Claude Code：我们所知的软件工程的终结

今天的嘉宾是 Boris Cherny，Anthropic 公司 Claude Code 的负责人。Claude Code 对世界的影响难以言表。这期节目播出时，正好是 Claude Code 发布一周年。在这么短的时间内，它彻底改变了软件工程师的工作。现在它也开始改变科技行业许多其他职能的工作，我们会在节目中聊到。Claude Code 本身也是 Anthropic 过去一年整体增长的重要驱动力。他们刚刚以超过 3500 亿美元的估值完成了一轮融资。正如 Boris 提到的，Claude Code 的增长仍在加速。仅在过去一个月，他们的日活跃用户就翻了一番。Boris 也是一个非常有趣、有思想、有深度的人。在这次对话中，我们发现我们出生在乌克兰的同一个城市。这太有趣了，我完全不知道。非常感谢 Ben Mann、Jenny Wen 和 Mike Krieger 为这次对话建议话题。别忘了访问 Lenny's Product Pass dot com，那里有专门为 Lenny 的新闻通讯订阅者提供的超值优惠。在简短地感谢我们的赞助商之后，我们开始吧。

Today my guest is Boris Cherny, head of Claude Code at Anthropic. It is hard to describe the impact that Claude Code has had on the world. Around the time this episode comes out will be the one-year anniversary of Claude Code. And in that short time, it has completely transformed the job of a software engineer. And it is now starting to transform the jobs of many other functions in tech, which we talk about. Claude Code itself is also a massive driver of Anthropic's overall growth over the past year. They just raised a round at over $350 billion. And as Boris mentions, the growth of Claude Code itself is still accelerating. Just in the past month their daily active users has doubled. Boris is also just a really interesting, thoughtful, deep-thinking human. And during this conversation we discovered we were born in the same city in Ukraine. That is so funny, I had no idea. A huge thank you to Ben Mann, Jenny Wen, and Mike Krieger for suggesting topics for this conversation. Don't forget to check out Lenny's Product Pass dot com for an incredible set of deals available exclusively to Lenny's newsletter subscribers. Let's get into it after a short word from our wonderful sponsors.

Boris，非常感谢你来做客播客。

Boris, thank you so much for being here and to the podcast.

谢谢邀请。

Yeah, thanks for having me on.

我想从一个劲爆的问题开始。大约 6 个月前，我不知道大家是否还记得，你实际上离开了 Anthropic，加入了 Cursor，然后两周后又回到了 Anthropic。发生了什么？我好像从来没听过完整的故事。

I want to start with a spicy question. About 6 months ago, I don't know if people even remember this, you actually left Anthropic, you joined Cursor, and then 2 weeks later you went back to Anthropic. What happened there? I don't think I've ever heard the actual story.

这是我换工作最快的一次。我加入 Cursor 是因为我非常喜欢他们的产品。老实说，我见了团队后印象非常深刻。他们是一个很棒的团队。我仍然认为他们很棒，他们正在构建非常酷的东西。而且我觉得他们比很多人更早看到了 AI 编程的发展方向。所以打造一个好产品的想法对我来说非常令人兴奋。但一到那里，我就开始意识到我真正想念 Anthropic 的是它的使命。这也是最初吸引我加入 Anthropic 的原因。在加入 Anthropic 之前，我在大科技公司工作，后来我想去一个实验室，以某种方式帮助塑造我们正在构建的这个疯狂事物的未来。吸引我加入 Anthropic 的是它的使命，那就是安全。当你和 Anthropic 的人聊天时，随便在走廊里找个人，问他们为什么在这里，答案总是安全。这种使命驱动深深地引起我的共鸣，而且我个人知道这是我获得幸福所需要的东西。这正是我真正想念的。我发现，无论工作是什么，无论多么令人兴奋，即使是构建一个很酷的产品，也无法替代那种使命感。所以对我来说，很快就明显感觉到我缺少了那个。

It's the fastest job change that I've ever had. I joined Cursor because I'm a big fan of the product. And honestly, I met the team and I was just really impressed. They're an awesome team. I still think they're awesome, and they're just building really cool stuff. And kind of they saw where AI coding was going, I think, before a lot of people did. So the idea of building a good product was just very exciting for me. I think as soon as I got there, what I started to realize is what I really missed about Anthropic was the mission. And that's actually what originally drove me to Anthropic, also. Before I joined Anthropic, I was working in big tech, and then at some point I wanted to work at a lab to just help shape the future of this crazy thing that we're building in some way. And the thing that drew me to Anthropic was the mission, and it was all about safety. When you talk to people at Anthropic, just find someone in the hallway, if you ask them why they're here, the answer is always going to be safety. So this kind of mission drivenness just really, really resonated with me, and I just know personally it's something I need in order to be happy. And that's just the thing that I really missed. And I found that, whatever the work might be, no matter how exciting, even if it's building a really cool product, it's just not really a substitute for that. So for me it was actually pretty obvious that I was missing that pretty quick.

好的，让我顺着你回到 Anthropic 以及你在那里所做的工作这条线继续聊。这期播客将在 Claude Code 发布一周年左右播出。所以我想花点时间回顾一下你所产生的影响。最近有一份报告，我相信你看到了，来自 SemiAnalysis，显示目前所有 GitHub 提交中有 4%是由 Claude Code 编写的。他们预测到今年年底，这个比例将达到五分之一。他们的说法是：“就在我们眨眼之间，AI 吞噬了所有软件开发。”在我们录制这期节目的当天，Spotify 刚刚发布了一个头条新闻，称他们最好的开发者自去年 12 月以来就没有写过一行代码，这要归功于 AI。越来越多最资深的高级工程师，包括你，都在分享一个事实：你们不再写代码了，所有代码都是 AI 生成的，而且很多人甚至不再看代码了——这就是我们目前达到的程度。

Okay, so let me follow the thread of just coming back to Anthropic and the work you've done there. This podcast is going to come out around the year anniversary of launching Claude Code. So I want to spend a little time just reflecting on the impact that you've had. There's this report that recently came out that I'm sure you saw by SemiAnalysis that showed that 4% of all GitHub commits are authored by Claude code now. And they predicted it'll be a fifth of all code commits on GitHub by the end of the year. The way they put it is, "While we blinked, AI consumed all software development." The day that we're recording this, Spotify just put out this headline that their best developers haven't written a line of code since December thanks to AI. More and more of the most advanced senior engineers, including you, are sharing the fact that you don't write code anymore, that it's all AI-generated, and many aren't even looking at code anymore is how far we've gotten.

我 100%的代码都是由 Claude Code 编写的。自去年 11 月以来，我没有手动编辑过一行代码。每天我提交 10、20、30 个拉取请求。所以，目前我有大约五个智能体在运行。

100% of my code is written by Claude Code. I have not edited it a single line by hand since November. Every day I ship 10, 20, 30 pull requests. So, at the moment I have like five agents running.

就在我们录制的时候？

While we're recording this?

是的，是的，是的。

Yeah, yeah, yeah.

你怀念写代码吗？

Do you miss writing code?

我从未像今天这样享受编程，因为我不必处理所有琐碎的细节。每个工程师的生产力提高了 200%。一直有这样一个问题：我该不该学编程？一两年后，这就不重要了。编程基本上已经被解决了。我想象一个每个人都能编程的世界。任何人都可以随时构建软件。软件编写方式的下一个重大转变是什么？Claude 开始提出想法。它查看反馈、查看错误报告、查看遥测数据以修复错误和发布内容。它更像一个同事。很多收听这期节目的人都是产品经理，他们可能正在冒汗。我认为到今年年底，每个人都会成为产品经理，每个人都会编程。软件工程师这个头衔将开始消失。它将被“构建者”取代，这对很多人来说将是痛苦的。

I have never enjoyed coding as much as I do today because I don't have to deal with all the minutiae. Productivity per engineer has increased 200%. There's always this question, should I learn to code? In a year or two it's not going to matter. Coding is virtually solved. I imagine a world where everyone is able to program. Anyone can just build software anytime. What's the next big shift to how software is written? Claude is starting to come up with ideas. Looking through feedback, it's looking at bug reports, it's looking at telemetry for bug fixes and things to ship. A little more like a co-worker or something like that. A lot of people listening to this are product managers and they're probably sweating. I think by the end of the year everyone's going to be a product manager and everyone codes. The title software engineer is going to start to go away. It's just going to be replaced by builder and it's going to be painful for a lot of people.

很大程度上，这要归功于你启动的这个项目，以及你的团队在过去一年里将其规模化。我很想听听你对过去一年的反思，以及你的工作所产生的影响。

In large part, thanks to this little project that you started and that your team has scaled over the past year. I'm curious just to hear your reflections on this past year and the impact that your work has had.

这些数字简直太疯狂了，对吧？比如，全球 4% 的提交量远超我的想象。而且就像你说的，这仍然感觉像是起点。这些还只是公开提交。所以我们认为，如果看私有仓库，这个比例还要高得多。对我来说，最疯狂的不是当前的数字，而是我们的增长速度——无论看哪个指标，Claude Code 的增长率都在持续加速。所以它不仅在增长，而且增长得越来越快。

These numbers are just totally crazy, right? Like, 4% of all commits in the world is just way more than I imagined. And like you said, it still feels like the starting point. These are also just public commits. So, we actually think if you look at private repositories, it's quite a bit higher than that. And I think the craziest thing for me isn't even the number that we're at right now, but the pace at which we're growing because if you look at Claude Code's growth rate kind of across any metric, it's continuing to accelerate. So, it's not just going up, it's going up faster and faster.

当我刚开始做 Claude Code 时，它本来只是一个小 hack。我们 Anthropic 大致知道想推出某种编程产品。很长一段时间里，我们构建模型的方式符合我们构建安全 AGI 的心智模型：模型先非常擅长编程，然后非常擅长工具使用，再然后非常擅长计算机使用。大致就是这样的轨迹。我们在这方面已经工作了很长时间。我最初加入的团队叫做 Anthropic Labs 团队。实际上 Mike Krieger 和 Ben Mann 刚刚重新启动了这支团队，进行第二轮开发。团队做出了一些很酷的东西。我们构建了 QuadCode、MCP 和桌面应用。所以你可以看到这个想法的种子：编程，然后是工具使用，再然后是计算机使用。这对 Anthropic 之所以重要，是因为安全。又回到这一点了。

When I first started Claude Code, it was just going to be a little hack. You know, we broadly knew at Anthropic that we wanted to ship some kind of coding product. And for Anthropic for a long time, we were building the models in this way that kind of fit our mental model of the way that we build safe AGI, where the model starts by being really good at coding, then it gets really good at tool use, then it gets really good at computer use. Roughly, this is like the trajectory. And we've been working on this for a long time. When you look at the team that I started on, it was called the Anthropic Labs team. And actually Mike Krieger and Ben Mann, they just kicked this team off again for kind of round two. The team built some pretty cool stuff. So, we built QuadCode, we built MCP, we built the desktop app. So, you can kind of see the seeds of this idea: coding, then tool use, then computer use. And the reason this matters for Anthropic is because of safety. It's kind of again, just back to that.

人工智能正变得越来越强大，越来越有能力。过去一年发生的变化是，至少对工程师来说，AI 不仅仅写代码。它不只是对话伙伴，而是实际使用工具，在世界上行动。我认为现在通过 Co-work，我们开始看到非技术人员也在经历这种转变。对于许多使用对话式 AI 的人来说，这可能是他们第一次使用真正能行动的东西。它真的可以用你的 Gmail，可以用你的 Slack，可以为你做所有这些事情，而且做得相当好。而且只会越来越好。

AI is getting more and more powerful. It's getting more and more capable. The thing that's happened in the last year is that for at least for engineers, the AI doesn't just write the code. It's not just a conversation partner, but it actually uses tools. It acts in the world. And I think now with Co-work, we're starting to see the transition for non-technical folks also. For a lot of people that use conversational AI, this might be the first time that they're using the thing that actually acts. It can actually use your Gmail. It can use your Slack. It can do all these things for you, and it's quite good at it. And it's only going to get better from here.

所以我认为 Anthropic 长期以来一直有种感觉：我们想构建点什么，但不太明确是什么。因此，当我加入 Anthropic 时，我花了一个月时间做各种 hack，构建了一堆奇怪的原型。大部分都没有发布，甚至接近发布都谈不上。那只是为了理解模型能力的边界。然后我又花了一个月做后训练，以了解研究方面。老实说，作为工程师，我发现要做好工作，你必须理解你所工作层之下的那一层。在传统工程工作中，如果你在做产品，你需要理解基础设施、运行时、虚拟机、语言，无论是什么，你构建其上的系统。但如果你做 AI，你真的需要在一定程度上理解模型才能做好工作。所以我绕了点路去做那件事，然后回来开始原型设计，最终变成了 QuadCode。

So, I think for Anthropic for a long time, there was this feeling that we wanted to build something, but it wasn't obvious what. So, when I joined Anthropic, I spent one month kind of hacking and built a bunch of weird prototypes. Most of them didn't ship, and weren't even close to shipping. It was just kind of understanding the boundaries of what the model can do. Then I spent a month doing post-training. So, to understand the research side of it. I think honestly, that's just for me as an engineer, I find that to do good work, you really have to understand the layer under the layer at which you work. With traditional engineering work, if you're working on a product, you want to understand the infrastructure, the runtime, the virtual machine, the language, whatever that is, the system that you're building on. But if you're working AI, you just really have to understand the model to some degree to do good work. So, I took a little detour to do that, and then I came back and just started prototyping what eventually became QuadCode.

它的第一个版本，我录了一个夏天的视频，因为我把这个演示录下来并发布了。当时它叫 Claude CLI。我只是展示了它如何使用几个工具。让我震惊的是，我给了它一个 bash 工具，它就能用那个工具写代码来告诉我我正在听什么音乐，当我问它‘我在听什么音乐’时。这是最疯狂的事。因为我没有指示模型用这个工具做这个或做那个。模型得到了这个工具，然后自己弄明白了如何用它来回答我的问题，而我甚至不确定自己能否回答‘我在听什么音乐’。所以我开始进一步原型设计。我发了一篇帖子，在内部宣布，得到了两个赞。这就是当时的反应。因为我认为内部的人，当你想到编程工具时，你会想到 IDE，想到所有这些非常复杂的环境。没人认为这个东西可以基于终端。那是一种奇怪的设计方式，而且也不是本意。但从一开始我就在终端里构建它，因为头几个月只有我一个人。所以那是最简单的构建方式。对我来说，这实际上是一个非常重要的产品教训。你希望在开始时资源投入少一点。

The very first version of it, there's a video recording of the summer because I recorded this demo and posted it. It was called Claude CLI back then. And I just showed off how it used a few tools. The shocking thing for me was that I gave it a bash tool and it just was able to use that to write code to tell me what music I'm listening to when I asked it like 'what music am I listening to.' This is the craziest thing. Because I didn't instruct the model to use this tool for this or do whatever. The model was given this tool and it figured out how to use it to answer this question that I had that I wasn't even sure if I could answer: what music am I listening to. So I started prototyping this a little bit more. I made a post about it and announced it internally and I got two likes. That's the reaction at the time. Because I think people internally, when you think of coding tools, you think of IDEs, you think of all these pretty sophisticated environments. No one thought that this thing could be terminal based. That's sort of a weird way to design it and that wasn't really the intention. But from the start I built it in a terminal because for the first couple months it was just me. So, it was just the easiest way to build. For me this is actually a pretty important product lesson. You want to under-resource things a little bit at the start.

然后我们开始考虑应该构建其他什么形态，但实际上我们决定暂时坚持用终端。最大的原因是模型改进得太快了。我们认为没有其他形态能跟上它的速度。老实说，这只是我在纠结应该构建什么，过去一年 QuadCode 就是我全部的心思。所以就像深夜，我一直在想：好吧，模型在持续改进。我们该怎么办？我们怎么能跟上？终端实际上是我唯一想到的主意。结果它流行起来了。发布后，很快在 Anthropic 内部大获成功，日活跃用户直线上升。实际上在发布之前，Ben Mann 就催我做 DAU 图表。我说，还有点早，也许现在真的该做吗？他说，是的。然后图表立刻直线上升。然后在二月份我们对外发布。实际上，人们不太记得的是 Claude Code 一开始并不火爆。发布时，它获得了一批用户，有很多早期采用者立刻接受了它。但真正让所有人理解这个东西是什么，花了好几个月。再说一次，它太不同了。

Then we started thinking about what other form factors we should build and we actually decided to stick with the terminal for a while. The biggest reason was the model is improving so quickly. We thought that there wasn't really another form factor that could keep up with it. Honestly, this was just me kind of struggling with what should we build, for the last year QuadCode has just been all I think about. So just like late at night, this is just something I was thinking about: okay, the model's continuing to improve. What do we do? How can we possibly keep up? And the terminal was honestly just the only idea that I had. And it ended up catching on. After I released it, pretty quickly it became a hit at Anthropic and the daily active users just went vertical. Really early on actually before I launched it, Ben Mann nudged me to make a DAU chart. And I was like, it's kind of early, maybe should we really do it right now? And he was like, yeah. And so the chart just went vertical pretty immediately. Then in February we released it externally. Actually, something that people don't really remember is Claude Code was not initially a hit. When we released it, it got a bunch of users. There was a lot of early adopters that got it immediately. But it actually took many months for everyone to really understand what this thing is. Just again, it's like it's just so different.

当我回想起来，Claude Code 之所以能成功，部分原因在于这种潜在需求——我们把工具带到人们所在的地方，让现有工作流程变得更轻松。但同时也因为它是在终端里，有点出人意料，有点陌生。所以你必须保持开放心态，学会使用它。当然，现在 Claude Code 已经可以在 iOS 和 Android 的 Claude 应用、桌面应用、网站、Slack 和 GitHub 的编辑扩展中使用——所有工程师所在的地方。它变得熟悉了一些。但一开始并非如此。所以，是的，一开始连这个东西是否有用都让人惊讶。随着团队壮大、产品发展，它开始对人们越来越有用，从世界各地的小初创公司到最大的 FAANG 公司都开始使用它，并给出反馈。回想起来，这是一次非常谦卑的经历，因为我们不断从用户那里学习，最令人兴奋的是，我们其实都不知道自己在做什么。我们只是和大家一起摸索，而最好的信号就是用户的反馈。所以那是最好的部分。我被惊讶了很多次。

And when I think about it, kind of part of the reason Claude Code works is this idea of latent demand where we bring the tool to where people are and it makes existing workflows a little bit easier. But also because it's in a terminal, it's a little surprising. It's a little alien in this way. So you have to be open-minded and learn to use it. And of course now, Claude Code is available in the iOS and Android Claude app, in the desktop app, on the website, as edit extensions in Slack and GitHub—all these places where engineers are. It's a little more familiar. But that wasn't the starting point. So yeah, at the beginning it was kind of a surprise that this thing was even useful. And as the team grew, as the product grew, as it started to become more and more useful to people, just people around the world from small startups to the biggest FAANG companies started using it and they started giving feedback. And I think just reflecting back, it has been such a humbling experience because we just keep learning from our users and the most exciting thing is that none of us really know what we're doing. And we're just trying to figure it out along with everyone else and the single best signal for that is just feedback from users. So that's just been the best. I've been surprised so many times.

你一年前推出了这个产品，这并非人们第一次能用 AI 编程，但一年之内，整个软件工程行业发生了翻天覆地的变化。比如那些预测说‘AI 将编写 100% 的代码’，大家都说‘不，那太疯狂了，你在说什么？’而我觉得，‘哦，当然，事情正如他们所说的那样发展。’只是现在事情变化得太快了。

You launched this a year ago, and it wasn't the first time people could use AI to code, but in a year, the entire profession of software engineering has dramatically changed. Like there's all these predictions, 'Oh, AI's going to write 100% of code.' Everyone's like, 'No, that's crazy. What are you talking about?' And I was like, 'Oh, of course, it's happening exactly as they said.' It's just that things move so fast and change so fast now.

是的，非常快。回想五月份的 Code with Claude 活动，那是我们 Anthropic 举办的第一次开发者大会。我做了一个简短的演讲，在问答环节，人们问‘你对年底有什么预测？’我在 2025 年 5 月的预测是：‘到年底，你可能不再需要 IDE 来编程了。我们将开始看到工程师不再做这个。’我记得房间里一片哗然。那真是个疯狂的预测。但我认为在 Anthropic，我们就是这样思考问题的——指数级思维。这深深植根于我们的 DNA。看看我们的联合创始人，其中三位是缩放定律论文的前三位作者。所以我们真的以指数方式思考。如果你看看当时 Claude 编写的代码百分比指数曲线，只要顺着趋势线看，很明显我们会在年底前超过 100%，尽管这完全不符合直觉。所以我只是顺着趋势线看，果然，在 11 月，我个人就达到了这个状态，从那以后一直如此，而且我们开始看到很多不同的客户也达到了这个状态。

Yeah, it's really fast. Back at Code with Claude back in May, that was like our first developer conference that we did as Anthropic. I did a short talk and in the Q&A after the talk, people were asking, 'What are your predictions for the end of the year?' And my prediction back in May of 2025 was, 'By the end of the year, you might not need an IDE to code anymore. And we're going to start to see engineers not doing this.' And I remember the room audibly gasped. It was such a crazy prediction. But I think at Anthropic, this is just the way we think about things—exponentials. And this is very deep in the DNA. If you look at our co-founders, three of them were the first three authors on the scaling laws paper. So we really just think in exponentials. And if you look at the exponential of the percent of code that was written by Claude at that point, and if you just trace the line, it's pretty obvious we're going to cross 100% by the end of the year, even if it does not match intuition at all. And so all I did was trace the line, and yeah, in November, that happened for me personally, and that's been the case since, and we're starting to see that for a lot of different customers, too.

我觉得你刚才分享的这段经历非常有趣。这种‘随便玩玩，看看会发生什么’的想法是不是很关键？在 Open Claude 上经常出现这种情况，就像 Peter 随便玩玩，然后事情就发生了。感觉这是许多重大 AI 创新的核心要素——人们只是坐下来尝试各种东西，把模型推得比大多数人都更远。

I thought that was really interesting what you just shared there about the journey. Is this kind of idea of just playing around and seeing what happens? This comes up with Open Claude a lot, just like Peter was playing around, and just like a thing happened. And it feels like that's a central ingredient to a lot of the biggest innovations in AI—people just sitting around trying stuff, pushing the models further than most other people.

我的意思是，这就是创新的本质，对吧？你不能强迫它。创新没有路线图。你只需要给人们空间。你可能需要给他们一种安全感。也就是说，心理上要安全，失败是可以的。80% 的想法是糟糕的也没关系。但你也需要让他们承担一点责任。所以，如果想法不好，你就止损，转向下一个想法，而不是继续投入。在 Claude Code 的早期，我完全不知道这个东西会有用。因为即使在二月份我们刚开始的时候，它可能只写了我 20% 的代码，不会更多。甚至在五月份，它可能只写了 30%。我大部分代码还在用 Cursor。直到 11 月才超过 100%。所以这花了一段时间，但即使从最早的日子开始，我就感觉我发现了什么，我每个晚上、每个周末都在捣鼓这个。幸运的是，我的妻子非常支持。但就是感觉我发现了什么，虽然不明显是什么。然后有时候，你找到一根线，你只需要顺着它拉。

I mean, this is the thing about innovation, right? You can't force it. There's no road map for innovation. You just have to give people space. You have to give them maybe the word is like safety. So, it's psychological safety that it's okay to fail. It's okay if 80% of the ideas are bad. You also have to hold them accountable a bit. So, if the idea is bad, you cut your losses, move on to the next idea. Instead of investing more. In the early days of Claude Code, I had no idea that this thing would be useful at all. Because even in February when we started, it was writing maybe 20% of my code, not more. And even in May, it was writing maybe 30%. I was still using Cursor for most of my code. And it only crossed 100% in November. So, it took a while, but even from the earliest day, it just felt like I was onto something and I was just spending every night, every weekend hacking on this. And luckily, my wife was very supportive. But it just felt like I was onto something. It wasn't obvious what. And then sometimes, you find a thread, you just have to pull on it.

那么，目前你 100% 的代码都是由 Claude Code 编写的。这是你现在的编程状态吗？

So, at this point, 100% of your code is written by Claude Code. Is that the current state of your coding?

是的，我 100% 的代码都是由 Claude Code 编写的。我是一个相当高产的编码者。即使在 Instagram 工作的时候也是如此，我是效率最高的几位工程师之一。在 Anthropic 这里也仍然如此。

Yeah, so 100% of my code is written by Claude Code. I'm a fairly prolific coder. And this has been the case even when I worked back at Instagram. I was like one of the top few most productive engineers. And that's actually still the case here at Anthropic.

哇，即使作为团队负责人。

Wow, even as head of the team.

是的，是的。我仍然做很多编码工作。所以每天我大概会提交 10、20、30 个拉取请求，差不多这样。

Yeah, yeah. I still do a lot of coding. And so every day I ship like 10, 20, 30 pull requests, something like that.

每天？

Every day?

每天。是的。

Every day. Yeah.

天哪。100% 由 Claude Code 编写。自从 11 月以来，我没有手动编辑过一行代码。是的，就是这样。我确实会看代码。所以，我认为我们还没有到可以完全放手的地步，尤其是当有很多人在运行程序时。你必须确保代码正确，必须确保它安全等等。然后我们还有 Claude 对所有内容进行自动代码审查。所以，在 Anthropic，Claude 审查 100% 的拉取请求。之后仍然有一层人工审查，但你仍然需要这些检查点。你仍然需要有人查看代码，除非是纯原型代码，不会在任何地方运行。那只是原型。

Good god. 100% written by Claude Code. I have not edited a single line by hand since November. And yeah, that's been it. I do look at the code. So, I don't think we're at the point now where you can be totally hands-off, especially when there's a lot of people running the program. You have to make sure that it's correct. You have to make sure it's safe and so on. And then we also have Claude doing automatic code review for everything. So, here at Anthropic, Claude reviews 100% of pull requests. There's still a layer of human review after it, but you still want some of these checkpoints. You still want a human looking at the code, unless it's pure prototype code that's not going to run anywhere. It's just a prototype.

下一个前沿是什么？目前，你 100% 的代码都是由 AI 编写的。这显然是软件工程领域每个人的发展方向。那曾经感觉是一个疯狂的里程碑。现在却觉得理所当然。这就是现在的世界。接下来软件编写方式会发生什么重大转变，要么你的团队已经在这样做了，要么你认为会朝着这个方向发展？

What's the next frontier? So at this point, 100% of your code is being written by AI. This is clearly where everyone is going in software engineering. That felt like a crazy milestone. Now it's just like, of course. This is the world now. What's the next big shift to how software is written that either your team's already operating in or you think will head towards?

我认为现在正在发生的一件事是，Claude 开始提出想法了。所以，Claude 正在查看反馈。

I think something that's happening right now is Claude is starting to come up with ideas. So, Claude is looking through feedback.

它现在会看 bug 报告、遥测数据之类的东西，然后开始提出修复 bug 和发布功能的点子。它越来越像一个同事了。第二件事是，我们开始跳出编程领域。现在可以说编程基本上被解决了，至少对我做的这类编程来说，Claude 已经能搞定。所以我们开始想下一步是什么。有很多跟编程相关的事情，但也有通用任务。比如我现在每天用 Claude 做各种跟编程无关的事，像前几天交停车罚单，还有所有项目管理——同步表格、在 Slack 和邮件上联系别人。所以我觉得前沿是这种方向，而不是编程，因为编程已经基本解决了。未来几个月，整个行业里，各种代码库和技术栈都会越来越被解决。

It's looking at bug reports, telemetry, and things like this, and it's starting to come up with ideas for bug fixes and things to ship. So it's just starting to get a little more like a co-worker. I think the second thing is we're starting to branch out of coding a little bit. So at this point, it's safe to say that coding is virtually solved. At least for the kinds of programming that I do, it's a solved problem because Claude can do it. So now we're starting to think about what's next. There are a lot of things adjacent to coding, but also general tasks. For example, I use Claude every day now to do all sorts of things unrelated to coding automatically. Like paying a parking ticket the other day, or all my project management—syncing spreadsheets, messaging people on Slack and email. So I think the frontier is something like this, not coding, because coding is pretty much solved. Over the next few months, across the industry, it will become increasingly solved for every code base and tech stack.

帮你决定做什么这个想法太有意思了。很多产品经理可能都在冒汗。你是怎么用 Claude 做这个的？就是跟它聊天吗？有没有什么巧妙的用法？

This idea of helping you come up with what to work on is so interesting. A lot of product managers are probably sweating. How do you use Claude for this? Do you just talk to it? Is there anything clever you've come up with?

说实话，最简单的办法就是打开 Claude Coder，然后指向一个 Slack 线程。我们有一个频道专门收集 Claude Coder 的内部反馈。从我们第一次发布，甚至 2024 年内部测试开始，反馈就像洪水一样。早期，只要有人提反馈，我就会尽快修复每一个问题——一分钟或五分钟内。这种快速反馈循环鼓励更多反馈，因为让人感觉被倾听。通常反馈会石沉大海，但如果你让人感觉被倾听，他们就愿意贡献。现在我也这么做，但 Claude 做了大部分工作。我把它指向那个频道，它就会说：‘我能做这几件事。我刚提了几个 PR，要看看吗？’我说：‘好。’

Honestly, the simplest thing is to open Claude Coder and point it at a Slack thread. For us, there's a channel with all the internal feedback about Claude Coder. Since we first released it, even in 2024 internally, it's been a firehose of feedback. In the early days, anytime someone sent feedback, I would fix every single thing as fast as possible—within a minute or five minutes. This fast feedback cycle encourages more feedback because it makes people feel heard. Usually, feedback goes into a black hole, but if you make people feel heard, they want to contribute. Now I do the same thing, but Claude does a lot of the work. I point it at the channel, and it says, 'Here are a few things I can do. I just put up a couple PRs. Want to take a look?' I say, 'Yeah.'

你有没有发现它在这方面进步很大？因为这是圣杯。构建解决了，代码审查成了下一个瓶颈——谁来审查所有 PR？下一个大问题是决定构建什么、优先做什么。你说 Claude Coder 已经开始帮你做这个了。它是不是在 Opus 4.6 上进步很大？趋势是怎样的？

Have you noticed it getting much better at this? Because this is the holy grail. Building is solved, code review became the next bottleneck—who reviews all those PRs? The next big question is figuring out what to build and prioritize. And you're saying Claude Coder is starting to help with that. Has it gotten a lot better with Opus 4.6 or what's the trajectory?

是的，进步很大。一部分是专门针对编程的训练——世界上最好的编程模型，越来越好。4.6 太不可思议了。但很多编程之外的训练也转化得很好。有一种迁移效应：你教模型做 X，它做 Y 也变好了。提升幅度惊人。在 Anthropic 过去一年，自从我们推出 Claude Coder，工程团队规模大概翻了四倍，但每个工程师的生产力在 PR 数量上提升了 200%。这个数字对任何做开发效率的人来说都疯狂。我以前在 Meta 负责所有代码库的代码质量——Facebook、Instagram、WhatsApp。很多工作都跟效率有关。几百个工程师干一年，效率提升也就几个百分点。现在看到几百个百分点的提升，简直不可思议。

Yeah, it's improved a lot. Some of it is training specific to coding—best coding model in the world, getting better and better. 4.6 is incredible. But also a lot of training outside of coding translates well. There's a transfer where you teach the model to do X and it gets better at Y. The gains have been insane. At Anthropic over the last year, since we introduced Claude Coder, we probably 4x the engineering team, but productivity per engineer has increased 200% in terms of pull requests. This number is crazy for anyone who works on dev productivity. In a previous life at Meta, I was responsible for code quality across all code bases—Facebook, Instagram, WhatsApp. A lot of that was about productivity. With hundreds of engineers working on it for a year, you'd see a gain of a few percentage points. Now seeing gains of hundreds of percentage points is absolutely insane.

同样疯狂的是这一切都变得习以为常了。我们听到这些数字，觉得‘当然 AI 能做到’。软件开发、产品构建和科技世界的变化程度前所未有。很容易就习惯了，但重要的是要认识到这很疯狂。

What's also insane is how normalized this has all been. We hear these numbers and think, 'Of course AI is doing this.' The amount of change to software development, building products, and the tech world is unprecedented. It's easy to get used to it, but it's important to recognize this is crazy.

有一个缺点：模型变化太快，我有时会陷入旧的思维方式。团队里的新人，甚至刚毕业的，做事方式比我更 AGI 导向。比如几个月前有个内存泄漏问题。Claude Coder 的内存占用上升然后崩溃。传统做法是取堆快照，放到特殊调试器里分析。我当时就在做这个，看跟踪信息。但一个新来的工程师直接让 Claude 处理：‘嘿 Claude，有个泄漏，你能查出来吗？’然后 Claude 做了跟我完全一样的事情。

There's a downside: the model changes so often that I sometimes get stuck in old ways of thinking. New people on the team, even new grads, do things in a more AGI-forward way than I do. For example, a couple months ago there was a memory leak. Claude Coder's memory usage was going up and it would crash. Traditionally, you take a heap snapshot, put it in a special debugger, and figure it out. I was doing that, looking through traces. But a newer engineer just had Claude code it: 'Hey Claude, there's a leak. Can you figure it out?' And Claude did exactly the same thing I was doing.

它抓取了堆快照，给自己写了个小工具来分析。这有点像即时程序。它比我更快地找到了问题并提交了拉取请求。所以，对于我们这些长期使用模型的人来说，你仍然需要把自己带到当下，不要停留在旧模型上，因为它不再是 Sonnet 3.5 了。新模型完全不同。这种思维转变非常不同。

It took the heap snapshot. It wrote a little tool for itself so it can analyze it itself. It was sort of like a just-in-time program. And it found the issue and put up a pull request faster than I could. So for those of us that have been using the model for a long time, you still have to transport yourself to the current moment and not get stuck back in old model because it's not Sonnet 3.5 anymore. The new models are just completely different. And this mindset shift is very different.

我听说你为团队制定了一些非常具体的原则。新人加入时，你会带他们过一遍。我相信其中一条是‘比亲自做更好的，是让 Claude 去做’。感觉这正是你在内存泄漏中描述的——你几乎忘了那个原则：‘好吧，让我看看 Claude 能不能帮我解决。’

I hear you have these very specific principles that you've codified for your team. When people join you, you walk them through them. I believe one of them is 'What's better than doing something, having Claude do it.' And it feels like that's exactly what you described with this memory leak—you almost forgot that principle of 'Okay, let me see if Claude can solve this for me.'

当你稍微资金不足时，会发生一件有趣的事——人们被迫‘Claudify’。我们看到了这一点。有时我们只给一个项目配一名工程师，他们因为想快速交付而真的很快。这是一种内在动机。如果你有个好主意，你就想把它做出来。没人强迫你。如果你有 Claude，你可以用它来自动化很多工作。这是我们反复看到的。所以一个原则是稍微资金不足。另一个原则是鼓励人们更快行动。如果你今天能做的事，今天就做。早期这很重要，因为只有我一个人。我们唯一的优势是速度——唯一能在拥挤的编程市场中竞争的方式。但现在，这仍然是我们的原则。如果你想更快，一个好方法就是让 Claude 做更多事。所以它鼓励了这一点。

There's this interesting thing that happens when you underfund everything a little bit—people are forced to Claudify. We see this. For work, sometimes we put just one engineer on a project, and they ship really quickly because they want to ship quickly. It's an intrinsic motivation from within. If you have a good idea, you just want to get it out there. No one has to force you. And if you have Claude, you can use that to automate a lot of work. That's what we see over and over. So one principle is underfunding things a little bit. Another principle is encouraging people to go faster. If you can do something today, you should do it today. Early on, it was really important because it was just me. Our only advantage was speed—the only way we could ship a product to compete in this crowded coding market. But nowadays, it's still a principle we have. And if you want to go faster, a really good way is to have Claude do more stuff. So it encourages that.

资金不足这个想法很有趣。总的来说，有种感觉是 AI 能让你减少员工和工程师。所以不仅是你更高效；你说的是如果资金不足，你实际上会做得更好。不仅仅是 AI 能让你更快；如果做某事的人更少，你从 AI 工具中得到的会更多。

This idea of underfunding is so interesting. In general, there's this feeling that AI will allow you to have fewer employees, fewer engineers. So it's not only that you can be more productive; you're saying you will actually do better if you underfund. It's not just that AI can make you faster; you will get more out of AI tooling if you have fewer people working on something.

是的，如果你雇佣优秀的工程师，他们会想办法做到。尤其是如果你赋予他们权力。我和 CTO 以及各种公司都谈过这个。我的建议通常是：不要试图优化。一开始不要削减成本。先给工程师尽可能多的 token。现在你开始看到像 Anthropic 这样的公司——每个人都可以用很多 token。我们开始看到这成为一些公司的福利：加入就给你无限 token。我非常鼓励这一点，因为它让人们自由尝试那些原本太疯狂的想法。然后如果某个想法有效，你再想办法规模化。那时才是优化和削减成本的时候——也许你可以用 Haiku 或 Sonnet 代替 Opus。但一开始，只管投入大量 token，看看想法是否可行。给工程师这样做的自由。

Yeah, if you hire great engineers, they'll figure out how to do it. Especially if you empower them. I talk about this with CTOs and all sorts of companies. My advice generally is: don't try to optimize. Don't try to cost cut at the beginning. Start by just giving engineers as many tokens as possible. Now you're starting to see companies like Anthropic—everyone can use a lot of tokens. We're starting to see this come up as a perk at some companies: if you join, you get unlimited tokens. I very much encourage this because it makes people free to try ideas that would have been too crazy. Then if an idea works, you can figure out how to scale it. That's the point to optimize and cost cut—maybe you can do it with Haiku or Sonnet instead of Opus. But at the beginning, just throw a lot of tokens at it and see if the idea works. Give engineers the freedom to do that.

听到这个的人可能会想：‘当然，他在 Anthropic 工作。他当然希望我们尽可能多用 token。’但你说的是，最有趣的创新想法会来自有人把它用到极致，看看什么是可能的。

People hearing this may think, 'Of course, he works at Anthropic. He would want us to use as many tokens as possible.' But what you're saying is the most interesting innovative ideas will come out of someone just taking it to the max and seeing what's possible.

是的，现实是，在小规模下，你不会得到巨额账单。如果是个别工程师在实验，token 成本相对于他们的工资或运营业务的其他成本仍然相对较低。所以这不是大开销。当事情规模化时——比如他们做出了很棒的东西，消耗大量 token——成本就会变得很大。那时才是优化的时机。但不要太早优化。

Yeah, and the reality is, at small scale, you're not going to get a giant bill. If it's an individual engineer experimenting, the token cost is still relatively low relative to their salary or other costs of running the business. So it's not a huge cost. As the thing scales up—say they built something awesome and it takes a huge amount of tokens—then the cost becomes pretty big. That's the point to optimize. But don't do that too early.

你见过 token 成本高于工资的公司吗？你认为我们会看到这种趋势吗？

Have you seen companies where their token cost is higher than their salary? Is that a trend you think we'll see?

在 Anthropic，我们开始看到一些工程师每月花费数十万 token。所以我们开始看到一点。一些公司也开始出现类似情况。

At Anthropic, we're starting to see some engineers spending hundreds of thousands a month in tokens. So we're starting to see this a little bit. Some companies are starting to see similar things.

回到编程。你怀念写代码吗？你会为此感到难过吗，因为它不再是软件工程师会做的事？

Going back to coding. Do you miss writing code? Is this something you're sad about, that it's no longer something you'll do as a software engineer?

这很有趣。我学工程时，它非常实用。我学工程是为了能造东西。我是自学的。我在学校学的是经济学，不是计算机科学。我很早就自学了工程——我初中就开始编程。从一开始，它就很实用。我学编程是为了在数学考试中作弊。那是第一件事。我们有那种图形计算器，我把答案编进了 TI-83 Plus。然后第二年，数学考试太难了——我无法编入所有答案，因为我不知道题目。所以我得写一个小求解器来解代数题。然后我发现可以用一根小电缆把程序传给全班同学，全班都得 A。但我们都被抓了，老师让我们别再干了。从一开始，编程对我来说一直很实用——它是造东西的方式，而不是目的本身。

It's funny. When I learned engineering, it was very practical. I learned engineering so I could build stuff. I was self-taught. I studied economics in school but didn't study CS. I taught myself engineering early on—I was programming in middle school. From the very beginning, it was very practical. I learned to code so I could cheat on a math test. That was the first thing. We had these graphing calculators, and I programmed the answer into the TI-83 Plus. Then the next year, the math test was too hard—I couldn't program all the answers because I didn't know the questions. So I had to write a little solver that would solve algebra questions. Then I figured out you can get a little cable, give the program to the rest of the class, and the whole class gets A's. But we all got caught, and the teacher told us to knock it off. From the very beginning, it's always been very practical for me—programming is a way to build a thing. It's not the end in itself.

在某个时候，我个人陷入了编程之美的兔子洞。我写了一本关于 TypeScript 的书。我创办了当时世界上最大的 TypeScript 聚会，因为我爱上了这门语言。我深入研究了函数式编程之类的东西。我觉得很多程序员会被这个分心。对我来说，编程有一种美，尤其是函数式编程，还有类型系统。解决一个复杂的数学问题会带来一种快感，类似于平衡类型或让程序变得优美。但这并不是终点。对我来说，编码是一种工具，一种做事的方式。话虽如此，并非每个人都这么想。例如，团队里有一位工程师 Lena，她周末还在手写 C++，因为她真的很喜欢。每个人都不一样。即使这个领域在变化，也总有空间去欣赏艺术，如果你愿意，也可以亲手去做。

At some point, I personally fell into the rabbit hole of the beauty of programming. I wrote a book about TypeScript. I started the world's biggest TypeScript meetup at the time because I fell in love with the language. I got deep into functional programming and all that. I think a lot of coders get distracted by this. For me, there is a beauty to programming, especially functional programming, and to type systems. There's a buzz you get when solving a complicated math problem, similar to balancing types or making a program beautiful. But that's not the end goal. For me, coding is a tool, a way to do things. That said, not everyone feels this way. For example, there's an engineer on the team, Lena, who still writes C++ by hand on weekends because she really enjoys it. Everyone is different. Even as this field changes, there's always space to enjoy the art and do things by hand if you want.

你担心自己的工程师技能会退化吗？这是你担心的事情，还是说这就是趋势？

Do you worry about your skills atrophying as an engineer? Is that something you worry about, or is it just how it's going to go?

我认为这就是趋势。我个人很担心。对我来说，编程是一个连续体。很久以前，软件相对较新。今天的程序运行在虚拟机上，这从 1960 年代就开始了。在那之前是打孔卡，再之前是开关，再之前是硬件，再之前是纸笔——满屋子的人用纸做数学。编程一直在变。在某种程度上，你仍然需要理解底层，因为这有助于你成为更好的工程师。这可能在未来一两年内成立，但很快就不重要了。它会像程序底层的汇编代码一样。情感上，我一直需要学习新东西。作为程序员，这并不觉得新鲜，因为总有新框架和新语言。我们对此很适应。但并非所有人都这样。有些人会感到更强烈的失落、怀旧或退化感。不知道你有没有看到 Elon 说：‘为什么 AI 不直接写二进制？所有这些编程抽象的意义何在？’如果你愿意，它完全可以做到。

I think it's just the way it happens. I worry about it too much personally. For me, programming is on a continuum. Way back, software is relatively new. Programs today run on virtual machines, which has been the way since the 1960s. Before that, punch cards; before that, switches; before that, hardware; before that, pen and paper—a room full of people doing math on paper. Programming has always changed. In some ways, you still want to understand the layer below because it helps you be a better engineer. That might be true for the next year or so, but pretty soon it won't matter. It'll be like assembly code running under the program. Emotionally, I've always had to learn new things. As a programmer, it doesn't feel that new because there are always new frameworks and languages. It's something we're comfortable with. But this isn't true for everyone. Some people will feel a greater sense of loss, nostalgia, or atrophy. I don't know if you saw Elon saying, 'Why isn't the AI just writing binary straight to binary? What's the point of all this programming abstraction?' It totally can do that if you wanted to.

所以我听到的是，总有一个问题：我应该学编程吗？学校里的人应该学编程吗？从你这里我听到，一两年后你其实不需要。我的看法是，对于今天使用代码智能体的人来说，你仍然需要理解底层。但一两年后，这就不重要了。

So, what I'm hearing is there's always this question: should I learn to code? Should people in school learn to code? From you, I hear that in a year or two you don't really need to. My take is that for people using code agents today, you still have to understand the layer below. But in a year or two, it won't matter.

我在想这件事的历史类比。我们必须把它放在历史中，看看我们经历过哪些类似的转变。对我来说最接近的是印刷机。15 世纪中叶的欧洲，识字率非常低——不到人口的 1%。抄写员负责所有的书写和阅读，受雇于通常不识字的领主和国王。然后古腾堡和印刷机出现了。印刷机发明后的 50 年里，产生的印刷材料比之前一千年还多。数量激增，成本在接下来的 50 年里下降了约 100 倍。识字率花了很长时间才提高，因为学习读写很难——需要教育体系、空闲时间，不能整天在农场干活。但在接下来的 200 年里，全球识字率上升到了 70%。我认为我们可能会看到类似的转变。有一份有趣的历史文献，采访了 15 世纪的一位抄写员关于印刷机的看法。他们很兴奋，因为他们不喜欢抄写书籍，他们喜欢画插图和装订。他们很高兴时间被解放了。作为一名工程师，我感到一种相似性。我不再需要做编码的繁琐工作了——那一直是细节，是繁琐的部分。有趣的部分是弄清楚要构建什么、与用户交流、思考大系统和未来、与团队合作。这就是我现在能更多做的事情。而令人惊叹的是，你正在构建的工具允许任何人做这件事——没有技术经验的人也能做到你描述的那些。

I was thinking about the right historical analogue for this. We have to situate this in history and figure out when we've gone through similar transitions. The closest thing for me is the printing press. In Europe in the mid-1400s, literacy was very low—sub 1% of the population. Scribes did all the writing and reading, employed by lords and kings who often weren't literate themselves. Then Gutenberg and the printing press came along. In the 50 years after the printing press, more printed material was created than in the thousand years before. Volume went way up, cost went down about 100x over the next 50 years. Literacy took a while because learning to read and write is hard—it requires an education system, free time, not working on a farm all day. But over the next 200 years, it went up to 70% globally. I think we might see a similar transition. There's an interesting historical document where a scribe in the 1400s was interviewed about the printing press. They were excited because they didn't like copying between books; they liked drawing art and bookbinding. They were glad their time was freed up. As an engineer, I feel a parallel. I don't have to do the tedious work of coding anymore—that was always the detail, the tedious part. The fun part is figuring out what to build, talking to users, thinking about big systems and the future, collaborating with the team. That's what I get to do more of now. And what's amazing is that the tool you're building allows anyone to do this—people with no technical experience can do exactly what you're describing.

就像我一直在做各种小项目，任何时候卡住了，就说‘帮我解决这个问题’，然后你就被解开了。我职业生涯早期做了 10 年工程师。我记得花了很多时间在库、依赖和小细节上，就像‘天哪，我该怎么办？’然后去查 Stack Overflow。现在只需要说‘帮我解决这个问题’，然后一步一步的指导就来了。好了，搞定了。

Like I've been doing a bunch of random little projects and anytime you get stuck, just like 'help me figure this out' and you get unblocked. I used to be an engineer earlier in my career for 10 years. I just remember spending so much time on libraries and dependencies and little things, just like 'Oh my god, what do I do?' and then looking on Stack Overflow. And now it's just like 'Help me figure this out' and here's step by step 1 2 3 4. Okay, we got this.

没错。我今天早些时候和一个工程师聊过。他们用 Go 写一个服务，已经一个月了，服务建起来了，运行得挺好。然后我问‘那你用 Go 写感觉怎么样？’他说‘其实我还是不太懂 Go，但是……’

Yeah, exactly. I was talking to an engineer earlier today. They're writing some service in Go and it's been like a month already and they built up the service. It's working quite well. And then I was like 'Okay, so how do you feel writing in Go?' He was like 'You know, I still don't really know Go, but...'

我觉得我们会越来越多地看到这种情况。就像如果你知道它正确高效地工作，你其实不需要知道所有细节。

And I think we're going to start to see more and more of this. It's like if you know that it works correctly and efficiently, then you don't actually have to know all the details.

显然，软件工程师的生活已经发生了巨大变化。过去一两年里，这简直成了一项全新的工作。你觉得接下来哪个角色会受到 AI 最大的影响？是在技术领域内，比如产品经理、设计师，还是技术领域外？你觉得 AI 下一步会走向哪里？

Clearly, the life of a software engineer has changed dramatically. It's like a whole new job now as of the past year or two. What do you think is the next role that will be most impacted by AI within either within tech, like product managers, designers, or even outside tech? Just like where do you think AI is going next?

我认为很多与工程相邻的角色都会受到影响。所以是的，可能是产品经理、设计、数据科学。它会扩展到几乎所有能在电脑上完成的工作，因为模型会在这方面越来越好。而 co-work 产品是接触这个的第一种方式，但这只是第一个。我认为它把 AI 带到了智能体式 AI，让以前没用过的人开始第一次体验。一年前，技术工程领域没人真正知道什么是智能体，也没人用它，但现在这就是我们工作的方式。而当我今天看非技术工作，比如半技术的产品工作和数据科学之类，人们用的 AI 都是对话式的，比如聊天机器人，但没人真正用过智能体。‘智能体’这个词被到处乱用，被严重误用，已经失去了意义。但智能体其实有一个非常具体的技术含义：它是一个 AI，是一个能使用工具的 LLM。所以它不只是说话，它还能实际执行操作，与你的系统交互。这意味着它可以用你的 Google Docs，可以发邮件，可以在你的电脑上运行命令，做各种事情。所以我认为任何以这种方式使用电脑工具的工作，都会是下一个被影响的。这是我们必须作为社会、作为行业去解决的问题。对我来说，这也是为什么在 Anthropic 做这项工作感觉非常重要和紧迫的原因之一，因为我们非常认真地对待这件事。所以现在我们有了经济学家、政策专家、社会影响专家。我们想多讨论这件事，这样社会才能决定怎么做，因为这不应该是我们单方面决定的。

I think it's going to be a lot of the roles that are adjacent to engineering. So yeah, it could be product managers, it could be design, could be data science. It is going to expand to pretty much any kind of work that you can do on a computer because the model is just going to get better and better at this. And this is the co-work product is kind of the first way to get at this, but it's just the first one. And it's the thing that I think brings AI to an agentic AI to people that haven't really used it before. And people are starting to get a sense of it for the first time. When I think about tech engineering a year ago, no one really knew what an agent was, no one really used it, but nowadays it's just the way that we do our work. And then when I look at non-technical work today, like semi-technical product work and data science and things like this, when you look at the kinds of AI that people are using, it's all conversational AI, like a chatbot or whatever, but no one really has used an agent before. And this word agent just gets thrown around all the time and it's just so misused, it's like lost all meaning. But agent actually has a very specific technical meaning, which is it's an AI, it's an LLM that's able to use tools. So it doesn't just talk, it can actually act and interact with your system. This means it can use your Google Docs, it can send email, it can run commands on your computer and do all this kind of stuff. So I think any kind of job where you use computer tools in this way, this is going to be next. This is something we have to figure out as a society, this is something we have to figure out as an industry. And I think for me also this is one of the reasons it feels very important and urgent to do this work at Anthropic because I think we take this very, very seriously. And so now we have economists, we have policy folks, we have social impact folks. This is something we just want to talk about a lot, so as a society we can figure out what to do because it shouldn't be up to us.

所以大问题，你也在暗示，就是工作和失业之类的事情。有个概念叫杰文斯悖论，就是我们能做得更多，反而雇佣更多，实际上并不像看起来那么可怕。到目前为止，随着 AI 成为工程工作的重要部分，你经历了什么？你们招聘的人数是不是比没有 AI 时更多？对工作有什么看法？

So the big question, which you're kind of alluding to, is jobs and job loss and things like that. There's this concept of Jevons paradox of just as we can do more, we hire more and it's not actually as scary as it looks. What have you experienced so far with AI becoming a big part of the engineering job? Just are you hiring more than if you didn't have AI and just thoughts on jobs?

是的，我们团队在招人。Claude Code 团队正在招聘。如果你感兴趣，可以看看 Anthropic 的招聘页面。就我个人而言，所有这些都让我更享受工作。我从未像今天这样喜欢编程，因为我不需要处理那些琐碎细节。所以对我来说，这非常令人兴奋。我们从很多客户那里听到，他们喜欢这个工具，喜欢 Claude Code，因为它让编程再次变得愉快。这对他们来说非常有趣。但很难预测这会走向何方，我又得借助历史类比。我觉得印刷机就是一个很好的例子：原本只有少数人掌握的技术，比如读写能力，变得人人可及。这本质上是民主化的。每个人都开始能做这件事。如果不是这样，文艺复兴根本不可能发生，因为文艺复兴很大程度上是知识的传播，是人们用来交流的文字记录。当时没有电话，没有互联网。所以关键是，这接下来会催生什么？对我来说，这是非常乐观的版本，也是我真正兴奋的部分。这简直难以想象。如果没有印刷机，我们今天不可能交谈。我们的麦克风不会存在。我们周围的一切都不会存在。没有它，就不可能协调如此庞大的人群。所以我设想几年后的世界，每个人都能编程。那会解锁什么？任何人都可以随时构建软件。我不知道。就像在 15 世纪，没人能预测到今天的局面。我认为是一样的。但我也认为，与此同时，这会非常具有颠覆性，对很多人来说会很痛苦。再次强调，作为社会，我们必须进行这场对话，必须一起解决这个问题。

Yeah, I mean, for our team, we're hiring. So Claude Code team is hiring. If you're interested, just check out the jobs page on Anthropic. Personally, all this stuff has just made me enjoy my work more. I have never enjoyed coding as much as I do today because I don't have to deal with all the minutia. So for me personally, it's been quite exciting. This is something we hear from a lot of customers where they love the tool, they love Claude Code because it just makes coding delightful again. And that's just so fun for them. But it's hard to know where this thing is going to go and I again, I have to reach for these historical analogs. I think the printing press is just such a good one because what happened is this technology that was locked away to a small set of people, like knowing how to read and write, became accessible to everyone. It was just inherently democratizing. Everyone started to be able to do this. And if that wasn't the case, then something like the Renaissance just could never have happened because a lot of the Renaissance was about knowledge spreading, about written records that people used to communicate. There were no phones or anything like this, no internet at the time. So it's about what does this enable next? And I think that's the very optimistic version of it for me and that's the part that I'm really excited about. It's just unimaginable. We couldn't be talking today if the printing press hadn't been invented. Our microphones wouldn't exist. None of the things around us would exist. It just wouldn't be possible to coordinate such a large group of people if that wasn't the case. And so I imagine a world a few years in the future where everyone is able to program. And what does that unlock? Anyone can just build software anytime. And I have no idea. It's just the same way that in the 1400s, no one could have predicted this. I think it's the same way. But I do think in the meantime, it's going to be very disruptive and it's going to be painful for a lot of people. And again, as a society, this is a conversation that we have to have. And this is a thing that we have to figure out together.

那么对于听到这些、想要在这个即将到来的疯狂动荡中成功并立足的人，有什么建议吗？是去玩 AI 工具，熟练掌握最新东西吗？还有什么其他推荐来帮助人们保持领先吗？

So for folks hearing this that want to succeed and make it in this crazy turmoil we're entering, any advice? Is it play with the AI tools, get really proficient at the latest stuff? Is there anything else that you recommend to help people stay ahead?

是的，差不多就是这样。去实验这些工具，了解它们，不要害怕。直接投入进去，尝试它们，站在最前沿，站在边界上。

Yeah, I think that's pretty much it. Experiment with the tools, get to know them, don't be scared of them. Just dive in, try them, be on the bleeding edge, be on the frontier.

也许第二条建议是，试着比过去更成为一个通才。例如，在学校里，很多学计算机科学的人只学编程，不太学别的。可能学一点系统架构之类的。但我每天共事的一些最有效的工程师和产品经理，他们跨越多个学科。在 Claude Code 团队，每个人都写代码。我们的产品经理写代码，工程经理写代码，设计师写代码，财务人员写代码，数据科学家也写代码。团队里每个人都写代码。再看具体的工程师，人们常常跨学科。一些最强的工程师是产品与基础设施混合型工程师，或者产品工程师有很好的设计感，也能做设计。或者一个工程师对业务有很好的理解，能据此决定下一步做什么。或者一个工程师喜欢和用户交流，能真正传达用户的需求。所以，我认为未来几年最受回报的人，不只是 AI 原住民，不只是会用这些工具，还要有好奇心，是通才，跨越多个学科，能思考他们解决的更广泛问题，而不仅仅是工程部分。

Maybe the second piece of advice is try to be a generalist more than you have in the past. For example, in school, a lot of people that study CS, they learn to code and they don't really learn much else. Maybe they learn a little bit of systems architecture or something like this. But some of the most effective engineers that I work with every day and some of the most effective, you know, like product managers and so on, they cross over disciplines. So, on the Claude Code team, everyone codes. You know, our product manager codes, our engineering manager codes, our designer codes, our finance guy codes, our data scientist codes. Like everyone on the team codes. And then if I look at particular engineers, people often cross different disciplines. So, some of the strongest engineers are hybrid product and infrastructure engineers. Or product engineers with really great design sense and they're able to do design also. Or an engineer that has a really good sense of the business and can use that to figure out what to do next. Or an engineer that also loves talking to users and can just really channel what users want to figure out what's next. So, I think a lot of the people that will be rewarded most over the next few years, they won't just be AI native and they don't just know how to use these tools really well, but also they're curious and they're generalists and they cross over multiple disciplines and can think about the broader problem they're solving rather than just engineering part of it.

你觉得这三个独立的学科——工程、设计、产品管理——作为思考团队的方式仍然有用吗？即使他们现在都在写代码，都在为构建做贡献，你觉得这三个角色至少目前会长期存在吗？

Do you find these three separate disciplines still useful as a way to think about the team? They're, you know, engineering, design, product management. Do you find like those even though they are now coding and contributing to thinking about to build, do you feel like those are three roles that will persist long-term at least at this point?

我认为短期内会持续，但我们开始看到这些角色可能有 50% 的重叠，很多人实际上在做同样的事情，有些人有专长。比如，我写代码多一点，或者产品经理做更多协调、规划、预测之类的事情。

I think in the short term it will persist, but one thing that we're starting to see is there's maybe a 50% overlap in these roles where a lot of people are actually just doing the same thing and some people have specialties. For example, I code a little bit more over his cat or PM does a little bit more, you know, coordination or planning or you know, forecasting or things like this.

利益相关者对齐。

Stakeholder alignment.

利益相关者对齐。没错。我确实认为未来，到今年年底，我们会开始看到这些角色变得更加模糊。在某些地方，软件工程师的头衔会开始消失，被“构建者”取代，或者可能每个人都成为产品经理，每个人都写代码之类的。

Stakeholder alignment. Exactly. I do think that there is a future where I think by the end of the year what we're going to start to see is these start to get even murkier where I think in some places the title software engineer is going to start to go away and it's just going to be replaced by builder or maybe it's just everyone's going to be a product manager and everyone codes or something like this.

谁说招聘必须公平？最近和我聊过的每一位创始人和招聘经理都感受到同样的压力。尽快招到最好的人，但招聘耗时，对齐困难，优秀人才的竞争越来越激烈。这就是为什么像 11 Labs、Brex、Replit、Deel 以及 5000 多家其他组织使用 Metaview，这家 AI 公司为高绩效团队提供了真正的招聘不公平优势。他们提供一套 AI 智能体，像招聘同事一样工作。他们根据你的确切标准为你寻找候选人，自动记录面试笔记，收集整个招聘过程中的洞察，并帮助你识别 pipeline 中的最佳候选人。AI 处理招聘的苦差事，给你一个真实的数据源。这意味着每次招聘节省数小时，团队专注于最重要的事情：赢得合适的候选人。别让你的竞争对手在招聘上超过你。Metaview 的客户关闭职位的速度快 30%。立即免费试用 Metaview，并在 metaview.ai/lenny 获得额外一个月的 sourcing。网址是 m e t a view.ai/lenny。

Who says hiring has to be fair? Every founder and hiring manager I've been speaking with these days is feeling the same pressure. Hire the best people as fast as possible, but recruiting is time consuming, alignment is hard, and competition for great talent keeps getting tighter. That's why teams like 11 Labs, Brex, Replit, Deel, and 5,000 other organizations use Metaview, the AI company giving high-performance teams a real unfair advantage in hiring. They give you a suite of AI agents that behave like recruiting co-workers. They find candidates for you based on your exact criteria, take interview notes automatically, gather insights across your hiring process, and help you identify the best candidates in your pipeline. AI handles the recruiting toil and gives you a real source of truth. That means hours saved per hire and a team focused on what matters most, winning the right candidates. Don't let your competitors out hire you. Metaview customers close roles 30% faster. Try Metaview today for free and get an extra month of sourcing at metaview.ai/lenny. That's m e t a view.ai/lenny.

你谈到你更享受写代码了。我实际上在 Twitter 上做了一个小型的非正式调查。不知道你看到没有，我做了三个不同的投票。我问工程师，自从使用 AI 工具后，你更享受工作还是更不享受？然后我分别问了产品经理和设计师。工程师和产品经理中，70% 的人说更享受工作，大约 10% 说更不享受。有趣的是，设计师中只有 55% 说更享受，20% 说更不享受。我觉得这很有意思。

You talked about how you're enjoying coding more. I actually did this little informal survey on Twitter. I don't know if you saw this where I just asked I did three different polls. I asked engineers, are you enjoying your job more or less since adopting AI tools? And then I did a separate one for PMs and one for designers. And both engineers and PMs, 70% of people said they're enjoying their job more. And about 10% said they're enjoying their job less. Designers interestingly, only 55% said they're enjoying their job more and 20% said they're enjoying their job less. Thought that was really interesting.

这非常有趣。我很想和这些人聊聊，无论是更享受的还是更不享受的，去理解原因。

That's super interesting. I'd love to talk to these people. You know, both in the more bucket and the less bucket just to understand.

你跟进过他们中的任何人吗？

Did you get to follow up with any of them?

有几个人回复了，我们实际上在做后续投票，会链接到节目笔记中，深入探讨这些内容。但有很多因素让工作更有趣或更无趣。设计师们实际上没有分享太多，那些被问到为什么更不享受工作的人，我没听到太多。所以我很好奇发生了什么。

A few people replied and we're actually doing a follow-up poll that we'll link to in the show notes of going deeper into some of this stuff. But a lot of there's like, you know, the factors that make it more fun and less fun. The designers, they didn't share a lot actually of just like the people that are actually asked just like why are you enjoying your job less? I didn't hear a lot. So I'm curious what's going on there.

是的，我在 Anthropic 也看到了一些。我认为每个人都很技术化。这是我们筛选的条件，当人们加入时。即使是非技术岗位，也要经过很多技术面试。我们的设计师实际上也写代码。所以我认为对他们来说，这是他们享受的事情，因为现在他们可以直接写代码，而不是依赖有 bug 的工程师，甚至一些以前不写代码的设计师也开始写了，这对他们来说很好，因为他们可以解放自己。但我真的很想听听更多人的经历，因为我打赌不是所有人都这样。

Yeah, I'm seeing this a little bit with Anthropic. I think everyone is fairly technical. This is something that we screen for, you know, when people join. We have a lot of technical interviews that people go through even for non-technical functions. And you know, our designers virtually code. So I think for them this is something that they have enjoyed from what I've seen because now instead of buggy engineers they can just like go in and code and even some designers that didn't code before have just started to do it and for them it's great because they can unlock themselves. But I'd be really interested just to hear more people's experiences because I bet it's not uniform like that.

是的，所以如果你在听这个，如果你觉得工作没那么有趣，更不享受，请留言评论。因为你说的话和我听到的大多数人——70% 的产品经理和工程师——更热爱他们的工作。如果你不在那个群体里，可能有些问题。

Yeah, so maybe if you're listening to this leave a comment if you're finding your jobs less fun and you're enjoying your job less because what you're saying and what I'm hearing from most people 70% of PMs and engineers are loving their job more. That's like if you're not on a bucket you can something's going on.

是的，是的。我们确实看到人们使用不同的工具。例如，我们的设计师更多地使用云桌面应用来写代码。你只需下载桌面应用，有一个代码标签，就在代码工作旁边，实际上和云代码是一样的，同一个智能体等等。我们已经有了这个功能好几个月了。所以你可以用它来写代码，而不必打开一堆终端。但你仍然拥有云代码的能力，最重要的是，你可以并行运行任意多个云会话。我们称之为多云化。这对非工程师的人来说更自然，这又回到了把产品带到人们所在的地方。你不想让人们使用不同的工作流程。

Yeah, yeah. We do see that people use also different tools. So for example our designers they use the Claude Desktop app a lot more to do their coding. So you just download the desktop app there's a code tab it's right next to code work and it's actually the same as that Claude Code so it's like the same agent and everything. We've had this for you know for many many months. And so you can use this to code in a way that you don't have to open a bunch of terminals. But you still get the power of Claude Code and the biggest thing is you can just run as many cloud sessions in parallel as you want. We call this multi clouding. So this is a little more native I think for folks that are not engineers and really this is back to bringing the product to where the people are. You don't want to make people use a different workflow.

你不想让他们费劲去学新东西。无论人们在做什么，如果你能让它变得稍微容易一点，那就会是一个更好的产品，人们会更喜欢。这就是潜在需求原则，我认为这是产品中最重要的原则。你能谈谈这个吗？因为我正想聊这个，解释一下这个原则是什么，以及当你释放这种潜在需求时会发生什么？

You don't want to make them go out of their way to learn a new thing. It's whatever people are doing if you can make that a little bit easier then that's just going to be a much better product that people enjoy more and this is just this principle of latent demand which I think is just the single most important principle in product. Can you talk about that actually cuz I was going to go there explain what this principle is and and and just what happens when you unlock this latent demand?

潜在需求是指，如果你以一种可以被用户“破解”或“误用”的方式来构建产品，让他们用它来做一些你原本没设计但对他们有用的事，那么这能帮助你作为产品构建者了解下一步该往哪走。举个例子，Facebook Marketplace。团队经理 Fiona，她实际上是 Marketplace 团队的创始经理，她经常讲这个。Facebook Marketplace 的起源是基于一个观察，大概在 2016 年左右，Facebook 群组中 40%的帖子都是买卖东西。这太疯狂了。人们是在滥用 Facebook 群组这个产品来做买卖，这不是安全意义上的滥用，而是没人设计这个产品来做这个，但他们就是想办法用上了，因为它实在太有用了。所以很明显，如果你建一个更好的产品让人们买卖东西，他们会喜欢的。Marketplace 会火是显而易见的。所以第一步是买卖群组，一种专门用来做这个的群组，第二步就是 Marketplace。Facebook Dating 我觉得也是从类似的地方开始的。观察是，如果你看 Facebook 上的个人主页浏览，60%的浏览来自不是朋友、且性别不同的人。这就像传统的约会场景，但人们就是在互相“窥视”。所以如果你能为此建一个产品，可能会有效。潜在需求这个概念非常强大。比如，Cower 也是这么来的。我们看到过去六个月左右，很多用 Claude Code 的人并不是在写代码。有人在 Twitter 上用它种番茄，有人用它分析基因组，有人用它从损坏的硬盘里恢复照片，比如婚礼照片，还有人用它分析 MRI。这些都是完全非技术性的用例。很明显，人们费尽周折用终端来做这些事。也许我们应该直接为他们建一个产品。我们很早就看到了这一点。大概去年五月，我走进办公室，看到我们的数据科学家 Brendan 的电脑上开着 Claude Code，他开着终端。我很震惊，我说：“Brendan，你在干嘛？你居然学会了打开终端，这可是一个非常工程化的产品。很多工程师都不愿意用终端。这是最低级的工作方式，非常底层。”他学会了用终端，下载了 Node.js，下载了 Claude Code，然后在终端里做 SQL 分析。这太疯狂了。然后下一周，所有数据科学家都开始做同样的事。所以当你看到人们以这种方式“滥用”产品，用它来做对他们有用但并非设计初衷的事，这是一个非常强烈的信号，你应该为此建一个产品，人们会喜欢的。一个专门用途的产品。现在我觉得潜在需求还有第二个有趣的维度。传统的框架是看人们在做什么，让它更容易，赋能他们。过去六个月我看到的现代框架有点不同：看模型想做什么，让它更容易。当我们刚开始构建 Claude Code 时，很多人用大语言模型设计产品的方式是把模型放在一个盒子里。他们说：这是我想构建的应用，我想让它做这个。模型，你来做这个组件。这是你与这些工具和 API 交互的方式。而对于 Claude Code，我们反过来了。我们说产品就是模型。我们想暴露它，给它最少的脚手架，最少的工具集。这样它就能做事了，它可以决定运行哪些工具，以什么顺序运行等等。我认为这很大程度上是基于模型想做什么的潜在需求。在研究里，我们称之为“在分布上”。你想看模型试图做什么。在产品术语里，潜在需求就是同一个概念，但应用在模型上。

Latent demand is this idea that if you build a product in a way that can be hacked or can be kind of mis used by people in a way it wasn't really designed for it to do kind of something that they want to do then this helps you as the product builder learn where to take the product next. So an example of this is Facebook Marketplace. So, the the manager for the team, Fiona, she she was actually the founding manager for the Marketplace team, and she talks about this a lot. Facebook Marketplace it started based on the observation back in this must have been like 20 2016 or or something like this, that 40% of posts in Facebook groups are buying and selling stuff. So, this is crazy. It's like people are abusing the Facebook groups product to buy and sell, and it's not it's not abuse in kind of like a security sense, it's abuse in that no one designed the product for this, but they're kind of figuring it out because it is just so useful for this. And so, it was pretty obvious if you build a better product to let people buy and sell, they're going to like it. And it was just very obvious that Marketplace would be a hit from this. And so, the first thing was buy and sell groups, so kind of special purpose groups to let people do that, and the second product was Marketplace. Facebook Dating I think started in a pretty similar place. And I think that would the observation was if you look at people looking at if you look at profile views, so people looking at each other's profiles on Facebook, 60% of profile views were people that are not friends with each other, that are opposite gender. And so, this is this kind of like, you know, like traditional kind of date dating setup, but you know, people are just like creeping on each other. So, maybe if you can build a product for this, it's, you know, it it might work. And so, this idea of latent demand, I think it's just so powerful. And for example, this is also where Cower came from. We saw that for the last 6 months or so, a lot of people using Claude Code were not using it to code. There was someone on Twitter that it to grow tomato plants, there was someone else using it to analyze their genome. Someone was using it to recover photos from a corrupted hard drive, it was like wedding photos. There was someone that was using it for I think like they were using it to analyze an MRI. So, there's just all these different use cases that are not technical at all. And it was just really obvious like people are jumping through hoops to use a terminal to do this thing. Maybe we should just build a product for them. And we saw this actually pretty early. Back in maybe May of last year, I remember walking into the office and our data scientist, Brendan, was had a Claude Code on his uh computer. He just had a terminal up. And I was like I was shocked. I was like, "Brendan, what are what are you doing? Like you you figured out how to open the terminal, which is you know, it's it's a very engineering product. Even a lot of engineers don't want to use a terminal. It's just like a it's like just like the lowest level way to to do your work. Um just really, really uh kind of in the weeds of the computer." And so he figured out how to use the terminal. He downloaded Node.js. He downloaded Claude Code. And he was doing SQL analysis in the terminal. And it was it was crazy. And then the next week all the data scientists were doing the same thing. So when you see people abusing the product in this way, using it in a way that it wasn't designed in order to do something that is useful for them, it's just such a strong indicator that you should just build a product and and people are going to like that. It's something that's special purpose for that. I think now there's there's also this kind of interesting second dimension to latent demand. This is sort of the traditional framing is look at what people are doing, make that a little bit easier, empower them. The modern framing that I've been seeing in the last 6 months is a little bit different and it's look at what the model is trying to do and make that a little bit easier. And so when we first started building Claude Code, I think a lot of the way that people approached designing things with LLMs is they kind of put the model in a box. And they were here's this application that I want to build. I want it to do. Model, you're going to do this one component of it. Here's the way that you're going to interact with these tools and APIs and whatever. And for Claude Code, we inverted that. We said the product is the model. We want to expose it. We want to put the minimal scaffolding around it, give it the minimal set of tools. So it can do the things. It can decide which tools to run it can decide in what order to run them in and so on. And I I think a lot of this was just based on kind of latent demand of what the model wanted to do. And so in research, we call this being on distribution. Uh you want to see like what the model is trying to do. In product terms, latent demand is just the same exact concept, but applied to the model.

你提到了 Claude Code。我记得你之前说过，你的团队在 10 天内就建成了它。这太疯狂了。它很快就有了数百万用户。10 天建成这样的东西，有什么故事吗？除了“我们用 Claude Code 建了它”之外？是的，很有趣。Claude Code，就像我说的，发布时并没有立刻火起来。它是随着时间的推移才火起来的，有几个转折点。一个是 Opus 4，它真的引爆了。然后 11 月又引爆了。它一直在引爆，增长曲线每天都在变得更陡。但头几个月它并不火。人们用它，但很多人不知道怎么用，不知道它是干什么的。模型本身也不是很好。Claude Code 发布时立刻就火了，比早期的 Claude Code 火得多。说实话，很多功劳要归功于 Felix、Sam、Jenny 和整个团队。他们是一个非常强大的团队。再次强调，Claude Code 的起源就是潜在需求。我们看到人们用 Claude Code 做那些非技术性的事情。

You talked about Claude Code. Something that I saw you talk about when you launched that initially is you your team built that in 10 days. That's insane. Uh I think that it came out I think it was like, you know, used by millions of people pretty quickly. Something like that being built in 10 days. Uh anything there any stories there other than just it was just, you know, we used Claude Code to build it. That's it. Yeah, it it it It's funny. Uh Claude Code, like I said, when we released it, it was not immediately a hit. It became a hit over time and there was a few inflection points. So, one was, you know, like Opus 4. Uh it just really really inflected. And then in November, it inflected. And it just keeps inflecting. It The growth just keeps getting steeper and steeper and steeper every day. But, you know, for the first few months, it wasn't a hit. Uh people used it, but a lot of people couldn't figure out how to use it. They didn't know what it was for. The model still like wasn't very good. Claude Code, when we released it, it was just immediately a hit. Much more so than Claude Code it was early on. I think a lot of the credit, honestly, just goes to like Felix and and Sam and the and Jenny and the the team that built this. It's just an incredibly strong team. And again, the the place Claude Code came from is just this latent demand. Like, we saw people using Claude Code for these non-technical things.

我们当时在想办法。团队探索了几个月，尝试了各种不同的方案。最后有人说：‘要不我们把 Claude Code 放到桌面应用里？’结果这招还真管用。于是，他们花了 10 天时间，完全用 Claude Code 把它搭建起来。Claude Code 内置了非常精密的安全系统，这些护栏能确保模型做正确的事，不会失控。比如，我们随它一起发布了一整套虚拟机。所有这些代码都是 Claude Code 写的。我们只需要考虑如何让它更安全、更易用，让非工程师也能上手。整个功能完全用 Claude Code 实现，花了大约 10 天。我们提前发布了，当时还很粗糙，现在也还有些粗糙。但这就是我们学习的方式——无论是产品还是安全方面——我们必须比预想更早地发布，这样才能获得反馈，与用户交流，了解人们想要什么，从而塑造产品的未来方向。

And we're trying to figure out what do we do. So, for a few months, the team was exploring. They were trying all sorts of different options. And in the end, someone was just like, 'Okay, what if we just take Claude Code and put it in the desktop app?' And that's essentially the thing that worked. So, over 10 days, they just completely used Claude Code to build it. Claude Code actually has a very sophisticated security system built in. Essentially, these guardrails make sure the model does the right thing and doesn't go off the rails. For example, we ship an entire virtual machine with it. And Claude Code just wrote all of this code. So we just have to think about how to make this a little safer, a little more self-guided for people who are not engineers. It was fully implemented with Claude Code. Took about 10 days. We launched it early. It was still pretty rough and it's still pretty rough around the edges. But this is the way we learn, both on the product side and on the safety side: we have to release things a little earlier than we think so that we can get feedback, talk to users, understand what people want, and that will shape where the product goes in the future.

对，我觉得这一点非常有趣，也很独特。一直有这种理念：尽早发布，从用户那里学习，获取反馈，迭代。但连 AI 能做什么、人们会怎么用它都很难预料，这本身就是一个独特的理由去尽早发布。就像你刚才说的，这能帮你发现我们之前不知道的潜在需求——把它放出去，看看人们会怎么用它。

Yeah, I think that point is so interesting and it's so unique. There's always been this idea: release early, learn from users, get feedback, iterate. The fact that it's hard to even know what the AI is capable of and how people will try to use it is a unique reason to start releasing things early. That will help you, as you exactly describe, this idea of what is the latent demand in this thing that we didn't really know. Let's put it out there and see what people do with it.

对，在 Anthropic 这个安全实验室里，另一个维度就是安全。说到模型安全，有很多不同的研究方法。最底层是对齐和机制可解释性。我们在训练模型时，要确保它是安全的。目前我们已经有了相当先进的技术，可以理解神经元里发生了什么，追踪它们。比如，如果有一个与欺骗相关的神经元，我们开始能够监控它，知道它被激活了。这就是对齐、机制可解释性，是最底层。第二层是评估。这基本上是一个实验室环境，模型在培养皿里，你研究它。你把它放在一个合成情境中，然后问：‘模型，你做什么？你做对了吗？对齐了吗？安全吗？’第三层是观察模型在真实世界中的行为。随着模型越来越复杂，这一层变得非常重要，因为模型在前两层可能表现很好，但在第三层可能就不那么好了。

Yeah, and at Anthropic as a safety lab, the other dimension of that is safety. When you think about model safety, there are a bunch of different ways to study it. The lowest level is alignment and mechanistic interpretability. This is when we train the model, we want to make sure it's safe. We at this point have pretty sophisticated technology to understand what's happening in the neurons, to trace it. For example, if there's a neuron related to deception, we can start to get to the point where we can monitor it and understand that it's activating. This is alignment, mechanistic interpretability. It's like the lowest layer. The second layer is evals. This is essentially a laboratory setting. The model is in a petri dish and you study it. You put it in a synthetic situation and just say, 'Okay, model, what do you do? Are you doing the right thing? Is it aligned? Is it safe?' And then the third layer is seeing how the model behaves in the wild. As the model gets more sophisticated, this becomes so important because it might look very good on these first two layers, but not great on the third one.

你们工作的领域真是疯狂，竞争激烈、节奏飞快。同时，又担心一旦搞砸，那个‘神’可能会逃逸并造成破坏。找到这个平衡一定非常困难。我听到的是这三个层次。我知道这可以单独做一期播客，聊聊你们怎么考虑安全。但我听到的是你们用了这三层：观察模型的思考和运作，测试和评估来发现它做坏事，然后尽早发布。第一层我之前没怎么听说过，太酷了。所以你们有一个可观测性工具，可以窥视模型的大脑，看它怎么思考、往哪个方向走。

It's such a wild space that you work in where there's this insane competition and pace. At the same time, there's this fear that if you get it wrong, the god can escape and cause damage. Finding that balance must be so challenging. What I'm hearing is there's these three layers. I know this could be a whole podcast conversation about how you all think about the safety piece. But just what I'm hearing is there's these three layers you work with: observing the model thinking and operating, tests and evals that tell you it's doing bad things, and then releasing it early. I haven't actually heard a ton about that first piece. That is so cool. So you guys have an observability tool that can let you peek inside the model's brain and see how it's thinking and where it's heading.

对，你应该找机会请 Chris Olah 上播客，他是这个领域的专家。他开创了机制可解释性这个领域。思路是：大脑是什么？就是一堆相互连接的神经元。在人类或动物大脑中，你可以从机制层面研究神经元在做什么。令人惊讶的是，很多方法也适用于模型。模型神经元和动物神经元不一样，但在很多方面行为相似。我们已经学到了很多关于这些神经元如何工作的知识，比如这一层或这个神经元映射到哪个概念，特定概念如何编码，模型如何做规划，如何提前思考。很久以前，我们不确定模型只是预测下一个词，还是做了更深层的事情。现在我认为有相当强的证据表明它确实在做更深层的事情。支持它做这些的结构现在已经相当复杂了。随着模型变大，不再是一个神经元对应一个概念。单个神经元可能对应十几个概念。当它与其他神经元一起被激活时，这叫做叠加。它们共同代表更复杂的概念。我们一直在学习这些东西。

Yeah, you should at some point have Chris Olah on the podcast because he is the industry expert on this. He invented this field of mechanistic interpretability. The idea is, like, what is your brain? It's a bunch of neurons that are connected. In a human brain or an animal brain, you can study it at this mechanistic level to understand what the neurons are doing. It turns out surprisingly a lot of this does translate to models also. Model neurons are not the same as animal neurons, but they behave similarly in a lot of ways. We've been able to learn a ton about the way these neurons work, about how this layer or this neuron maps to this concept, how particular concepts are encoded, how the model does planning, how it thinks ahead. A long time ago we weren't sure if the model was just predicting the next token or doing something a little deeper. Now I think there's actually quite strong evidence that it is doing something a little deeper. The structures that allow it to do this are pretty sophisticated now. As the models get bigger, it's not just a single neuron that corresponds to a concept. A single neuron might correspond to a dozen concepts. And if it's activated together with other neurons, this is called superposition. Together it represents this more sophisticated concept. It is just something we're learning about all the time.

我们很早就发布了 Claude Code，因为我们想研究安全性。我们在 Anthropic 内部用了大概四五个月才发布，因为我们不太确定。这是当时大家发布的第一个智能体，也是第一个被广泛使用的编程智能体。我们不确定它是否安全。我们在内部研究了很长时间，才觉得放心。从那以后，我们学到了很多关于对齐和安全的知识，并反馈到了模型和产品中。Claude 的工作也类似。模型处于新环境中，执行非工程任务。它是一个代表你行动的智能体。在对齐和评估上表现良好。我们在内部试了，效果不错。我们和几个客户试了，效果也不错。现在我们需要确保它在真实世界中是安全的。这就是为什么我们提前发布，并称之为研究预览。它在不断改进。这确实是确保模型长期对齐并做正确事情的唯一方法。

We released Claude Code really early because we wanted to study safety. We actually used it within Anthropic for 4 or 5 months before we released it because we weren't really sure. This is the first agent that I think folks had released at that point. It was definitely the first coding agent that became broadly used. We weren't sure if it was safe. We had to study it internally for a long time before we felt good about that. Even since, there's a lot that we've learned about alignment and safety that we've been able to put back into the model and the product. For Claude work it's pretty similar. The model is in this new setting, doing tasks that are not engineering tasks. It's an agent acting on your behalf. It looks good on alignment, it looks good on evals. We tried it internally, it looks good. We tried it with a few customers, it looks good. Now we have to make sure it's safe in the real world. That's why we released a little early, that's why we call it a research preview. It's constantly improving. This is really the only way to make sure that over the long term the model is aligned and doing the right things.

对于 Anthropic 来说，当我们思考这个领域的发展方式时，以安全和对世界有益的方式来做这件事就是我们存在的理由。Anthropic 的每个人都是为此而来。我们开源了很多这项工作，并自由地发表，以激励其他实验室安全地做事。我们称之为“竞相向上”。对于 Claude Code，我们发布了一个开源沙箱，可以在有边界的情况下运行智能体，而且它适用于任何智能体，不仅仅是 Claude Code。我们希望让其他人也能轻松做到同样的事情。

For Anthropic, as we think about the way this space evolves, doing this in a way that is safe and good for the world is the reason we exist. Everyone at Anthropic is here for that reason. We open source a lot of this work and publish it freely to inspire other labs to do things safely. We call this the race to the top. For Claude Code, we released an open source sandbox that runs the agent with boundaries, and it works with any agent, not just Claude Code. We want to make it easy for others to do the same.

我注意到工程师和产品经理在智能体不工作时会感到焦虑——好像他们在损失生产力。你有这种感觉吗？

I've noticed anxiety among engineers and product managers when agents aren't working—like they're losing productivity. Do you feel that?

我总是同时运行一堆智能体。现在我有五个在跑。醒来时我就启动一堆。今天第一件事就是打开手机，用 Claude iOS 应用的代码标签，让智能体做点什么，因为我昨天写了代码，有点怀疑自己。结果是对的。现在太容易了。我不怎么焦虑，因为智能体一直在运行。我不再被终端束缚了。大概三分之一的代码在终端，三分之一用桌面应用，还有三分之一在 iOS 应用上——这很令人惊讶，因为我没想到即使在 2026 年我也会这样编程。

I always have a bunch of agents running. Right now I have five agents running. When I wake up, I start a bunch. The first thing I did today was open my phone, Claude iOS app, code tab, and tell the agent to do something because I wrote code yesterday and was double-guessing myself. It was correct. It's so easy now. I don't feel much anxiety because I have agents running all the time. I'm not locked into a terminal anymore. Maybe a third of my code is in the terminal, a third using the desktop app, and a third on the iOS app—which is surprising because I didn't think I'd code that way even in 2026.

你把和 Claude Code 对话描述为编程。现在的编程就是描述你想要什么，而不是写代码。我想知道打孔卡程序员会怎么想。

You describe talking to Claude Code as coding. Coding now is describing what you want, not writing code. I wonder what punch-card programmers would think.

我记得读过早期的 ACM 杂志，人们说‘不，那不是真正的编程’。他们称之为程序设计。编程是个较新的词。我家来自苏联；我出生在乌克兰。我祖父是那里最早的程序员之一，用打孔卡。他把成堆的卡片带回家，我妈妈会在上面画画。他从未看到软件转型。我认为老一辈程序员不把软件当回事。他们会说这不是真正的编程。但这个领域一直在变化。

I remember reading early ACM magazines where people said, 'No, that's not really coding.' They called it programming. Coding is a newer word. My family is from the Soviet Union; I was born in Ukraine. My grandfather was one of the first programmers there, using punch cards. He brought stacks home, and my mom would draw on them. He never saw the software transition. I think older programmers didn't take software seriously. They'd say it's not real coding. But this field has always changed.

我也出生在乌克兰，来自敖德萨。

I was born in Ukraine too, from Odessa.

我也是！太巧了。也许我们有亲戚关系。你什么时候离开的？

Me too! That's crazy. Maybe we're related. When did you leave?

我们 95 年来的。

We came in '95.

我们 88 年离开的。留下来会是多么不同的生活。我每天都为能在这里长大感到幸运。我家用伏特加为美国干杯。

We left in '88. What a different life it would have been to stay. I feel lucky every day to grow up here. My family toasts to America with vodka.

同样的干杯，还是伏特加。

Same toast, still vodka.

你分享了一些技巧，比如给团队很多 token，以及朝着模型发展的方向构建。你还有什么其他建议？

You shared tips like giving your team many tokens and building towards where the model is going. What other advice do you have?

不要试图限制模型。很多人试图让模型以非常特定的方式行事，比如叠加严格的工作流。但通过给模型工具和目标，让它自己解决，你会得到更好的结果。一年前你可能需要脚手架，但现在不需要了。不要问模型能为你做什么——想想如何给它工具去做事。不要过度策划。

Don't try to box the model in. Many people try to make it behave a very particular way, like layering strict workflows. But you get better results by giving the model tools and a goal, and letting it figure it out. A year ago you needed scaffolding, but now you don't. Ask not what the model can do for you—think about how to give it tools to do things. Don't over-curate.

不要试图把它装进盒子里。不要试图一开始就给它一堆上下文。给它一个工具，让它自己获取所需的上下文，这样效果会更好。第二个原则可能更通用，就是‘苦涩的教训’。对于 Claude Code 团队，希望听众读过 Rich Sutton 大约 10 年前写的博客文章《苦涩的教训》。这个想法很简单：越通用的模型总是胜过越专用的模型。他当时说的是自动驾驶等领域，但有很多推论。对我来说，最大的推论就是永远押注更通用的模型。长期来看，不要用小型模型，不要微调，不要做这些事。虽然有些情况需要这么做，但如果你有灵活性，几乎总是押注更通用的模型。这些工作流本质上是围绕模型的脚手架，脚手架可能提升 10-20%的性能，但这些收益往往被下一代模型抹平。所以，几乎最好直接等下一代。最后一个原则是 Claude Code 事后看来做对的事：从一开始，我们就押注为 6 个月后的模型构建，而不是为今天的模型。在早期版本中，我几乎不写自己的代码，因为我不信任它。模型当时编码能力还不行。所以那时模型自动化了一些事情，但没有做太多编码工作。我们的赌注是，总有一天模型会足够好，能写大量代码。我们第一次看到这一点是在 Opus 4 和 Sonnet 4 上。Opus 4 是我们 5 月发布的第一个 ASL 3 级模型，我们看到一个转折点，所有人都开始使用 Claude Code，我们的增长呈指数级。我给创业者的建议是：前 6 个月会很痛苦，因为产品市场契合度不好，但如果你为 6 个月后的模型构建，当那个模型出来时，你会立即进入状态。

Don't try to put it into a box. Don't try to give it a bunch of context up front. Give it a tool so that it can get the context it needs. You're just going to get better results. I think a second one is maybe a more general version of this principle: the bitter lesson. For the Claude Code team, hopefully listeners have read Rich Sutton's blog post from maybe 10 years ago called 'The Bitter Lesson.' It's a really simple idea: the more general model will always outperform the more specific model. He was talking about self-driving cars and other domains, but there are many corollaries. For me, the biggest one is always bet on the more general model. Over the long term, don't try to use tiny models, don't try to fine-tune, don't try to do any of this stuff. There are some reasons to do it, but almost always bet on the more general model if you have that flexibility. These workflows are essentially scaffolding around a model, and in general, scaffolding might improve performance by 10-20%, but often these gains get wiped out with the next model. So it's almost better to just wait for the next one. A final principle that Claude Code got right in hindsight: from the very beginning, we bet on building for the model 6 months from now, not for the model of today. For the very early versions of the product, I wrote very little of my code because I didn't trust it. The models just weren't very good at coding yet. So back then, the model automated some things but wasn't doing a huge amount of my coding. The bet was that at some point the model gets good enough to write a lot of the code. We first saw this with Opus 4 and Sonnet 4. Opus 4 was our first ASL 3 class model released in May, and we saw an inflection where everyone started using Claude Code for the first time, and our growth went exponential. I give this advice to folks building startups: it's going to be uncomfortable because your product-market fit won't be great for the first 6 months, but if you build for the model 6 months out, when that model comes out, you'll hit the ground running.

你说为 6 个月后的模型构建，你认为人们可以假设会发生什么？只是它会普遍变得更好吗？比如，它几乎够好了，这就是它可能会更好的信号？有什么建议吗？

And when you say build for the model 6 months out, what do you think people can assume will happen? Is it just that it will generally get better at things? Like, it's almost good enough, and that's a sign it'll probably get better? Any advice there?

我觉得这是个好方法。当然，在 AI 实验室里，我们能看到它具体在哪些方面变好，所以有点不公平。但我们尽量公开讨论。一个它会变好的方面是使用工具和计算机。这是我愿意下的赌注。另一个是它会越来越擅长长时间运行。有相关研究。如果你追踪轨迹，或者从我自己的经验来看，一年前我用 Sonnet 3.5 时，它可能运行 15 到 30 秒就开始出问题，你得手把手教。现在用 Opus 4.6，平均能无人值守运行 10、20、30 分钟，我会再开一个 Claude 让它做别的事。我总是同时跑很多个 Claude。它们能运行几小时甚至几天。有些例子中它们运行了好几周。随着时间的推移，这会越来越正常：模型长时间运行，不需要人盯着。

I think that's a good way to do it. Obviously, within an AI lab, we get to see the specific ways it gets better, so it's a little unfair. But we try to talk about this. One way it's going to get better is using tools and computers. This is a bet I would make. Another is it's going to get better at running for long periods of time. There are studies about this. If you trace the trajectory, or from my own experience, when I used Sonnet 3.5 a year ago, it could run for maybe 15 or 30 seconds before going off the rails, and you had to hold its hand. Nowadays with Opus 4.6, on average it'll run 10, 20, 30 minutes unattended, and I'll start another Claude and have it do something else. I always have a bunch of Quads running. They can run for hours or even days. There are examples where they ran for many weeks. Over time, this will become more normal: models running for very long periods without needing babysitting.

你刚讲了构建 AI 产品的技巧。对于第一次使用 Claude Code 的人，或者已经在用但想用得更好的人，有什么建议？分享几个专业技巧吧。

So you just talked about tips for building AI products. Any tips for someone just using Claude Code for the first time, or someone already using it that wants to get better? What are a couple pro tips?

我先说一句：使用 Claude Code 没有唯一正确的方法。我可以分享一些技巧，但说实话，这是一个开发工具。开发者各不相同，有不同的偏好和环境。没有唯一正确的方法。你得找到自己的路。幸运的是，你可以问 Claude Code；它能给出建议，编辑你的设置，并帮助你。我觉得有用的几个技巧：第一，使用最强大的模型。目前是 Opus 4.6。我总是开启最大努力模式。有时人们会尝试更便宜的模型比如 Sonnet，但因为不够智能，完成同样的任务反而需要更多 token。所以它不一定更便宜。通常，使用最强大的模型实际上更便宜、token 消耗更少，因为它能更快地完成任务，需要的纠正和指导更少。所以第一个技巧：用最好的模型。第二：使用计划模式。我几乎 80%的任务都从计划模式开始。计划模式很简单：我们在模型提示中注入一句话‘请先不要写任何代码。’就这些。没什么花哨的。对于终端用户，按两次 shift+tab。桌面应用和网页上都有按钮。移动端也即将推出。

I'll give a caveat: there's no one right way to use Claude Code. I can share some tips, but honestly, this is a dev tool. Developers are all different, with different preferences and environments. There's no one right way. You have to find your own path. Luckily, you can ask Claude Code; it can make recommendations, edit your settings, and help with that. A few tips I find useful: number one, use the most capable model. Currently that's Opus 4.6. I always have maximum effort enabled. Sometimes people try a less expensive model like Sonnet, but because it's less intelligent, it actually takes more tokens to do the same task. So it's not obvious that it's cheaper. Often, using the most capable model is actually cheaper and less token-intensive because it can do the same thing much faster with less correction and hand-holding. So first tip: use the best model. Second: use plan mode. I start almost all my tasks in plan mode, maybe 80%. Plan mode is simple: we inject one sentence into the model's prompt saying 'Please don't write any code yet.' That's it. Nothing fancy. For terminal users, it's shift+tab twice. For desktop app, there's a button. On web, there's a button. Coming soon to mobile too.

嗯，我们刚刚也推出了 Slack 集成。所以，计划模式是第二个。本质上，模型会和你来回交流。一旦计划看起来没问题，你就让模型执行。之后我会自动接受编辑，因为如果计划没问题，它就会一次性搞定。几乎每次用 Opus 4.6 都能第一次就做对。然后第三个建议可能就是多试试不同的界面。我觉得很多人想到 Claude Code 时，就会想到终端。当然，我们支持所有终端。我们支持 Mac、Windows，不管什么终端，都能完美运行。但我们也支持很多其他形式。我们有 iOS 和 Android 应用，有桌面应用，还有 Slack 集成。我们支持各种各样的东西。所以，我建议你多试试这些。再说一次，每个工程师都不一样。每个做开发的人都不一样。找到适合你的方式，然后用它。你不一定非要用终端。同样的 Claude 智能体在到处运行。

Uh and we just launched it for the Slack integration, too. So, plan mode is the second one. And essentially, the model will just go back and forth with you. Once the plan works good, then you let the model execute. I auto-accept edits after that because if the plan works good, it's just going to one-shot it. It'll get it right the first time almost every time with Opus 4.6. And then maybe the third tip is just play around with different interfaces. I think a lot of people when they think about Claude Code, they think about a terminal. And of course, we support every terminal. We support Mac, Windows, whatever terminal you might use, it works perfectly. But we actually support a lot of other form factors, too. We have iOS and Android apps, we have a desktop app. There's the Slack integration. There's all sorts of things that we support. So, I would just like play around with these. And again, it's like every engineer is different. Everyone that's building is different. Just find the thing that feels right to you and use that. You don't have to use the terminal. It's the same Claude agent running everywhere.

太棒了。好的，还有几个问题来收尾。你对 Codex 怎么看？你觉得那个产品怎么样？你觉得他们发展方向如何，以及在这个竞争激烈的编程智能体领域竞争的感觉如何？

Amazing. Okay, just a couple more questions to round things out. What's your take on Codex? How do you feel about that product? How do you feel about where they're going and just kind of competing in this very competitive space in coding agents?

嗯，我其实没用过。但我觉得它刚出来时我可能用过。它看起来很像 Claude Code，所以还挺受宠若惊的。有更多竞争其实是好事，因为人们应该有选择，而且希望这能迫使我们都做得更好。不过说实话，对我们团队来说，我们只专注于解决用户的问题。所以我们不会花很多时间看竞争对手的产品。我们不会真的去试用其他产品。你当然要知道它们的存在。但对我来说，我就是喜欢和用户交流，喜欢把产品做得更好，喜欢根据反馈行动。所以，真的就是打造一个好产品。

Yeah, I actually haven't really used it. But I think I did use it maybe when it came out. It looked a lot like Claude Code to me, so that was kind of flattering. It's actually good to have more competition because people should get to choose and hopefully it forces all of us to do an even better job. Honestly for our team though, we're just focused on solving the problems that users have. So for us, we don't spend a lot of time looking at competing products. We don't really try the other products. You kind of want to be aware of them. You want to know they exist. But for me, I just love talking to users. I love making the product better. I love just acting on feedback. So, it's really just about building a good product.

最后一个问题。我和 Anthropic 的联合创始人 Ben Mann 聊过。他想让我问你一个问题，我已经融入我们的对话了。他问你的问题是：你 AGI 之后的计划是什么？你觉得你会做什么？一旦我们达到 AGI，不管那意味着什么，你的生活会是什么样？

Maybe a last question. So, I talked to Ben Mann, co-founder of Anthropic. What to talk to you about here about just suggestions which I've integrated throughout our chat. One question he had for you is what's your plan post AGI? What do you think you're going to be doing? What's your life like once we hit AGI, whatever that means?

在加入 Anthropic 之前，我其实住在日本乡下，那是一种完全不同的生活方式。我是镇上唯一的工程师，也是镇上唯一说英语的人。氛围完全不同。每周我会骑几次自行车去农贸市场。你会经过稻田之类的地方。这和旧金山完全相反。我特别喜欢的一点是，我们通过交换腌菜来认识邻居、建立友谊。在我们住的那个小镇，几乎每个人都会做味噌，每个人都会做腌菜。所以我也变得挺擅长做味噌的。我做了好几批，现在还在做。味噌有趣的地方在于，它教会你以很长的时间尺度来思考。这和工程学完全不同。因为一批白味噌至少需要三个月才能做好。而我们的红味噌需要两三年甚至四年。你必须非常有耐心。你把它混合好，然后放着。你必须非常非常有耐心。所以我喜欢它的地方就是这种长时间尺度的思考。嗯，我觉得 AGI 之后，或者如果我不在 Anthropic 了，我可能会去做味噌。

So, before I joined Anthropic, I was actually living in rural Japan and it was like a totally different lifestyle. I was the only engineer in the town. I was the only English speaker in the town. It was just a totally different vibe. A couple times a week I would bike to the farmers market. And you bike by rice paddies and stuff. It was just completely opposite of San Francisco. One of the things that I really liked is a way that we got to know our neighbors and we kind of built friendships is by trading pickles. So, in that town where we lived, it was actually like everyone made miso, everyone made pickles. And so I actually got decently good at making miso. And I made a bunch of batches and this is something that I still make. Miso is this interesting thing where it teaches you to think on these long time scales. That's just very different than engineering. Because a batch of white miso takes at least 3 months to make. And our red miso is like two, three, four years. You just have to be very patient. You kind of mix it up and then you just let it sit. You have to be very very patient. So, the thing that I love about it is just thinking in these long time scales. And yeah, I think post-AGI or if I wasn't at Anthropic, I'd probably be making miso.

我喜欢这个回答。Ben 让我问你关于你和味噌的事。我很高兴你回答了。好的，所以未来可能就是深入味噌，把做味噌做得特别好。太棒了。Boris，这太棒了。我觉得我们现在是乌克兰兄弟了。在我们进入非常激动人心的收尾之前，你还有什么想分享的吗？有什么想留给听众的吗？有什么想强调的吗？

I love this answer. Ben asked me to ask you about what's the deal with you and miso. And so I love that you answered it. Okay, so the future might be just going deep into miso. Getting really good at making miso. Amazing. Boris, this was incredible. I feel like we're brothers now from Ukraine. Before we get to our very exciting landing ground, is there anything else that you wanted to share? Is there anything you want to leave listeners with? Anything you want to double down on?

嗯，我想强调一下，对于 Anthropic 来说，从一开始，从编码开始，然后到工具使用，再到计算机使用，这就是我们思考问题的方式。这也是我们知道模型会如何发展，或者说我们想要构建模型的方式。这也是我们学习安全、研究安全并最大程度改进安全的方式。所以，现在围绕 Claude Code 发生的一切，它变成了一个巨大的、价值数十亿美元的业务。现在我的朋友都用 Claude Code，他们总是发短信告诉我。所以，这东西变得有点大了。从某些方面来说，这完全是个惊喜。因为我们不知道它会成为这个产品。我们不知道它会从终端开始或类似的东西。但从某些方面来说，这完全不足为奇，因为这是我们公司长期以来的信念。同时，感觉仍然非常早期。世界上大多数人仍然不使用 Claude Code。世界上大多数人仍然不使用 AI。所以，感觉这就像完成了 1%，还有更多要做。

Yeah, I think I would just like underscore that for Anthropic since the beginning, this idea of starting at coding then getting to tool use then getting to computer use has just been the way that we think about things. And this is the way that we know the models are going to develop or the way that we want to build our models. And it's also the way that we get to learn about safety, study it, and improve it the most. So, everything that's happening right now around Claude Code becoming this huge, multi-billion dollar business. And now all my friends use Claude Code and they just text me about it all the time. So, this thing getting kind of big. In some ways it's a total surprise. Because we didn't know that it would be this product. We didn't know that it would start in a terminal or anything like this. But, in some ways it's just totally unsurprising because this has been our belief as a company for a long time. At the same time it just feels still very early. Most of the world still does not use Claude Code. Most of the world still does not use AI. So, it just feels like this is 1% done and there's so much more to go.

是啊，老兄。看到这些数字真是难以置信。你们刚融了一大笔钱。我觉得光是 Claude Code 就有 20 亿美元的收入。你觉得 Anthropic，我记得你们公布的数字是 150 亿美元的收入。想到现在还这么早期，而我们已经看到这些数字，真是疯狂。

Yeah, man. That's insane to think seeing the numbers that are coming out. You guys just raised a bazillion dollars. I think Claude Code alone is making $2 billion in revenue. You think Anthropic, I think the number you guys put out you're making 15 billion in revenue. It's insane to just think this is how early it still is and just the numbers we're seeing.

是啊，太疯狂了。Claude Code 持续增长的原因其实就是用户。很多人用它，他们非常热情，爱上了这个产品。然后他们告诉我们哪些地方不好用，他们想要什么。所以，它不断改进的唯一原因就是每个人都在使用它，每个人都在谈论它，每个人都在不断反馈。这是唯一最重要的事情。对我来说，这就是我喜欢度过每一天的方式——和用户交流，为他们把产品做得更好。还有做味噌。嗯，味噌不需要太多参与，只需要等待。

Yeah, it's crazy. And the way that Claude Code has kept growing is honestly just the users. So many people use it. They're so passionate about it. They fall in love with the product. And then they tell us about stuff that doesn't work, stuff that they want. And so, the only reason that it keeps improving is because everyone is using it. Everyone is talking about it. Everyone keeps giving feedback. And this is just the single most important thing. And for me, this is the way that I love to spend my days just talking to users and making it better for them. And making miso. Well, the miso's not super involved. They just got to wait.

稍等一下。好了，Boris，我们到了非常刺激的快问快答环节。我有五个问题要问你。准备好了吗？开始吧。第一个问题，你经常向别人推荐哪两三本书？

Just got to wait. Well, Boris, with that we've reached our very exciting lightning round. I've got five questions for you. Are you ready? Let's do it. First question, what are two or three books that you find yourself recommending most to other people?

我特别喜欢读书。先推荐一本技术书：《Scala 函数式编程》。这是我读过最好的技术书。虽然你可能不会用 Scala，而且我不知道这在未来有多重要，但函数式编程和类型思维有一种优雅，这正是我编程的方式，也是我无法停止思考编程的方式。你可以把它看作历史产物，或者能提升你的东西。我很喜欢这本书，之前没提过，是我的最爱。

I'm a big reader. I would start with a technical book: Functional Programming in Scala. This is the single best technical book I've ever read. It's very weird because you're not going to use Scala and I don't know how much this matters in the future now, but there's this elegance to functional programming and thinking in types, and this is just the way that I code and the way that I can't stop thinking about coding. So you could think of it as a historical artifact or something that will level you up. I love this. Never before mentioned book, my favorite.

哦，太棒了，太棒了。

Oh, amazing, amazing.

第二本是斯特罗斯的《加速》。这大概是我最喜欢的类型，科幻。《加速》是一本不可思议的书，节奏极快。节奏越来越快，我觉得它比任何我读过的书都更能捕捉我们当下的本质。故事从起飞开始，我们接近奇点，最后以围绕木星轨道的集体龙虾意识结束。这发生在几十年间，节奏太棒了。我真的很喜欢。

Okay, second one is Accelerando by Stross. This is probably my big genre, sci-fi. Accelerando is just this incredible book, so fast-paced. The pace gets faster and faster, and I feel like it captures the essence of this moment we're in more than any other book I've read. It starts as a lift-off is starting to happen, we're approaching the singularity, and it ends with this collective lobster consciousness orbiting Jupiter. This happens over a few decades, so the pace is incredible. I really love it.

也许我再推荐一本：刘慈欣的《流浪地球》。他是《三体》的作者。我觉得《三体》很棒，但我更喜欢他的短篇小说。《流浪地球》是短篇小说集，里面有一些非常精彩的故事。看中国科幻也很有趣，因为它和西方科幻视角很不一样，他的思维方式读起来很有意思，文笔也很美。

Maybe I'll do one more book: The Wandering Earth by Cixin Liu. He's the guy who did Three-Body Problem. I think Three-Body Problem was awesome, but I actually like his short stories even more. Wandering Earth is one of the short story collections, and he has some really amazing stories. It's also quite interesting to see Chinese sci-fi because it has a very different perspective than Western sci-fi, and the way he thinks is really interesting to read, beautifully written.

科幻小说让我们思考未来的方向，这太有趣了。就像有无数模型说：‘好的，我明白了。我读过这种世界。’

It's so interesting how sci-fi has prepared us to think about where things are going. Just like crazy amounts of models of, 'Okay, I see. I've read about this sort of world.'

是的，对我来说，这其实是我加入 Anthropic 的原因。我当时住在乡下，思考着很长的时间尺度，因为那里一切都很慢。你做的所有事情都围绕季节和需要好几个月才能收获的食物。社交活动是这样组织的，时间也是这样安排的。你去农贸市场，是柿子季节，因为有 20 个柿子摊，下周就是葡萄季节。所以是这种长时间尺度。我当时也在读很多科幻小说，就在那个时刻，思考这些长时间尺度，我知道事情会如何发展，我觉得我必须为它变得更好一点做出贡献。这就是我最终来到 Anthropic 的原因。Ben Mann 也起了很大作用。

Yeah, I think for me this is the reason that I joined Anthropic actually. I was living in this rural place, thinking these long time scales because everything is so slow out there. All the things you do are based around the seasons and food that takes many months. That's how social events were organized, how you organize your time. You go to the farmers market and it's persimmon season because there are 20 persimmon vendors, then the next week it's grape season. So these long time scales. I was also reading a bunch of sci-fi at the time, and just being in this moment, thinking about these long time scales, I knew how this thing can go, and I felt like I had to contribute to it going a little bit better. That's actually why I ended up at Anthropic. Ben Mann was also a big part of that.

我觉得我想做一整期播客，专门聊聊你在日本的时光，以及 Boris 从日本到 Anthropic 的旅程，但我们还是简短点。我快速给你推荐一本科幻书，如果你没读过的话。你读过《深渊上的火》吗？

I feel like I want to do a whole podcast just talking about your time in Japan and the journey of Boris through Japan to Anthropic, but we'll keep it short. I'll quickly recommend a sci-fi book to you if you haven't read it. Have you read Fire Upon the Deep?

这是文奇写的，对吧？是的，很棒。

This is Vinge, right? Yeah, it's great.

是的。那本书从 AI 和 AGI 的角度来看非常有趣。读过的人很少。我自己读过。

Yes. Okay. That one is so interesting from an AI AGI perspective. So few people have read that. I read it myself.

是的，我很喜欢这本。我也喜欢《天渊》。我觉得那些部分是最棒的。

Yeah, it's one I like a lot. I like Deepness in the Sky also. I think those parts are the greatest.

续集更好。是的，是的，是的。我也这么认为。它很长，进入状态很复杂，但非常好。

The sequel is greater. Yeah, yeah, yeah. I think so. It's very long and complex to get into, but so good.

好的，我们继续快问快答。你最近有没有特别喜欢看的电影或电视剧？

Okay, we'll keep going through our lightning round. Do you have a favorite recent movie or TV show you've really enjoyed?

我其实不怎么看电视或电影。最近真的没时间。不过我看了 Netflix 上的《三体》剧集，非常喜欢。我觉得它是对原著系列很好的改编。

I actually don't really watch TV or movies. I just don't really have time these days. I did watch The Three-Body Problem series on Netflix, and I really loved it. I thought it was a great rendition of the book series.

这是 AI 领导者的共同模式：没时间看电视或电影，我完全理解。你最近有没有发现特别喜欢的产品？

It's a common pattern across AI leaders: no time to watch TV or movies, which I completely understand. Is there a favorite product you've recently discovered that you really love?

我要稍微推销一下，就说 Coda 吧。这真的是一个对我生活改变很大的产品，因为我一直开着它，特别是 Chrome 集成非常出色。它帮我付了交通罚单，取消了几个订阅。它省去了大量繁琐的工作，太棒了。我还喜欢另一个播客：Ben 和 David 的《Acquired》。他们深入商业历史并让它活灵活现的方式非常好。如果你没听过，我建议从任天堂那期开始。

I'm going to shill a little bit and just say Coda. This is legitimately the one product that's been pretty life-changing for me, because I have it running all the time, and the Chrome integration in particular is just really excellent. It paid a traffic fine for me, canceled a couple subscriptions for me. The amount of tedious work it gets out of the way is awesome. I also have another podcast that I really love: Acquired by Ben and David. The way they get into business history and bring it alive is really good. I would start with the Nintendo episode if you haven't listened to it.

好建议。关于 Coda，为了让没试过的人理解：基本上你输入想完成的事情，它就能启动 Chrome 并为你操作。我看到有人从 Anthropic 休陪产假，你用 Coda 帮他填了那些医疗表格，那些很烦人的 PDF，它直接加载浏览器、登录、填写、提交。

Great tip. With Coda, just so people understand if they haven't tried this: basically you type something you want to get done and it can launch Chrome and just do things for you. I saw someone went on pat leave from Anthropic and you had it fill out these medical forms for him, these really annoying PDFs where it just loads up the browser, logs in, fills them out, submits them.

是的，没错。而且它真的能行。我们大约一年前试过这个实验，但不太成功，因为模型还没准备好，但现在它真的能用了，太棒了。我觉得很多人不太理解这是什么，因为他们以前没用过智能体。对我来说，它很像一年前的 Claude Code。但就像我说的，它比早期的 Claude Code 增长快得多。所以我觉得它开始有点突破了。

Yeah, exactly. And it actually just kind of works. We tried this experiment like a year ago and it didn't really work because the model wasn't ready, but now it actually just works and it's amazing. I think a lot of people just don't really understand what this is because they haven't used the agent before. It feels very similar to me to Claude code a year ago. But like I said, it's just growing much faster than Claude code did in the early days. So I think it's starting to break through a bit.

还有你提到的那个 Chrome 扩展，可以独立使用，就在 Chrome 里，你可以直接跟 Claude 对话，让它看着你的浏览器屏幕，帮你做事，比如告诉你你在看什么、总结内容等等。

And there's also this Chrome extension that you mentioned that you could just use standalone that sits in Chrome and you could just talk to Claude, looking at your screen at your browser and have it do stuff, have it tell you about what you're looking at, summarize what you're looking at, things like that.

没错。对于刚想用 Cowerk 的人，我建议：下载 Claude 桌面应用，进入 Cowerk 标签，就在代码标签旁边。我推荐的做法是，先让它用工具，比如清理桌面、总结邮件，或者回复前三封邮件。它现在真的能帮我回复邮件了。第二件事是连接工具。比如你说‘查看我的重要邮件，然后发 Slack 消息’，或者把它们放进电子表格。例如，我用它做所有项目管理。我们团队有一个共享电子表格，每个工程师一行，每周每个人填写状态。每周一 Cowerk 会自动检查，给没填状态的工程师发 Slack 消息。我再也不用做这个了。这只是其中一个问题，它能做所有事。第三件事是并行运行多个任务。用 Cowerk 你可以同时运行任意数量的任务。所以启动一个任务，比如这个项目管理在跑，然后我再做别的，再启动别的，然后去喝杯咖啡。我会分享一篇文章，里面有很多人使用之前叫 Claude Code 或现在通过 Cowerk 能做的事。因为很多人会说‘哇，我没想到还能这么用’。一旦你看到这些例子，我觉得大家需要听到‘哇，我不知道还能这样’。

Exactly. For people that are just running to use Cowerk, the thing I recommend is: download the Claude desktop app, go to the Cowerk tab, it's right next to the code tab. The thing I recommend doing is start by having it use a tool. So like clean up your desktop or summarize your email or something like this, or respond to the top three emails. It actually just responds to emails for me now, too. The second thing is connect tools. So if you connect, like if you say, 'Look at my top emails and then send Slack messages,' or put them in a spreadsheet or something. For example, I use it for all my project management. We have a single spreadsheet for the whole team. There's a row per engineer. Every week everyone fills out a status. And every Monday Cowerk just goes through and messages every engineer on Slack that hasn't filled out their status. So I don't have to do this anymore. And this is just one problem. It will do everything. And then the third thing is just run a bunch of tasks in parallel. With Cowerk, you can have as many tasks running as you want. So start one task, I have this project management thing running. Then I'll have to do something else, then something else, and then I'll kick these off. And then I just go get a coffee while it runs. There's a post I'll link to that shares a bunch of ways people use what was previously Claude Code or now just you could do through Cowerk. Because a lot of this is just like, 'Oh wow, I hadn't thought I could use it for that.' And once you see these examples, I think people need to hear of just like, 'Oh wow, I didn't know I could do that.'

是啊，我觉得很多也受到了你 Lenny 的启发。你发过一篇关于 Claude Code 的 50 个非技术用例之类的帖子。所以我们的一位产品经理在发布 Cowerk 前就用它来评估。当 Cowerk 能完成 50 个中的 48 个时，我们就觉得‘不错了’。

Yeah, I think a lot of this was also inspired by you, Lenny. You had this post about 50 non-technical use cases for Claude Code or something like this. So we actually one of our PMs used that as a way to evaluate Cowerk before we released it. And I think at the point where we hit where Cowerk was able to do like 48 out of the 50, we were like, 'Okay, it's pretty good.'

哇，我都不知道。太棒了。我成了评估标准了。感觉如何？

Wow, I did not know that. That is awesome. I'm becoming an eval. Yeah? How does that feel?

太棒了。我觉得自己对 AI 的未来很有价值。这就像反向基准测试。

Amazing. I feel like I'm valuable to the future of AI. This is like reverse benchmarking.

哇，太酷了。好吧，我好奇最后两个是什么。不管了，还有两个问题。

Wow, that is so cool. Wow, okay. I wonder what those last two are. Anyway, okay. Two more questions.

你工作或生活中有没有经常想起的人生格言？

Do you have a favorite life motto that you often come back to in work or in life?

运用常识。我觉得我看到的很多失败，尤其是在工作环境中，是人们没有运用常识。比如，他们不假思索地遵循流程，或者做一个产品，但产品本身不好或想法不好，他们只是随大流而不思考。我看到的最好结果是那些从第一性原理思考、培养自己常识的人。如果某件事感觉不对劲，那很可能不是好主意。所以这也是我给同事最多的建议。

Use common sense. I think a lot of the failures that I see, especially in a work environment, is people just failing to use common sense. Like, they follow a process without thinking about it. They just do a thing without thinking about it, or they're working on a product that's not a good product or not a good idea. And they're just following the momentum and not thinking about it. I think the best results that I see are people thinking from first principles and just developing their own common sense. Like if something smells weird, then it's probably not a good idea. So I think this is the single advice that I give to co-workers more than anything too.

我觉得光这个就能单独做一期播客了。什么是常识？怎么培养？但我们简短点。

And I feel like that alone could be its own podcast conversation. What is common sense? How do you build? But we'll keep this short.

最后一个问题。你在 Twitter/X 上更活跃了。我好奇为什么，以及你在 Twitter 上的体验如何，因为你得到了很多互动。

Final question. You've been more active on Twitter slash X. I'm curious why and what your experience has been with Twitter, the world of Twitter, because you get a lot of engagement on Twitter slash X.

很长一段时间我只用 Threads，因为我以前参与过 Threads 的开发。而且我也喜欢它的设计，很干净的产品。我开始用 Threads 是因为无聊。去年十二月我在欧洲，和妻子到处旅行，去了哥本哈根等几个国家。对我来说就像编程假期，每天 coding，那是我最喜欢的假期。但后来我无聊了，几个小时没想法，就打开 Twitter，看到有人在发关于 Claude Code 的推文，就开始回复。然后我想，也许我应该找找有 bug 或反馈的人。于是我自我介绍，问大家有没有 bug 和反馈。他们对我们处理反馈的速度感到惊讶。对我来说这很正常：如果有人报 bug，我几分钟就能修好，因为我刚写了代码，只要描述清楚，它就能自动处理，然后我继续做下一件事。但很多人觉得惊讶，所以很酷。Twitter 上的体验很棒，和人们互动，了解他们想要什么，听到 bug 和功能需求。前几天我在 Twitter 上看到 GitHub issue 的抱怨，说发太多线程导致崩溃，我就想‘哦，怎么回事？’

For a long time I used Threads exclusively because I actually helped build Threads a little bit back in the day. And I also just like the design. It's a very clean product. I started using Threads because I was bored. So in December I was in Europe. My wife and I were traveling around Europe for December, just kind of nomading around. We went to Copenhagen, went to a few different countries. For me it was just like a coding vacation. So every day I was coding and that's my favorite kind of vacation. It's just code all day. It's the best. And at some point I just got bored and ran out of ideas for a few hours. I was like, 'Okay, what do I want to do next?' So I opened Twitter and saw people tweeting about Claude Code and then I just started responding. Then I was like, 'Okay, maybe I should look for people who have bugs or feedback.' So I introduced myself and asked if people had bugs and feedback. I think they were surprised by the pace at which we're able to address feedback nowadays. For me it's just so normal. If someone has a bug, I can probably fix it within a few minutes because I just wrote the code and as long as the description is good, it would just go and do it and then I'll answer the next thing. But I think for a lot of people it was pretty surprising. So it's really cool. And the experience on Twitter has been pretty great. It's been awesome just engaging with people and seeing what people want, hearing about bugs, hearing about features. I saw a complaint in the GitHub issue the other day on Twitter just so you could see you're posting many threads and it was breaking and just like, 'Oh man, what's going on here?'

是啊，有个 bug。希望现在修好了。太棒了。Boris，我可以跟你聊几个小时。先放你走。非常感谢你来做客。你太棒了。大家在网上哪里能找到你？听众怎么帮你？

Yeah, there was a bug. I hope it's fixed now. Amazing. Oh man, Boris, I could chat with you for hours. I'll let you go. Thank you so much for doing this. You're wonderful. Where can folks find you online? How can listeners be useful to you?

在 Threads 或 Twitter 上找我，那是最容易的地方。请随时 @ 我。发 bug，发功能请求。缺什么？我们怎么改进产品？你想要什么？我很爱听。

Find me on Threads or on Twitter. That's the easiest place. And please just tag me on stuff. Send bugs, send feature requests. What's missing? What can we do to make the products better? What do you want? I love hearing.

太棒了。Boris，非常感谢你来做客。

Amazing. Boris, thank you so much for being here.

酷。谢谢 Lenny。大家再见。

Cool. Thanks, Lenny. Bye, everyone.

非常感谢收听。如果你觉得有价值，可以在 Apple Podcasts、Spotify 或你喜欢的播客应用上订阅。也请考虑给我们评分或写评论，这能帮助其他听众找到这个播客。

Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review as that really helps other listeners find the podcast.

你可以在 lennyspodcast.com 找到所有过往节目或了解更多信息。下期再见。

You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.