [重复内容]

[59:36]And then we're able to extract data to test that. [重复内容]

[59:37]And then we're able to extract data to test that. [重复内容]

[59:38]And then we're able to extract data to test that. [重复内容]

[59:39]And then we're able to extract data to test that. [重复内容]

[59:40]And then we're able to extract data to test that. [重复内容]

[59:41]And then we're able to extract data to test that. [重复内容]

[59:42]And then we're able to extract data to test that. [重复内容]

[59:43]And then we're able to extract data to test that. [重复内容]

[59:44]And then we're able to extract data to test that. [重复内容]

[59:45]And then we're able to extract data to test that. [重复内容]

[59:46]And then we're able to extract data to test that. [重复内容]

[59:47]And then we're able to extract data to test that. [重复内容]

[59:48]And then we're able to extract data to test that. [重复内容]

[59:49]And then we're able to extract data to test that. [重复内容]

[59:50]And then we're able to extract data to test that. [重复内容]

[59:51]And then we're able to extract data to test that. [重复内容]

[59:52]And then we're able to extract data to test that. [重复内容]

[59:53]And then we're able to extract data to test that. [重复内容]

[59:54]And then we're able to extract data to test that. [重复内容]

[59:55]And then we're able to extract data to test that. [重复内容]

[59:56]And then we're able to extract data to test that. [重复内容]

[59:57]And then we're able to extract data to test that. [重复内容]

[59:58]And then we're able to extract data to test that. [重复内容]

[59:59]And then we're able to extract data to test that. [重复内容]

[60:01]People really like that Claude's low ego. 大家真的很喜欢 Claude 的低调。

[60:03]And so if you tell it, Hey, you did this thing wrong. 所以如果你告诉它，嘿，你做错了这件事。

[60:05]It's like, truly sorry. 它会真心道歉。

[60:06]It's like, Oh shoot. 就像，哎呀。

[60:07]Like, thanks for telling me, like, let me fix it. 就像，谢谢你告诉我，让我来修一下。

[60:10]Let's work together. 我们一起解决。

[60:11]It's also very positive. 而且它也非常积极。

[60:12]So if you're feeling like, Oh, this is like an insurmountable task. 所以如果你觉得，哦，这任务太难了。

[60:16]I don't know how to get started. 我不知道怎么开始。

[60:19]Claude is like, okay, it's okay. Claude 就会说，好的，没关系。

[60:21]These, these are like the steps that I think we should take. 我觉得我们应该按这些步骤来做。

[60:24]Like, do you want me to get started on it for you? 要不要我先开始帮你做？

[60:28]I think part of what makes a great coworker is this positivity, this 我觉得一个好同事的特质就是这种积极性，

[60:32]like bias towards action, this, this ability to give you like earnest 这种行动导向，这种能给你真诚

[60:37]feedback, not just agreeing with every single thing that you say. 反馈的能力，而不是你说什么就同意什么。

[60:41]And so we try to imbue this into Claude because we think it makes it 所以我们努力把这种特质融入到 Claude 中，因为我们觉得

[60:44]a lot more enjoyable to work with. 这样会让跟它合作变得更加愉快。

[60:45]There's something I want to come back to. 有个话题我想回来聊聊。

[60:47]You talked about how, when new models come out, you often have to 你谈到每当新模型出来的时候，你经常需要

[60:50]kind of revisit things you've built. 重新审视之前构建的东西。

[60:52]That's so interesting. 太有意思了。

[60:53]And so like frustrating, maybe just like, Oh, goddammit, we ship this thing. 而且可能还有点沮丧，就像，天啊，我们已经发布了这个东西。

[60:56]Now I have to rethink it. 现在又得重新思考。

[60:57]Talk about just like how often. 聊聊大概多久会发生一次。

[60:58]You have to come back with a new model and they're like, okay, we have to redo 你需要回过头来处理新模型，然后说，好的，我们要重做

[61:01]this product that we launched a few months ago, a lot of the changes that 几个月前发布的产品。很多时候，

[61:06]we make with a new model is removing features that are no longer needed. 新模型带来的变化是移除不再需要的功能。

[61:11]So a lot of times we add features to the product as a crutch for the model, 很多时候我们给产品添加功能是为了弥补模型的不足，

[61:17]because it's not naturally doing itself. 因为它本身做不到。

[61:19]So the classic example for this is a to-do list. 典型的例子就是待办事项列表。

[61:22]When we first launched Claude code, people would ask it to do these 我们刚推出 Claude Code 的时候，人们会要求它做

[61:25]large refactors and Claude code would say, okay, cool. 大型重构，Claude Code 就会说，好的，没问题。

[61:28]I need a. 我需要——

[61:28]And then we would change these like 20 call sites and it would go and 然后它要修改 20 个调用点，

[61:31]change five of them and then stop. 结果只改了 5 个就停了。

[61:33]And then we were like, okay, how do we like force it to remember to 然后我们就在想，怎么才能

[61:37]get every single one of these 20? 让它记住要改完所有 20 个？

[61:39]And so Sid on our team was like, okay, what if we just like think 所以我们团队的 Sid 就说，好吧，

[61:43]about what a human would do? 我们想想人类会怎么做？

[61:44]A human would like make a list of everything that they need to change. 人类会先把所有需要修改的东西列一个清单。

[61:46]Similar to how in VS code, you would look up all the call sites and it'll be a 就像在 VS Code 里，你会查找所有调用点，

[61:50]list on the left side and you would like go through them one by one and replace all. 左边会显示一个列表，然后你一个个去替换。

[61:54]How do we give this kind of like a tool to Claude? 我们怎么给 Claude 这种工具呢？

[61:56]And so he added the to-do list. 所以他就添加了待办事项列表。

[61:58]And we found that with that, Claude was actually able to fix all these 20 call 我们发现有了这个之后，Claude 确实能够修改完所有 20 个

[62:02]sites, but then with Opus four and later models, we realized that we didn't need 调用点，但后来有了 Opus 4 和更新的模型，我们意识到

[62:07]to force it to use this to-do list. 不需要再强制它使用待办事项列表了。

[62:10]It would like naturally use it itself for the earlier models. 对早期的模型来说，它会自己用。

[62:13]We had to keep reminding it. 我们需要不断提醒它。

[62:14]Hey, did you finish everything on to-do list? 嘿，待办事项都做完了吗？

[62:16]You can't finish until you're done with everything on the to-do list. 你得把待办事项上所有的事都做完才行。

[62:18]And for the later models without prompting, it just like naturally thinks 而对于后来的模型，不需要提示，它就自然地

[62:22]to do everything on the to-do list these days, the to-do list is still 会把待办事项上的所有事情都做完，现在待办事项

[62:27]nice to have as like a user, um, because then you can more clearly see what 对用户来说还是有用的，因为你可以更清楚地看到 Claude

[62:32]Claude is working on, but honestly, it's such a de-emphasized part of the product 在做什么，但说实话，它现在已经是产品中

[62:36]right now that, um, the model may use it. 非常弱化的一部分了，模型可能会用，

[62:39]The model may not use it. 也可能会不用。

[62:40]It's like really not necessary for it to make thorough changes anymore. 它已经不再是做出彻底修改所必需的了。

[62:44]I forget who said this on the podcast, um, that the model 我忘了是谁在播客上说的，

[62:48]will eat your harness for breakfast. 模型会把你的脚手架当早餐吃掉。

[62:49]And what I'm hearing here is essentially you. 我在这里听到的基本上就是你——

[62:52]You remove things over time that you've had to add on top of the model where 你随着时间推移移除了那些不得不加在模型之上的东西，

[62:57]it was not operating the way you want it. 因为它之前没有按你想要的方式运作。

[62:59]And essentially as the models get smarter, you just, it becomes simpler 基本上随着模型变得更聪明，它就变得越来越简单，

[63:01]and simpler for it just to do the thing you want it to do. 就能直接做你想让它做的事了。

[63:04]Yeah. 是的。

[63:04]Um, we can move, remove a lot of prompting interventions every 每次模型变聪明的时候，我们可以移除很多提示干预。

[63:09]time the model gets smarter. 模型变得更聪明的时候。

[63:10]And we actually do this every time we launch a model, we read through the 其实每次发布模型我们都会做这件事，我们会通读

[63:13]entire system prompt and we reflect on, okay, for each of these sections, does 整个系统提示，然后反思，好的，对于每个部分，

[63:18]the model really need this reminder anymore? 模型还需要这个提醒吗？

[63:19]And if not, we'll remove it. 如果不需要了，我们就把它移除。

[63:21]The most exciting thing that new models unlocks though, is 不过新模型最令人兴奋的是

[63:24]just like entirely new features. 全新功能。

[63:26]So there's a lot of features that we've been testing out with prior models 有很多功能我们之前用旧模型测试过，

[63:31]and the accuracy wasn't high enough for us to want to launch them. 但准确率不够高，我们不想发布。

[63:34]And so one example of this is code review. 其中一个例子就是代码审查。

[63:37]We tried to build a code review product a few times, and we've launched like 我们尝试过几次构建代码审查产品，

[63:43]simpler versions of code review, which is the slash code review command in the past. 也发布过一些简单版本的代码审查，

[63:46]And it was only with the most recent models that we felt like, okay, 就是过去的斜杠代码审查命令。

[63:50]this code review is so good that our engineering team relies on this code review 这个代码审查太好了，以至于我们的工程团队

[63:56]to pass before we merge PRs. 在合并 PR 之前都要依赖代码审查通过。

[63:58]And we found that this was, we've always dreamed of Cloud being able to be a 我们发现，我们一直梦想着 Claude 能成为一个

[64:05]reliable code reviewer that can actually, that we can like confidently feel 可靠的代码审查者，能够——我们有信心它能

[64:09]catches the majority of bugs. 发现大部分的 bug。

[64:11]And it was only with like Opus 4.5 and 4.6 and Sonnet 4.6 that we felt like, 直到 Opus 4.5、4.6 和 Sonnet 4.6，我们才觉得，

[64:17]okay, we are now able to like run multiple code review agents simultaneously to traverse 好的，我们现在可以同时运行多个代码审查 agent，

[64:25]the entirety of the code base and to synthesize a set of like real issues 遍历整个代码库，然后综合出

[64:32]that an engineer needs to address before merge. 工程师在合并前需要处理的

[64:34]And so this is like a new capability that the, the newest models have unlocked. 一系列真实问题。

[64:39]This is another trend that is very common on this podcast of build something that 所以这就是最新的模型解锁的一项新能力。

[64:43]will possibly be possible in the next six months, be kind of at the edge of what's 这是本播客上非常常见的一个趋势：构建那些

[64:48]working sort of, and then it'll catch up and then it'll be an amazing product 在未来六个月内可能实现的东西，稍微处于

[64:52]and you'll be ahead of everyone. 可行性的边缘，然后模型会追上来，产品就会变得非常好，

[64:53]Yeah, exactly. 而你已经领先所有人了。

[64:54]Um, it's pretty important. 没错。

[64:55]You build products that don't necessarily work yet so that you know, okay, what is 构建那些暂时还不能用的产品很重要，

[65:01]missing, um, for this product to work. 这样你就知道，好的，这个产品要能用

[65:04]And then with the newest model, you can just swap it into the prototype you've 还缺什么。

[65:09]already made and see, okay, does this new model close that gap? 然后有了最新的模型，你可以直接把它

[65:12]How much are you able to speak to just kind of where things are going with 换进你已经做好的原型里，看看，

[65:15]Cloud and co-work as kind of the vision of it? 新模型是否填补了那个差距。

[65:18]I imagine you don't wanna give away too much about the goal, but it 你能在多大程度上谈谈 Claude 和 Cowork 的

[65:21]feels like you're, there's all these awesome features being added on top dispatch, 发展方向和愿景？

[65:25]control from phone and all these mobile app, all these things, what's kind of 我想你不想透露太多目标，

[65:29]just like a way to understand the vision for all these things long-term. 但感觉你——有所有这些很棒的功能在上面叠加，调度、

[65:32]We think about this in terms of building blocks. 手机控制，还有移动应用，所有这些，

[65:35]So for both cloud code and co-work, the core building block is 怎么理解这些产品长期的整体愿景？

[65:39]making individual tasks successful. 我们用构建模块的方式来思考这件事。

[65:42]So you, you want it to produce some output. 对于 Claude Code 和 Cowork，核心构建模块就是

[65:45]You give it a clear prompt description. 让单个任务成功。

[65:48]Is it able to consistently produce acceptable output that you're able to either merge or share with your colleagues 所以你想要它产生某个输出。

[65:54]or external audience? 你给它一个清晰的提示描述。

[65:55]So the task is the core building block. 它能持续产出你能接受的输出吗？你可以合并，或者分享给同事，

[65:57]As the models get smarter, the task success rate gets a lot higher. 或者给外部用户？

[66:01]And then we see people moving towards doing multiple tasks at the same time. 所以任务是核心构建模块。

[66:05]So multi-clouding was this big thing and towards the end of 2025, and it's only increased since then. 随着模型变得更聪明，任务成功率大幅提高。

[66:10]And so we see this as, okay, great. 然后我们看到人们开始同时做多个任务。

[66:13]One task works, and now you can do like six tasks at a time. 所以多 Claude 并行是 2025 年底的一大趋势，而且此后只增不减。

[66:17]As the models get even smarter, the way that we are extrapolating this is, okay, next, maybe you're 所以我们把这看作，好的，很好。

[66:24]going to run like 50 clouds at a time or hundreds of clouds at a time. 单个任务搞定了，现在你可以同时跑六个任务。

[66:28]And so what is the infrastructure we need to build to enable that? 随着模型变得更聪明，我们的推演是，好的，接下来，也许你会

[66:30]At that point, you're probably not going to run everything locally on your machine anymore. 同时运行 50 个 Claude，或者几百个 Claude。

[66:35]There's just like not enough RAM to do it. 那么我们需要构建什么基础设施来支持这个？

[66:38]And so we're, we're thinking about how do we make it easier for you to manage all these? 到那时，你可能不会在本地机器上运行所有东西了。

[66:45]These will probably run remotely. 就是没有足够的内存来做这件事。

[66:46]How do we build the interface so that you as a human know which tasks you need to look, look into, how do we 所以我们正在思考如何让你更轻松地管理所有这些？

[66:54]make sure that the agent is fully verifying its work so that when you look at a task and it says it's done, you like, can very 这些可能会远程运行。

[67:01]quickly verify and fully trust that it is done to your spec. 我们怎么构建界面，让你作为人类知道哪些任务需要查看，我们怎么

[67:04]And how do we make sure that this like process is self-improving so that when you do see a task that isn't done to your liking, you 确保 agent 完整地验证了它的成果，这样当你看到一个任务显示完成时，

[67:13]can give it feedback and the model will know for every future run to incorporate that feedback. 你能很快验证并且完全信任它达到了你的要求。

[67:17]So it never makes that mistake again. 我们怎么确保这个过程是自我改进的，这样当你确实看到

[67:19]So this is the progression that we're, we're bringing our users along for. 一个不符合你要求的任务时，你可以给它反馈，

[67:23]There's a lot of people. 模型会在未来的每次运行中 incorporate 那个反馈。

[67:24]People listening, a lot of product managers, a lot of maybe founders, a lot of other cross-functional folks listening. 这样它就再也不会犯同样的错误了。

[67:30]There's a lot of worry about just how their role, the future of their careers, what advice would you have for just people to not 这就是我们带用户一起前进的路径。

[67:40]just survive this transition to this very AI driven world, but to be really successful to essentially just to thrive in this future? 有很多——

[67:46]What are just like things people need to hear, need to be doing? 听众里有很多产品经理、很多创始人、很多其他跨职能的人。

[67:49]I think AI gives everybody a ton more leverage than they 很多人担心自己的角色、职业生涯的未来，

[67:53]used to. 你对这些人有什么建议，不只是

[67:54]And so I would push you towards anytime you realize that you're doing some manual task multiple times, think about how you can use 在这个 AI 驱动的世界里生存下来，而是真正成功，在这个未来中蓬勃发展？

[68:02]cloud code, cowork or other AI tools to automate that for you. 有什么大家需要听到、需要去做的？

[68:07]Most people have like creative parts of their job that they absolutely love. 我觉得 AI 给了每个人比以前

[68:12]And then like tedious parts of their job that they really hate doing. 多得多的杠杆。

[68:16]I think the beauty of AI is that it can do those tedious parts for you. 所以我建议你，每当你发现自己在反复做某个手动任务时，

[68:21]It can learn from every time that you've done that. 想想怎么用 Claude Code、Cowork 或其他 AI 工具

[68:23]It can do the manual task and generalize and then run it automatically and so that you can focus on the creative parts. 来自动化它。

[68:30]And that means you can do a lot more than you used to be able to do. 大多数人都有工作中

[68:33]So I think my like immediate push for people is figure out the repetitive parts that you can pass to cloud, iterate on those 非常喜欢的创造性部分。

[68:41]automations until the success rate is very high and then focus on, okay, what more can you be doing for your team, for your 然后还有一些非常讨厌的繁琐部分。

[68:48]product, for your company that like people haven't had the bandwidth to pick up so far? 我觉得 AI 的美妙之处就在于它可以帮你做那些繁琐的部分。

[68:53]Like, what is that like pet project that you always thought the company should do that like you've never had bandwidth to do? 它可以从你每次做的过程中学习。

[69:00]If AI can take care of the like grunt work, then you have you have this extra 20% time now that you might not have before. 它可以做手动任务，总结规律，然后自动运行，

[69:09]So 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. 这样你就可以专注于创造性部分。

[69:16]And then as a result, you'll be able to do so much more. 这意味着你能做到比以前多得多的事情。

[69:19]Something core to what you just shared, which I fully agree with, is find problems. 所以我对大家的直接建议是，找出可以交给 Claude 的重复性工作，

[69:23]Problems to solve with AI. 迭代这些自动化，直到成功率非常高，

[69:25]There's all this potential what all these tools can do. 然后专注于，好的，你还能为你的团队、你的产品、你的公司

[69:29]Some of the hard, like for a lot of people, the hardest part is just like, what should I actually do? 做什么，那些人们一直没精力去做的事情？

[69:32]And what you're saying here is just pay attention to things that you are doing constantly. 比如那个你一直觉得公司应该做的 pet project，但你从来没时间做？

[69:36]You can automate, pay attention to just like ideas that have been floating around that you haven't had time to do. 你可以自动化的，关注那些一直想去做但没时间的想法。

[69:41]It's basically it's like solve a problem for yourself is kind of the core advice there. 基本上就是为自己解决一个问题，这是核心建议。

[69:45]Exactly. 没错。

[69:46]I would also push listeners towards focusing on bringing your automations from, okay, this is a great tool. 我也想鼓励听众，把你的自动化从，好的，这是一个好工具，

[69:53]This is a cool concept to like, Hey, this actually works a hundred percent of the time. 这是一个很酷的概念，变成，嘿，这个真的百分之百能用了。

[69:56]Like sometimes I see users trying to automate something, getting it to like 90, 95% accuracy, and then giving up on it. 有时候我看到用户尝试自动化某件事，做到 90%、95% 的准确率，然后就放弃了。

[70:04]And this, if an automation doesn't work a hundred percent of the time, it's not really an automation. 如果一个自动化不能百分之百工作，它就不算是真正的自动化。

[70:11]And that last five to 10% does take more time. 而最后那 5% 到 10% 确实需要更多时间。

[70:15]Also building the automation is often a lot slower than you doing it yourself. 而且构建自动化往往比你亲自动手慢很多。

[70:20]I would encourage listeners to put. 我鼓励大家投入那个时间。

[70:23]Put in that time to scope some automation that you really want to get to a hundred percent, put in the elbow grease to teach quad your preferences, to like give it feedback so that it can improve its skill so that it can get to that a hundred percent. 投入时间去定义你真正想要做到百分之百的自动化，下功夫教 Claude 你的偏好，

[70:38]And then like really, then you'll be able to rely on it. 给它反馈，让它提高技能，直到达到百分之百。

[70:41]There there's just not much value in a 95% there automation. 然后你就真的能依赖它了。

[70:44]I am super guilty of that. 一个 95% 的自动化其实没多大价值。

[70:46]This is really good advice for me. 我太有这个毛病了。

[70:48]I am guilty of this too. 这个建议对我太有用了。

[70:50]I've been teaching it. 我也有这个毛病。

[70:51]I've been teaching co-work to. 我一直在教它。

[70:52]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 realize. 我一直在教 Cowork——

[71:02]Yeah, I funny enough. 试着帮我达到 Gmail 收件箱归零，但效果一直——非常耗时，而且肯定还没做到，你可能也发现了。

[71:03]That's exactly where my mind goes. 是的，有趣的是。

[71:05]I have this, uh, 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? 我想到的也是这个。

[71:14]Or what about the spot? 我有一个工作流，每封邮件进来后，它会找出那些垃圾性质的邮件，就是那些，"嘿，我能上你的播客吗？"

[71:15]Like all these things, I'm just like, I don't have time for these sorts of things and I have it categorized it into a folder called spammy. 或者"要不要做个广告？"

[71:21]And it's just like, it's. 之类的，我就是没时间处理这些，所以我让它把这些分到一个叫"垃圾邮件"的文件夹里。

[71:22]95% great. 而且它——

[71:23]But then there's like, oh, wow. 95% 做得很好。

[71:24]I missed an email cuz it went in there. 但有时候就会，哦，天哪。

[71:25]So this is a good push for me to like, I'm gonna work on this. 我错过了一封邮件，因为它被分到那里去了。

[71:28]I'm gonna get it to perfect. 所以这对我来说是一个很好的推动，我要把它做好。

[71:29]Yeah. 我要把它做到完美。

[71:29]We also are working on making the flow for customizing these commands a lot easier. 是的。

[71:35]Cuz right now I think you have to like know too many concepts. 我们也在努力让定制这些命令的流程变得更加简单。

[71:37]You have to know to define a skill. 因为现在我觉得你需要了解太多概念了。

[71:39]You have to know to like use this skill and give it feedback. 你需要知道怎么定义一个 skill。

[71:42]And then you have to know to tell co-work to update the skill based on all the feedback that you gave. 你需要知道怎么使用这个 skill，还要给它反馈。

[71:47]And then you also have to know where to read the skill to like make sure that the feedback was incorporated the way that you. 然后你还需要知道告诉 Cowork 基于你给的所有反馈来更新这个 skill。

[71:52]You want the, it's also our job to make this flow really seamless so that it doesn't feel painful to do. 然后你还需要知道去哪里读这个 skill，确保反馈被按照你的要求

[71:58]Amazing. 纳入了。这也是我们的工作，让这个流程变得非常顺畅，不会让人觉得很痛苦。

[71:58]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? 太棒了。

[72:05]We haven't already touched on before we get to our very exciting lightning round. 还有什么想分享的吗，Cat？还有什么想留给听众的？有什么想进一步强调的，

[72:08]I see a lot of people playing around with AI, um, and building like prototype apps and tinkering with building workflows. 在我们进入非常精彩的闪电问答之前？

[72:17]I would really push people towards building apps that you're actually using. 我看到很多人在玩 AI，构建一些原型应用，摆弄各种工作流。

[72:22]Every single day cuz I think only through that usage, are you actually getting the value? 我真的很建议大家去构建你真正在用的应用。

[72:26]Like if you build a prototype app that isn't helping you get more done, then the, the AI isn't really adding value to your, to your day. 每天在用的那种，因为我觉得只有通过这种使用，你才能真正获得价值。

[72:38]And there's only so much you learn from that when it's like, okay, I just did one shot at something. 如果你构建了一个原型应用但并不能帮你完成更多工作，那 AI 并没有真正为你的日常生活增加价值。

[72:42]Oh, that's cool. 你只做了一次就觉得，好的，我就试了一次。

[72:43]And then you never come back to it. 哦，挺酷的。

[72:44]It's like, you're not learning a lot and you're not getting like much leverage from it and actual leverage. 然后你再也不碰它了。

[72:48]Yeah. 你学不到多少东西，也得不到什么真正的杠杆。

[72:48]That's such a good point. 真正的杠杆。

[72:49]I also think there's a lot of people who spend a lot of time. 是的。

[72:52]Like customizing their workflow. 这个观点太好了。

[72:54]So there's like, I think there's like two ends of the spectrum. 我也觉得有很多人花了很多时间

[72:56]One is like people who never customize or never build automations, but there's like this polar opposite end of people who like obsessed around customizing their tool, like adding a ton of skills and MCPs and, um, these like workflow improvements. 在定制他们的工作流上。

[73:12]And I think sometimes that can even distract from your core goal of like launching some product or building some feature. 所以就像，我觉得有两端。一端是从来不定制、从来不构建自动化的人，

[73:18]I think there's a lot of fun in customizing and we. 但还有另一端的人痴迷于定制他们的工具，比如加一堆 skill 和 MCP，

[73:21]Definitely wanna make our products very hackable so that you, you can make it work really well for you, but there is a limit to how much it's useful. 还有这些工作流改进。

[73:30]Um, and I think there, 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. 我觉得有时候这甚至会分散你做核心目标的注意力，

[73:41]I see a lot of that on Twitter. 比如发布某个产品或构建某个功能。

[73:43]Just like, look at my setup. 定制化确实很有趣，而且我们

[73:45]It's out of control. 当然希望我们的产品非常可 hack，这样你可以让它很好地为你工作，

[73:46]It's so optimized. 但它的实用性是有限度的。

[73:47]And what do you, what are, what are you actually building? 我觉得有一类人可能花太多时间在定制上了，以至于没在睡觉，

[73:49]No, but my setup is so awesome. 也没在做他们最初想做的核心任务。

[73:51]I could get so much done. 我在 Twitter 上看到很多这样的。

[73:52]I think the simple setups actually work better. 就是那种，"看看我的设置。"

[73:55]Slash power up, get, take level up a little bit. "简直失控了。"

[73:59]Yeah. "优化到极致了。"

[73:59]Yeah. 然后你问，你到底在构建什么？

[73:59]There's this Karpathy tweet that just, uh, to came out on yesterday where he talked about this divide. "不是，但我的设置太棒了。"

[74:04]That's interesting between people that tried ChatGPT cloud back in the day. "我能做超多事。"

[74:10]It was like, okay. 我觉得简单的设置实际上效果更好。

[74:11]And they're like, nah, this is this terrible. 就是稍微提升一下就好。

[74:13]And they kind of gave up on like what AI could do for them. 是的。

[74:15]And they just like, so cynical of like, no way. 是的。

[74:17]It's not actually that big of a deal. Karpathy 昨天发了一条推文，他谈到了这种分化。

[74:18]And then there's people that are using it to code essentially. 很有意思，就是当年试过 ChatGPT/Claude 的人，

[74:21]Yeah. 觉得，好的。

[74:21]Who see the full intense power of it and how good it is. 然后说，"不，这太差了。"

[74:26]And people on both sides don't understand the other side and why they like how much they, how they see the world. 两边的人都不理解对方，不理解对方为什么那样看待世界。

[74:32]And so your advice is really good here. 所以你的建议在这里真的很好。

[74:34]Just like actually use it for real things and see how good it actually has gotten. 就是真正去用它做实际的事情，看看它到底变得多好了。

[74:38]Yeah. 是的。

[74:39]I think the big shift is that the 2024 generation of products were chat based and the cloud code generation of products is action-based and the. 我觉得最大的转变是，2024 年那一代产品是基于对话的，而 Claude Code 这一代产品是基于行动的。

[74:51]So like big aha moment people have is when cloud can just like do things on your behalf. 所以大家的大 aha 时刻就是当 Claude 可以替你做事的时候。

[74:56]It is, it is an amazing feeling to know that the agent is capable of doing so much more than telling you what to do. 那种感觉非常奇妙，知道 agent 能做到的远不止是告诉你该怎么做。

[75:04]Like the agent can actually just do it itself. agent 可以真的自己去做。

[75:06]And when people feel that, I, I think that's the eye opening moment. 当人们感受到这一点时，我觉得那就是豁然开朗的时刻。

[75:10]Shout out, uh, Chrome extension, the cloud called Chrome extension, which you could just watch it doing stuff that you'd be like, fill out this form for me. 顺便提一下 Chrome 扩展，Claude 的 Chrome 扩展，你可以看着它在做事，比如你让它帮你填个表格。

[75:17]And I'm like, all right, here I go. 它就会说，好的，我开始了。

[75:18]Exactly. 没错。

[75:19]Okay. 好的。

[75:20]Uh, anything else before we get. 在我们进入

[75:21]A very exciting lightning round. 非常精彩的闪电问答之前，还有什么吗？

[75:22]No, let's do it. 没有了，开始吧。

[75:24]Let's do it. 开始吧。

[75:25]Uh, Kat, I've got five questions for you. 好的，Cat，我有五个问题要问你。

[75:28]Welcome to the lightning round. 欢迎来到闪电问答。

[75:29]There's this animation that place I have to make sure to say it. 有个动画——我得说一下。

[75:31]Uh, are you ready? 你准备好了吗？

[75:32]I'm ready. 我准备好了。

[75:33]First question. 第一个问题。

[75:35]What are two or three books that you find yourself recommending most other people? 你经常推荐给别人的两三本书是什么？

[75:38]I really like how Asia works. 我很喜欢《How Asia Works》。

[75:41]Um, it's a story about economic development and what are the, like the policies and, uh, governments that make, uh, long. 它是关于经济发展的故事，以及什么样的政策、政府能创造长期的——

[75:51]That. 那个——

[75:51]lasting, successful economies. The other books that I'm really into are The Technology Trap. 持久的、成功的经济体。我喜欢的另一本书是《The Technology Trap》。

[75:57]So this is actually about the past few technology revolutions, so the industrial revolution and the 它其实是关于过去几次技术革命的，工业革命和

[76:03]computer revolution, and how this has affected workers. The reason that I really like this is 计算机革命，以及这些如何影响了工人。我喜欢它的原因是

[76:10]because I think there's a lot we can learn from history to make sure that this transition goes 我觉得我们可以从历史中学到很多，来确保这次的转型

[76:15]well. And maybe on like a fun note, I really like Paper Menagerie. It's just like a book of short 进展顺利。也许轻松一点的推荐，我很喜欢《Paper Menagerie》。它就是一本

[76:22]stories about like coming of age and AI and just like self-discovery. Favorite recent movie or TV 短篇故事集，关于成长、AI 和自我发现。最喜欢的近期电影或电视

[76:32]show you have really enjoyed? I really like Drive to Survive. There's no like deeper meaning to it. 节目？我很喜欢《Drive to Survive》。没有什么深层含义。

[76:39]I just, there's just something very satisfying about people being so, 我就是觉得那些人对一个单一的工程目标

[76:45]so obsessed with like a singular engineering goal and just like the purity of their pursuits. 如此痴迷，追求的纯粹性，让人非常满足。

[76:52]And I also really love Free Solo, which is about Alex Honnold climbing El Capitan without a harness. 我还很喜欢《Free Solo》，关于 Alex Honnold 不带安全绳攀登 El Capitan 的故事。

[77:00]And I think similarly, it's just such a pure achievement to be able to climb this extremely 我觉得类似地，能够攀爬这条极其

[77:08]challenging, dangerous route and to be able to have the mental focus to do it, 挑战性、危险的路线，还能保持那种精神专注力，

[77:15]knowing that if you make a single mistake, you die. 知道犯一个错误就会死——这太纯粹了。

[77:17]It's insane. Yeah, that movie is out of control. And it's interesting how these 太疯狂了。是的，那部电影真的太震撼了。有趣的是这些

[77:20]relate in some way to the work you do. I actually am a rock climber. 在某种程度上跟你做的工作也有关联。我其实是个攀岩爱好者。

[77:23]I first watched Free Solo before I climbed rocks. And so I thought it was impressive, 我在攀岩之前就看了《Free Solo》。所以当时觉得很厉害，

[77:29]but I didn't understand how impressive it was. It's one of the rare movies where like, 但不知道到底有多厉害。这是那种罕见的电影，你越了解，

[77:32]the more you know about it, the more you're, you're blown away by how insane this is. Like 就越被震撼到。像他在墙上做的那些动作，

[77:37]the kinds, the kinds of movies he's doing on the wall are things that like, I don't think I will 我觉得我这辈子在攀岩馆里、离地一英尺，

[77:42]ever be able to do in my lifetime if we're set in a gym, like one feet off the ground. 都做不到。还系着绳子。系着绳子。

[77:47]With a rope. With a rope. 你看了那个关于另一个更年轻的攀冰者的纪录片吗？

[77:49]Did you see the documentary on that other guy, the younger one that went on like ice mountains? 我看了。那个很令人难过。

[77:54]I did. That one was very sad. 但那个真的很疯狂。好的。你最近发现并非常喜欢的

[77:56]But that was, that was wild. Okay. Favorite product you recently discovered that you really 产品？

[78:00]love? 除了 Claude 相关产品之外，对我的生活改变最大的产品

[78:01]The product that is like most changed my life outside of 可能就是 Waymo 了。我是一个铁杆 Waymo 用户。每天用两次，上下班通勤。

[78:05]cloud products is probably Waymo. Like I'm a diehard Waymo user. Use it twice a day, get to 是的。

[78:12]and from work. 我喜欢它的两点，一是如果 Waymo 在等我，我不会觉得不好意思。

[78:12]Yeah. 所以我觉得不用那么有压力非得在它到达的那一刻就站在路边。

[78:12]So the two things that I really like about it are one, I don't feel bad if a Waymo is waiting 第二是我觉得它让我更高效了。当我和另一个人

[78:18]for me. And so I feel like I feel less pressure to be right at the curbside the moment it arrives. 在车里时，我通常不会打工作电话。我觉得如果在车上

[78:25]And the second thing is I feel like it lets me be a bit more productive. When, when I'm in the car 一直用笔记本电脑有点不礼貌。但 Waymo 的好处是我可以

[78:32]with another human, I, I typically try not to like do any work calls. I, I feel a little rude if I'm 接工作电话。不用担心有人偷听。不用担心，

[78:37]like on my laptop the whole time. But one thing I really appreciate about the Waymo is I can 嘿，这样会不会很没礼貌？我是不是说话太大声了？要不要让人家

[78:42]call into a work call. I'm not worried about someone overhearing me. I'm not worried about, 换个音乐？所以这真的——我觉得每天帮我省了 30 分钟。

[78:47]hey, is this like rude? Am I talking too loud? Do I need to tell, ask someone to like change 技术的这些二阶效应。太有意思了。

[78:51]the music? And so this has been like, I feel like this has given me back like 30 minutes every day. 是的。我一直以为 Waymo 需要定价比 Uber 和 Lyft 低

[78:55]All these second order effects of, of technology. It's so interesting. 才能成功，但实际上我非常乐意付两倍的溢价。

[78:59]Yeah. I always thought Waymo needed to be priced 我爱 Waymo。就是——你一旦体验过，你就会觉得，啊，这太——太

[79:01]lower than Uber and Lyft to succeed, but actually I'm like very happy to pay a 2X premium for it. 疯狂了。然后你就习惯了。你坐进去的时候会觉得，

[79:07]I love Waymo. It's just like, like you, once you see it, you're just like, ah, this is in, in, 太不可思议了。然后你就忘了这种感觉了。完全正确。而且我觉得它也改变了

[79:12]insane. And, and then you get used to it. Like you get in there and you're like, 语言习惯。Anthropic 很多人喜欢 Waymo。过去你可能说，

[79:15]this is crazy. And then you forget about it. Totally. And I think it's also changed the 嘿，叫个某某打车软件。现在大家就直接说，好的，

[79:19]vernacular. Like a lot of people at Anthropic love Waymo. And I think in the past you'd be like, Waymo 到了吗？好的。还有两个问题。

[79:25]hey, like let's call like blah blah rideshare app. And now like everyone's just like, okay, 你有最喜欢的工作或生活中的座右铭吗？

[79:29]is Waymo here? Okay. Two more questions. 去做就好。我觉得第一性原理思考非常有价值。

[79:31]Do you have a favorite like motto that you often come back to in work or in life? 如果你知道你在优化什么，而且你有很强的

[79:35]Just do things. I think there's a lot of value in like first principles thinking. And 第一性原理，那你通常能推导出正确的行动方案，

[79:42]if, if you like, if you know what you're optimizing for and you have like strong first 并且能向所有利益相关者清楚地表达。

[79:46]principles, then you can normally deduce what the right, like course of action is and be able to 然后你就应该去做。我觉得工作头衔是假的。

[79:51]clearly articulate that to all the stakeholders. And then you should just like do it. Like I think 如果你理解了约束条件，你就能弄清楚你能做什么，

[79:56]jobs are fake. If you understand the constraints, you can figure out what you can do and then just 然后就去做，快速学习，如果做错了就道歉或修正。

[80:02]like try to do it quickly, learn from the mistakes and apologize or fix them if you did something 比如快速去做、从错误里学习，做错了就道歉或者修复

[80:07]wrong. You could just do things. Whoever said that. I think it's liberating actually. 做错了。你就可以去做事情。谁说的来着。我觉得这其实是一种解放。

[80:12]To like tell people this, I think a lot of companies like roles are very strictly defined, 告诉大家这个，我觉得很多公司的角色定义非常严格，

[80:18]like, okay, this is what the PM does is what the designer does is what engineer does. And then 好的，PM 做这个，设计师做这个，工程师做这个。然后

[80:22]even team scopes are very rigidly defined. So, hey, like this corner of the code base we touch 连团队范围都定义得很死板。所以，嘿，代码库的这个角落我们碰，

[80:27]and this corner, like we're not allowed to touch. And I think what just do things lets people do is 那个角落我们不能碰。我觉得"去做就好"让大家

[80:32]they feel like empowered to make these decisions, empowered to operate across team boundaries, 觉得自己有权力做这些决定，有权力跨团队操作，

[80:37]just to like get something done. That feels like a big, important skill. 就是为了把事情做好。这感觉是一项非常重要的技能。

[80:41]To be good at, people call it agency. Just like do the things that need to be done. 要擅长的，人们称之为"主动性"。就是去做需要做的事。

[80:46]Bias towards action. 行动导向。

[80:47]Bias towards action. All these ways of describing just like, don't wait for permission. 行动导向。所有这些描述方式都是在说，不要等许可。

[80:50]Yeah. I think this is my favorite reason to work at a startup at some point in your life, 是的。我觉得这是在人生某个阶段去创业公司工作的最好理由，

[80:55]because like one thing that was like very life-changing for me was actually working at 因为对我来说非常改变人生的一件事就是

[80:59]scale when we were 20 people. And so there was just no process and we had like really big problems 在 Scale 只有 20 个人的时候工作。那时候完全没有流程，

[81:04]that we needed to solve. And it was like, I really appreciate Alex and the rest of the team for like 但我们需要解决非常大的问题。我真的很感谢 Alex 和团队其他人

[81:10]empowering me. Alex: 给了我这么大的自主权。

[81:11]And the rest of the team to just like figure things out without any boundaries for what sales is 让我和团队其他人可以不受边界限制地去解决问题，

[81:16]supposed to do, what office is supposed to do, what engineer is supposed to do. Just like you 不管销售应该做什么，运营应该做什么，工程师应该做什么。

[81:20]have all the tools at your disposal. You have some like ambitious, hairy problem statement and you can 就像你拥有所有可用的工具，面对一些雄心勃勃的棘手问题，

[81:26]do whatever you need to like get to a good solution. Like you almost need that experience 你可以做任何需要做的事来找到好的解决方案。你几乎需要那种经历

[81:30]to build that skill, to feel comfortable doing that. Cause a lot of people, you know, they go 来培养那种技能，让自己习惯这样做。因为很多人，你知道的，

[81:34]through school or in college and all these, like do the thing we tell you to do, and then you will 从学校或大学一路走来，都是"做我们让你做的事，然后你就会

[81:38]get a good grade. And you have to kind of unlearn that of like, okay, I'm just going to 得到好成绩"。你需要忘掉这种思维，就像，好的，我就去

[81:41]do the thing that needs to be done. And even if people think it's dumb, 做需要做的事。即使别人觉得这很蠢，

[81:45]I think it's the right thing to do. Yeah, exactly. 但我觉得这是对的事情。是的，没错。

[81:47]Okay. I actually have two more quick questions. Two more final questions. One is, 好的。其实我还有两个快问快答。最后两个问题。一个是，

[81:50]when Claude thinks there's all these, I don't know if you call them verbs, Claude 会有那些——我不知道你们叫不叫动词，

[81:54]what's the term for these things? Thinking words. 这些东西叫什么？思考词。

[81:56]Thinking words. And interestingly, these all leaked in the source code. 思考词。有趣的是，这些词在源代码里泄露了。

[82:00]Is it, do you have a favorite thinking word? I really like manifesting. It's also 你有最喜欢的思考词吗？我非常喜欢"manifesting"。这也是

[82:06]like the sticker that I have on my laptop. Oh, amazing. Clearly the winner. 我笔记本电脑上的贴纸。哦，太棒了。显然是赢家。

[82:11]Okay. Final question asked for us this too, with AGI potentially arriving in our lifetime, 好的。最后一个问题，AGI 有可能在我们有生之年到来，

[82:18]when you don't potentially have to work, what are you going to do? What are you going to do 当你不用工作的时候，你会做什么？你会用你所有的时间做什么？

[82:23]with all your time? I think it will take a long time for AGI to diffuse across society. So I think 我觉得 AGI 扩散到整个社会还需要很长时间。所以我觉得

[82:29]the immediate thing is actually just like helping bring the world along. I think my like non-serious 眼前的其实是帮助世界跟上步伐。我的不太认真的

[82:34]answer for after this happens is I'll probably just do a lot of rock climbing. I'll probably just live in 回答是，那之后我可能就是去攀岩。我可能就搬到

[82:41]some, I'll probably move to like Fountain Blue and just like live amongst 10,000 boulders and climb Fontainebleau，住在那一万块巨石中间，爬一阵子。

[82:48]for a bit. There's also so many books I want to read that my goal is to be able to read one or two 还有好多书想读，目标是每周能读一到两本书。

[82:55]books a week. And I'm currently at probably like 0.5. The backlog is pretty big. I think there's 现在大概只能读 0.5 本。积压的量很大。我觉得

[83:04]just so much we can learn from history and so much that I don't understand as well as I would love to. 历史有太多我们可以学习的，太多我还不够了解但很想深入了解的。

[83:11]Anything about physics and or like robotics or like any hardware or like aerospace or there's 物理学，或者机器人学，或者硬件，或者航空航天——有

[83:18]just so many interesting topics. So I'm excited to learn even even knowing that the AGI will already 太多有趣的话题了。所以我很期待去学习，即使知道 AGI

[83:24]know it. Kat, this was amazing. You're awesome. Two follow up questions. Where can folks find 已经都知道了。Kat，太棒了。你太厉害了。两个后续问题。

[83:31]you online if they want to reach out and just follow what you're up to? And how can listeners 大家在网上哪里可以找到你，如果想联系你或关注你的动态？

[83:34]be useful to you? The best way to reach out is I am underscore Kat Wu on Twitter. 听众怎么帮到你？最好的联系方式是我在 Twitter 上的账号：_KatWu。

[83:40]Feel free to like tag me in things, feel free to DM me. I read all my DMs. I don't always respond 可以在推文里 tag 我，也可以给我发私信。我会读所有私信。

[83:48]to every single one, but I will read them all. And then the thing that is most helpful is tell 虽然不一定会每条都回复，但我都会看。然后最有帮助的是告诉我们

[83:55]us where Cloud Code and CoWork aren't working well for you. We are very grateful for the amount of Claude Code 和 Cowork 在哪里对你不好用。我们非常感激

[84:02]positive feedback. But the thing that we thrive on is edge cases, errors, like specific tasks that 大量的正面反馈。但我们最需要的是边界情况、错误，那些我们能

[84:10]we can reproduce where Cloud Code or CoWork fail. Because if you are able to share that with us and 复现的具体任务，Claude Code 或 Cowork 在哪里失败的。因为如果你能分享给我们，

[84:17]we're able to reproduce it, then this is something that we're able to actively improve for our next 我们能复现，那这就是我们能主动改进的，为下一代模型

[84:21]generations of models and for our next harnesses. Extremely cool. Everyone on people on Twitter 和下一代工具。太酷了。Twitter 上的人分享这些反馈一点都不害羞。

[84:28]are not shy with sharing this feedback. So keep it coming. Share, share. Please, 所以继续来吧。分享，分享。请，

[84:32]please share the problems that you're having with us. Yeah. And it's really cool to see all you, 请把你们遇到的问题分享给我们。是的。而且很酷的是你们

[84:36]your team being on so active on Twitter and responding to people. And so, 整个团队在 Twitter 上这么活跃，回复大家。所以——

[84:40]so like what I'm hearing, like, this is actually stuff you guys actually see and react to. So. 我听到的就是，这些确实是你们真正会看到并做出反应的东西。

[84:44]Yeah. We appreciate everyone being so engaged with us. It gives the team a ton of energy. 是的。我们感谢大家的积极参与。这给团队带来了很多能量。

[84:49]We, we have this channel of like user love. And so whenever you guys share a success story, 我们有一个"用户之爱"频道。所以每当你们分享成功故事，

[84:54]we post it there. And whenever you guys share like issues with our product, we 我们就会发到那里。每当你们分享产品的问题，

[84:59]put it into our feedback channel. That way our broader team is able to act on it. 我们就会放进反馈频道。这样我们更大的团队就能据此行动了。

[85:02]That is so cool to know. Thanks for sharing that. Well, Kat, thank you so much for being here. 知道这个太酷了。谢谢分享。Kat，非常感谢你来。

[85:07]Thanks for having me. 谢谢你的邀请。

[85:09]Bye everyone. 大家再见。

[85:10]Thank you so much for listening. If you found this valuable, you can subscribe to the 非常感谢收听。如果你觉得有价值，可以在

[85:15]show on Apple podcasts, Spotify, or your favorite podcast app. Also please consider giving us a Apple Podcasts、Spotify 或你最喜欢的播客应用上订阅。也请考虑给我们

[85:21]rating or leaving a review as that really helps other listeners find the podcast. You can find 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。你可以在

[85:26]all past episodes or learn more about the show at Lenny's podcast.com. See you in the next episode. Lenny's Podcast.com 找到所有往期节目或了解更多关于这个节目的信息。下期再见。