AI 工程师：与 Scott Wu 一起探讨编程的未来

今天的嘉宾是 Scott Wu。Scott 是 Cognition 的联合创始人兼 CEO，Cognition 开发了一款名为 Devon 的产品，这是全球首个自主 AI 软件工程师。与我在这档播客中介绍过的其他 AI 工具不同，Devon 的设计目标是像一个真正的远程工程师那样工作，你可以像与任何其他人类工程师交流一样，通过 Slack 或它的专用网站与它聊天。大约一年前 Devon 发布时，它还非常像一个初级工程师。在过去一年里，他们取得了很大进展，现在 Devon 已被大量公司用于生产环境。我们聊了他们的 15 人工程团队如何用 Devon 来开发 Devon，包括每位工程师如何同时使用大约五个 Devon 来帮助编码和加速开发；如今他们四分之一的需求（PR）是由 Devon 提交的，而且他们预计到年底这一比例将超过 50%。我们还聊了 Scott 如何设想软件工程的未来，以及工程师的角色如何从编码者转变为架构师。我们还深入探讨了他们在找到这条道路之前经历的八次转型，以及为什么 Scott 相信这类 AI 工具会带来更多而非更少的工程师招聘。此外，我们还聊了 Devon 这个名字的由来，以及更多内容。这期节目会让你大开眼界。如果你对工程、产品构建和 AI 的未来走向感兴趣，我强烈推荐你收听。非常感谢 Claire Vo 为这次对话提出了一系列很棒的问题。如果你喜欢这档播客，别忘了在你最喜欢的播客应用或 YouTube 上订阅并关注。另外，如果你成为我通讯的年度订阅者，你将免费获得一年的 Linear、Superhuman、Notion、Perplexity 和 Granola 使用权。请访问 lennisnewsletter.com 并点击 bundle 查看。接下来，有请 Scott Wu。

Today, my guest is Scott Wu. Scott is the co-founder and CEO of Cognition, which makes a product called Devon, the world's first autonomous AI software engineer. Unlike other AI tools that I've highlighted on this podcast, Devon is designed to act like an actual remote engineer that you chat with like you would with any other human engineer through Slack or through its dedicated website. When Devon launched about a year ago, it was very much a junior engineer. Over the past year, they made a lot of progress and Devon is now being used by tons of companies in production. We chat about how their engineering team of 15 uses Devons to build Devon, including how every engineer uses about five Devons each to help them code and move faster. How a quarter of their poll requests today are committed by Devons and that they expect this to be over 50% by the end of the year. We also talk about how Scott imagines software engineering is going to look in the future and how the role of an engineer changes from a coder to an architect. We also get into the eight pivots that they went through before landing on this path. why Scott believes AI tools like this will lead to more engineer hiring versus less. Also, where the name Devon comes from, and so much more. This episode is going to blow your mind. I highly recommend you listen to it if you're at all interested about where engineering, product building, and AI is going. A huge thank you to Claire Vo for suggesting a bunch of great questions for this conversation. If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube. Also, if you become an annual subscriber of my newsletter, you get a year free of linear superhuman notion perplexity and granola. Check it out at lennisnewsletter.com and click bundle. With that, I bring you Scott Wu.

Scott，非常感谢你来到这里，欢迎来到播客。

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

非常感谢邀请我，很兴奋能来参加。

Thanks so much for having me. Excited to be on.

我很高兴你能来，因为你在构建的、并且一直在构建的东西，与许多其他 AI 公司长期以来所做的非常不同。尽管他们开始向你们现在所处的位置靠拢。我们接下来会聊这个。而且这也是 AI 历史和 AI 发展历程中一个非常独特的时刻。所以现在能和你聊天真的很酷，我觉得几年后我们还会再聊，到时候可能会说：“哇，我们在很多方面都对，也在很多方面都错了。”

I'm really excited to have you here because you are building and you've been building something that is very different from what a lot of other AI companies have been doing for a long time. Although they are starting to converge to where you guys are now. We're going to talk about that. And it's also just such a unique point in the history of AI and just the journey of AI. And so it's really cool to be chatting right now and I feel like we're going to chat again in a few years and be like, "Wow, we were so right about so much and so wrong about so much."

是的。所以我很高兴你能来。让我们从 Devon 开始聊吧。

Yeah. And so I'm excited to have you here. Let's start with talking about Devon.

我们整个团队只有大约 15 名工程师。你知道，我们在开发 Devon 时大量使用了 Devon。团队中大多数人肯定同时与多达五个 Devon 一起工作。因此，Devon 每个月会在 Devon 的代码库中合并数百个拉取请求到生产环境。

Our whole team is only like 15 engineers. You know, we use a ton of Devon when we're building Devon. Most folks on the team are definitely working with up to five Devons at once. And so Devon merges like several hundred pull requests into production in the Devon code bases every month.

目前你们的 PR（拉取请求）中，Devon 和人类各占多少比例？

What percentage of your PRs are Devon versus humans right now?

大约在四分之一左右。

It's in the neighborhood of a quarter or so.

你认为到今年年底这会达到什么水平？

Where do you think this will be at the end of the year?

说实话，我们预计会超过一半，而且会多不少。

Honestly, we expect it to be a decent bit more than half.

你们在如何与 AI 工程师协作方面远远领先于其他公司。

You guys are so ahead of how companies work with AI engineers.

AI 将是我们一生中最大的技术变革。过去 50 年里我们经历的大多数重大技术革命，比如个人电脑、互联网和手机，它们都有很大的硬件成分，而硬件是分发的重要组成部分。那些为这些行业构建产品的人，随着手机用户数量的增加，随着互联网连接人数的增加，他们的市场基本上逐年稳步增长，对吧？我认为 AI 已经不同的一个地方在于，这项技术可能具有爆炸性。它不受硬件分发的限制。这意味着这个领域正在以指数级的速度增长。

AI is going to be the biggest technology shift of our lives. Most of the big tech revolutions that we've had over the last 50 years like personal computer and the internet and the mobile phone. They all had this big hardware component that was a big part of the distribution. Folks who were building for those industries kind of saw their market grow and grow and grow basically steadily year-over-year as the number of people with mobile phones increased, right? As the number of people connected to the internet increased. One of the things which is already I'd say different in AI is just how explosive the technology can be. There's no weight on hardware distribution. It means that the space is just growing so exponentially.

作为工程师和构建者的行为正在如何改变？

How is the act of being an engineer and building changing?

我认为几年后，程序员和工程师的数量会迅速增加。程序员的工作形态显然会改变。但归根结底，这门学科的核心就是能够告诉你的计算机该做什么。因此从这个角度看，我确实认为随着 AI 变得更强大，编程只会变得越来越重要。

I think there's going to be way more programmers and way more engineers a few years from now pretty quickly. The form factor of what it means to be a programmer obviously is going to change. But at the end of the day, of course, the discipline is all about just being able to tell your computer what to do. And so in that lens, I really think that programming is only going to become more and more important as AI gets more powerful.

让大家明白 Devon 到底是什么。这是你们打造的主要产品。理解 Devon 最简单的方式是什么？

Giving people an understanding of just what the heck Devon is. This is the main product that you guys build. What is the simplest way to understand what is Devon?

当然。Devon 是一个完全自主的软件工程师，能够端到端地处理任务。AI 代码工作流的各个层面已经有很多出色的工具，而 Devon 提供的是一个完整的异步工作流。你可以在 Slack 的某个 issue 上 @Devon，比如你在讨论一个问题时 @ 它；也可以在 Linear 里 @Devon，然后 Devon 会在你的 GitHub 里创建拉取请求。所以它很大程度上就是作为你的初级工程师，与工程团队协作而构建的。

Absolutely. And so Devon is a fully autonomous software engineer that is going to work on tasks end to end. And so there are a lot of great tools for all parts of the stack of the AI code workflow. What Devon does is it is a full asynchronous workflow. And so you can tag Devon on an issue in Slack. You know, you're talking about an issue and you tag Devon. You can tag Devon in Linear. You can have Devon and Devon will make pull requests in your GitHub. And so it's very much built to work with engineering teams as your junior engineer.

太棒了。好的。我记得你们发布这个产品时，有一个很大的宣传口号：这是你的新 AI 工程师。它在很多方面确实很擅长，但在其他方面并不出色。现在距离你们发布已经大约一年了，对吗？

Amazing. Okay. So I remember when you guys launched this, there was like this big pitch of this is your new AI engineer. And it was really good at a lot of stuff. It wasn't great at other things. It's been a year now about since you guys launched. Is that right?

是的，没错。

Yeah. Yeah.

最好的思考方式是什么？比如，你们刚发布时那个工程师的资历水平，以及今天这个工程师的资历水平，如果这是一种衡量 Devon 的方式的话。

What's the best way to think about like the level of seniority that engineer had back in the day when you guys launched and then the level of seniority of engineer today if that's I don't know a measure of how to think about Devon.

是的，没错。顺便说一句，这想起来很疯狂，因为一年前我们最初发布时，人们真的不相信智能体是可能的，对吧？那是一个非常不同的时代，你知道，那是 2024 年初，模型能力肯定还早得多，尤其是推理能力还早得多。自那以后，它显然发展了很多。我认为就实际技能而言，我们有一些比较。有时我们会说，刚开始时它有点像高中计算机科学的学生，然后随着时间的推移，它变得更像大学实习生，现在它就像初级工程师了。但我想说，这些更像是粗略的指导，因为我非常喜欢“锯齿状智能”这个说法。因为显然有些事情它比人类做得好得多，有些事情它比人类差得多。我认为在过去一年里，我们学到了很多，尤其是关于不仅仅是编码智能体，而是普遍的智能体，就像真正构建出我们所有人应该如何作为工作流的一部分与智能体工作和互动。所以我们构建的很多东西，我的意思是，当时没有 Slack，没有 GitHub 集成，没有 Linear，没有来回互动的规划阶段，也没有办法修改 Devon 的代码。所以自那以后，我们在产品端构建的很多功能，基本上都是为了弄清楚如何让与 Devon 合作以及向 Devon 交接任务变得尽可能顺畅。

Yeah. Yeah. And it's crazy to think about, by the way, because, you know, a year ago when we did the initial launch, I mean, people didn't really believe that an agent was possible, right? And it was a very different time, you know, it's like start of 2024, you know, things with model capabilities were definitely quite a bit earlier on. Reasoning especially was quite a bit earlier on. And in the time since then, it's obviously developed a lot. I think in terms of practical skills, you know, there are some comparisons we make. Sometimes we kind of say, well, when we got started it was kind of like a high school CS student and then as time went on it became more of like a college intern and now it's like a junior engineer. But I would say though that those are more like rough guidelines because I really like the phrase jagged intelligence for example. Because there are obviously certain things that it is much better at than a human. There are certain things that it's much worse at than a human. And I think over the last year we've learned a lot especially about not just coding agents but agents in general just like really building out like how all of us should be working and interacting with agents as part of our flow. And so a lot of the things that we built, I mean, there was no Slack, there was no GitHub integration, there was no Linear, there was no interactive planning phase working back and forth, there was no way to touch up Devon's code. And so a lot of the features that we built on the product side since then have really been about basically, yeah, figuring out how to make working with Devon and handing off tasks to Devon as smooth of an experience as possible.

这太有趣了。所以，很多工作并不是仅仅让 Devon 成为最好的工程师，而是如何与这种我们从未合作过的新型实体合作。

That's so interesting. So, a lot of the work has gone not into how do we just make Devon the best possible engineer, but it's how to work with this new type of entity that we haven't ever worked with.

我认为两者各占一半。你知道，能力显然提升了很多，我们看到了这些改进，而且是可衡量的改进，但我认为另一方面则完全关乎产品界面和工具等等。而且我认为，如今人们普遍知道如何使用聊天机器人并与它们协作，对吧？那是一个人们熟悉的界面。而显然对于智能体，我认为学习如何使用它们并充分利用它们仍然是一个真正的学习曲线。所以看到很多其他人也开始在智能体领域构建和做更多事情，真的很令人兴奋。但我认为这是我们整个领域真正一起摸索的事情。

I think it's a 50-50 of both. You know, I think the capabilities obviously have improved a ton and we've seen these get better and get measurably better, but I think the other side of it is everything to do with yeah, really the product interface and the tools and so on. And I think, you know, today folks generally know how to use chat bots and to work with chat bots, right? And that's an interface that people are familiar with. And obviously with agents, it's still like a real curve, I think, to learn how to use them and how to get the most out of them. And so it's really exciting to see a lot of others starting to build and do a lot more in the agent space as well. But I think this is the kind of thing that we're all really figuring out together as a space.

关于 Devon 目前的规模，你能分享些什么？只要是你方便分享的，然后你认为 Devon 的编码能力水平一年后会达到什么程度？

What can you share about just the scale of Devon at this point? Whatever you're comfortable sharing, and then just where do you think the level of Devon's coding abilities will be in a year?

我们与各种阶段和规模的公司合作。最小的一端是只有一两个人的初创公司，他们用 Devon 来构建大量初始原型或初始产品，一直到大型上市公司、财富 100 强公司或公共银行等，他们在整个工程团队中使用 Devon。总的来说，我们看到了各种各样的用例，显然在一两个人的初创公司所做的工程工作与在公共银行所做的工作非常不同。但自始至终，它基本上都是你的那个初级伙伴，让你更快，真正地倍增你的能力。我认为它可以倍增你作为工程师的能力，显然就像让你与自己的一队 Devon 合作，而不是必须完全同步地处理单个任务。然后它也在倍增你的团队和倍增你团队的知识库，因为 Devon 确实从与你团队每个成员的合作中积累了大量知识，并能够将这些知识带入每个新的会话。

So we work with companies of all stages and sizes. You know on the smallest end it goes to startups of just one or two people who are using Devon to build out a lot of their kind of like initial prototype or initial product all the way up to big public companies, Fortune 100 companies or public banks or things like that who are using Devon across their engineering teams. In general, we've seen a huge range of use cases there and obviously the kinds of engineering work that you're doing at a one or two person startup is very different from the kind of work that you're doing at a public bank. But throughout it's all been basically yeah being that junior buddy of yours that makes you go faster and really multiplies you I would say. I think there's it can multiply you as an engineer obviously by just like letting you work with your own team of Devons instead of having to be kind of like fully synchronous on a single task. And then it's also kind of like multiplying your team and multiplying your team's knowledge base because Devon really accumulates a lot of the knowledge from working with every member of your team and is able to bring that into each new session.

太棒了。我们稍后会在播客中向大家展示它是如何实际工作的。你将做几个现场演示，但让我们先回到旅程的起点。Devon 的起源故事是什么？这一切是如何开始的？

Awesome. We're going to show people how it actually works later in the podcast. You're going to do a few live demos, but let's actually go to the beginning of the journey. What's just the origin story of Devon? How did this all begin?

创始团队，我的意思是，我们大多数人实际上已经认识很多很多年了。对几乎所有人来说，这是我们第一次一起工作，但我们认识很久了。而且我们实际上在过去十年左右都有自己的人工智能之旅。所以对我自己来说，你知道，我之前经营一家叫 Lunch Club 的公司，那是一个用于职业社交产品的人工智能。我经营了大约五年。你知道，我的联合创始人，其中一位叫 Steven，是 Scale AI 公司的首批工程师之一，这家公司显然已经发展壮大并做得非常好。我的另一位联合创始人 Walden，是 Cursor 公司的早期工程师，这家公司显然也发展壮大并做得非常好。我们整个团队都是这样。你知道，我们很多人是从竞争性编程和数学竞赛中认识的，但自那以后的几十年里，我们一直保持非常密切的联系。而且我们都有各自的旅程。

The founding team, I mean, most of us have known each other for years and years and years actually. And for almost everyone, this is our first time working together, but we've known each other a long time. And we all actually had our own kind of journeys in AI for the last decade or so. And so for myself, you know, I ran a company called Lunch Club before this, which was an AI for a professional networking product. And I ran that for about five years. And you know, my co-founders, one of my co-founders, Steven, was one of the first engineers at a company called Scale AI, which has obviously grown a lot and done very well. My other co-founder, Walden, was an early engineer at a company called Cursor, which has also obviously grown a lot and done really well. And our whole team kind of was like that. You know, many of us knew each other from competitive programming and math competitions, but we had stayed very closely in touch, you know, in the decades since then. And we've all kind of had our own journeys.

所以，我们团队里有一个人之前在 Neuro 带团队，一个在 Waymo，还有一个自己做过机器学习方向的 YC 工具创业。我们当时真的很想一起做点东西。那大概是 2023 年底，到现在差不多一年半了。刚开始的时候，有几件事我们特别笃定。一是强化学习真的在起作用，而且会成为能力上下一波大的范式转变。那时候，2022 年 ChatGPT 刚发布，那些模型一阶近似就是我们说的 AI 里的模仿学习，对吧？基本上就是让模型读完互联网上能找到的所有文本，然后训练它像网上的人那样说话。当然还有很多细节，但那是第一阶的粗略做法。而且效果惊人，对吧？它通过了图灵测试，能回答问题，对很多事情都有百科全书式的知识。我觉得过去一年到一年半我们进入的这个新范式，其实是高算力强化学习，这是一个非常不同的范式。基本上就是能够去执行任务，把东西拼起来，然后被评估对错，用那个反馈来决定下一步怎么做，并从中学习。所以我们当时非常确信这会发生。对我们来说，代码是自然而然的选择，有几个原因。一是因为我们自己都是程序员宅男，所以教 AI 写代码对我们来说是最酷的事。但也因为代码有完整的自动化反馈循环，你可以运行代码，这种自动化反馈正好能喂给强化学习，让这些模型在编程上变得这么强。另一个我们特别坚信的是，产品体验会从所谓的文本补全转向智能体。一阶来看，文本补全有很多很好的体验，用在营销、客服、教育，还有代码领域，GitHub Copilot 就是那一波浪潮的主导产品。但我认为我们真正看到的大转变是从这种文本到文本的模型，转向一个真正的自主系统，它能做决策、和真实世界交互、接收反馈、迭代、多步解决问题。现在我们叫它智能体，但当时那就是我们最兴奋的东西。所以一直是编程，一直是智能体。某种程度上，感觉从一开始就该是清楚的。但即便如此，我觉得在过去一年半里，我们在编程智能体这个方向上大概转了八次向。

And so, you know, we had one person who was running teams at Neuro, we had one person who was at Waymo, someone who had their own YC tools startup for machine learning. And we were really excited to build something together. And this was around late 2023, so about a year and a half ago at this point. And when we got started, there were a couple things that we felt really strongly about. One was that reinforcement learning was really working and was going to be the next big paradigm shift in capabilities. Back then, it was the initial ChatGPT launch in 2022, and those models were, to first order, what we would call imitation learning in AI, right? Which is basically you have the model read all the text that you can find on the internet and then train it to talk like somebody on the internet would talk. There are obviously a lot more details on top of that, but that's kind of the first order pass of what was really done. And it was amazing, right? It passed the Turing test. It was able to respond and have encyclopedic knowledge about a lot of things. And I think this new paradigm, which we've gotten into over this last year or year and a half, is really high compute RL, which is a very different paradigm. It's basically the ability to go and do work on a task, put something together, and then be evaluated on whether that was correct or incorrect, and use that knowledge to decide what to do and learn from that. So we felt very strongly that that was going to happen. For us, code was the natural thing to work on for a couple reasons. One, because we're all programmer nerds ourselves, so teaching AI to code is about as cool as it gets for us. But also because code has this whole automated feedback loop where you can run the code, and that is the kind of automated feedback that really feeds into the RL, which makes these models so great at coding. And the other thing that we felt very strongly about was that the product experience was going to shift from what I'll call text completion to agents basically. To first order, there have been a lot of great experiences in text completion, used for marketing, customer support, education, and in code obviously, as GitHub Copilot was really the dominant product of that initial wave. But I think the big shift that we really felt we would see is moving from this text-to-text model to an actual autonomous system that can make decisions, interact with the real world, take in feedback, iterate, and take multiple steps to solve problems. Now we call that agents, but that was what we were really excited about at the time. So it was always coding, it was always agents. And in some ways, it feels like it should have been clear from the start. But even with that, I feel like we've pivoted like eight times or something within coding agents over the last year and a half.

所以我最近注意到，所有顶尖 AI 公司，不是全部但很多，它们胜出的产品和公司名字不一样，这很不寻常。Cursor 的 AnySphere，Bolt 到 StackBlitz，你们是 Cognition Labs，Vercel，这让我觉得这些产品都是在公司发展后期才出现的，他们试了一堆东西，然后发现，哇，这个成了。而且很有意思的是，这在顶尖公司里这么普遍。还有 OpenAI 的 ChatGPT，Anthropic 的 Claude，Google 也是。

So I just noticed recently all the top AI companies, not all but many of them, the product that is winning is different. It has a different name from the company, which is not typical. Cursor's AnySphere, Bolt to StackBlitz, you guys are Cognition Labs, Vercel, and it just tells me like these all emerged later in the company's journey and they tried a bunch of stuff and like, oh wow, this thing worked. And it's so interesting that it's so common amongst these top companies. And there's even OpenAI, ChatGPT, Anthropic, and Claude, and Google.

是啊，挺有意思的。我同意。所以当我们刚开始的时候，那甚至不算是公司，更像一个项目或者黑客松。我们基本上订了一个 Airbnb，在感恩节前后待了几周，召集了一群热衷于捣鼓项目、做点酷东西的人。有意思的是，我们最初做的东西其实是解决竞赛编程问题，用智能体循环来做得更好。显然，如果你在测试用例上运行代码，你可以评估，而且你可以做很多智能体式的工作来尝试做得更好。我们一开始花了一些时间在这上面。然后我们基本上是从一个黑客屋到另一个黑客屋，整个公司的故事在某种意义上就是这样。之后我们又有了另一个黑客屋，那里诞生了 Devon 的一些最初想法，真正构建一个软件工程智能体，而不仅仅是编程智能体，让它和很多工具交互。但即便如此，也有非常多的迭代。比如和 Devon 对话这个想法，就是我们后来才想到的。最初就是你把任务交出去，它工作，然后给你展示完整的代码。现在显然你可以随时介入，对计划给出反馈，和 Devon 一起界定任务范围。很多这些东西显然是我们后来开发的，而且我们确实学到了很多关于用例、产品形态的东西，我们做了很多大的改进，在能力上、在 Devon 使用工具、调试和决策的能力上都有了阶跃式的提升。所以，这是一段有趣的旅程。我觉得我们一直思考的根本问题是：软件工程的未来是什么，我们应该如何与 AI 合作写代码？因为归根结底，这是我们所有产品决策的基础。

Yeah, it's funny. Yeah, I agree. So when we got started, it wasn't even really a company. It was more like a project or a hackathon almost. We booked an Airbnb basically for a couple weeks around Thanksgiving time, and just got a bunch of people together who were excited to hack on some projects and build something cool. And it's funny, actually, the first thing that we were building for was more like solving these contest programming problems and using an agentic loop to really do better on that. So obviously, if you run your code on test cases, you can evaluate, and there's a lot of agentic work that you can do there to try and do better. We spent some time on that initially. And then we've kind of gone from, I mean, the story of the whole company for us in some sense has been going from hacker house to hacker house. So after that, we had another hacker house, and that's where some of the initial ideas for Devon came, and really building like a software engineering agent, not just a coding agent, and having it interact with a lot of these tools. But even then, there were so many iterations. Even the idea of talking to Devon, for example, was something that we had to come up with. Initially, it was just like you hand off a task, and then it works, and then it shows you this whole finished code. And now obviously it's like you can jump in at any time, you can get feedback on the plan, you guys can scope out the task together when you're working with Devon. A lot of these things we had to develop obviously, and certainly we've learned a lot about the use cases, the form factor, we've made a lot of big improvements, and step function improvements on the capabilities, and Devon's ability to use tools, debug, and make decisions. So yeah, it's been a fun journey. I think the grounding question for us really is one that we think about all the time: what is the future of software engineering, and how should we be working with AI to write code? Because at the end of the day, that's what underlies all the product decisions that we make.

所以我很喜欢你问这个关键问题，我想在问之前先确认一下，就是为历史记录。你们大概什么时候开始捣鼓的，Devon 是什么时候发布的？那是什么情况？那段时间有多长？

So I like that you're asking the juicy question I wanted to get to before I ask it is just for the history books. How when did you guys start kind of hacking around and when did Devon launch? What was that? How long was that period?

是的，我们从 2023 年 11 月开始，当时是黑客松模式。我们大约在 2024 年初正式成立公司，然后首次发布是在 3 月，所以基本上是马不停蹄。

Yeah, so we started in November of 2023, which was in hackathon mode. We officially made it into a company around the start of 2024, and then our initial launch was in March, so it was like non-stop.

我的意思是，过去整整 17 个月一直没停过，但到了发布阶段，然后显然与企业合作、大量开发产品、构建它并让它适用于许多实际用例，然后在去年 12 月全面自助上线，现在几周前我们又推出了 2.0。所以对我们来说这是一段非常忙碌的时期，这么说都是轻的。

I mean it's been non-stop for the entire last 17 months, but getting to the launch, and then obviously working with enterprises and developing the product a lot more, building it and getting it to work for a lot of practical use cases, and then making it fully available self-serve in December of last year, and now we've rolled out 2.0 just a few weeks ago. So it's been a very busy time for us, understatement of the century.

让我问这个问题，因为你稍微提到了。把 Devon 当作一个人、为 Devon 创造个性的整个想法。我相信这不同于任何其他 AI 应用。没有其他人有名字，你不会把它当作一个人。是什么让你们决定采用这种方法，你们又是如何设计让它这样运作良好的？

Let me ask this question because you touched on it a bit. This whole idea of Devon as a person and this idea of creating a personality for Devon. It's unlike any other I believe AI app. No one else has like a name and you don't think of it as a person. What made you guys decide to go that approach and just how do you design it to work well that way?

我想说这是我们相当自豪的一个决定。我认为市面上有很多不同的产品体验，而真正让 Devon 在其功能上独一无二的是，你真的可以把任务交托给它。而且说实话，我们越来越多地看到，向人们解释 Devon 的体验，其实就是把它解释为“这是你的初级伙伴”。这适用于流程的很多部分，例如在入职引导中，最初我们确实有很多用户进来，只是看到空白屏幕，不知道该怎么办，或者他们会问：“嘿，我要对整个代码库进行大规模重构。”基本上，我们随着时间学到的是，让人们更多地想：“哇，我们先来把仓库设置好吧。让我们确保先给 Devon 一些简单的任务，让它熟悉代码库。如果 Devon 需要能够测试代码或运行 linter 或 CI 或类似的东西，我们要确保 Devon 有自己的虚拟机来做到这一点。”同样，我认为使用模式，通常并不清楚，显然你可以坐下来看着 Devon 一步步操作，以那种方式工作。但我们发现，作为一个构建大量东西的团队，最好的工作流程实际上是同时使用多个 Devon，异步运行它们，启动它们，然后基本上只在需要提供反馈或调整计划时才介入。所以在很多方面，我认为 Devon 这个名字确实是我们在产品中捕捉这种灵魂的尝试，它确实更像一个自主实体，你可以把任务交托给它，与它合作，并且随着时间推移应该教它、与它共同学习。

I would say it's a decision we're pretty proud of. I think there are a lot of different product experiences out there, and the thing that really makes Devon unique in what it does is that you can really hand off tasks. And more and more what we've seen honestly is that a lot of explaining the Devon experience to folks is really just explaining it as, yeah, this is your junior buddy. And that goes for a lot of the parts of the flow, where in the onboarding for example, initially we've definitely had a lot of users come in and just see the blank screen and not really know, or they'd ask, hey, I'm going to do this whole big rearchitecture of the whole codebase. And basically what we've learned over time is to get folks to think more like, whoa, let's work on getting the repository set up first. Let's make sure we hand Devon a couple one-pointer tasks so it can get familiar with the codebase. Let's get it the thing if Devon needs to be able to test the code or run the linter or CI or things like that. Obviously, we want to make sure Devon's got its own virtual machine set up to be able to do that. And similarly, I think the usage pattern, often it wasn't clear, and obviously you can sit and just watch Devon do it action by action and work that way. But we found that the best workflow really as a team building a lot of stuff was to work with multiple Devons and to run them asynchronously, to kick them off and only jump in basically as you needed to provide feedback or steer the plan or anything like that. And so in many ways, I think Devon as a name really is our attempt to capture the soul of that as a product, where it really is treating it like a bit more of an autonomous entity that you can hand off tasks to, that you can work with, that you should be teaching and learning with over time.

我想回到你之前开始讲、然后我岔开的话题，那就是对软件工程的影响，以及软件工程将如何变化。所以这大概有两部分。就像今天人们使用 Devon 时，比如说今年，对于这些公司来说，作为一名工程师和构建的过程正在如何改变？那是什么样子？

I want to come back to an area you started us down and then I took us away from, which is impact on software engineering and then how software engineering is going to change. So there's kind of two parts of this. Just like when people are using Devon today, say in this year, how is the act of being an engineer and building changing for those companies? What does that look like?

你知道，顺便说一下，我们自己都是软件工程师。我受过编程训练，当然内心仍然是一名程序员。我认为我们一直以来的思考方式是，有抽象层和工具。从高层次来说，我想到的一点是，我认为 AI 总体上，是的，计算机显然变得越来越智能，能够做越来越多的事情，而且有可能有一天计算机真正做我们所做的一切，人类不再负责其中任何部分。我不期望这很快就会到来。但在那之前，只要我们仍然是等式的一部分，最重要的事情之一显然是我们作为人类要指示我们的计算机，告诉它们我们想要什么、我们想构建什么、我们想做什么，对吧？而软件工程，我们今天显然把它看作是 Python、C++、JavaScript 以及所有这些，但归根结底，这门学科完全在于能够告诉你的计算机该做什么。所以从这个角度看，我真的认为编程，如果说有什么变化的话，只会随着 AI 变得更强大而变得越来越重要。而真正让我们兴奋的是看到那种迭代式的转变。所以你问今天的情况如何。我会说，这真的就像有一个初级伙伴，或者实际上是一支你可以合作的初级伙伴团队，对吧？所以我们团队中的每个工程师，我们在构建 Devon 时大量使用 Devon。Devon 每个月会在 Devon 代码库中合并几百个拉取请求到生产环境，而我们整个团队只有大约 15 名工程师。所以这是我们编写的所有代码中相当大的一部分。我们使用它的方式基本上是，是的，每个人都有自己的整个开发团队。如果你要查看各种问题，如果你要处理功能请求、处理 bug、处理你想构建的新范式，那么自然会有很多交接点，你只需说：“嘿，Devon，这是怎么回事。你能处理一下这个吗？”有时 Devon 能够 100% 自主完成任务，直接创建 PR，然后你合并 PR，那就太好了。有时你想介入那真正需要你帮助的 10% 或 20%。也许有一些细节关于你希望如何界定范围或如何架构这个功能，或者你可能想在最后自己去测试前端，确保它看起来完全符合你的要求，然后给出你的一两条反馈。但很多实际上是学习与 Devon 合作，以便能够并行做更多事情，构建更多。

You know, by the way, we're all software engineers ourselves. I'm a programmer by training and still a programmer at heart certainly. And I think the way that we have always thought about it is there's layers of abstraction and there's tools. And one way I would say it at a high level is, I think of AI in general as, yeah, computers are obviously getting more and more intelligent and are able to do more and more, and it's possible there may come a day where computers truly do everything that we do and humans are not responsible for any of it. I don't expect that to come particularly soon. But until that point, for as long as we're still part of the equation, one of the most important things to do obviously is for us as humans to instruct our computers on what we want and what we want to build and what we want to do, right? And software engineering is, we think of it today obviously as Python and C++ and JavaScript and all these things, but at the end of the day, of course, the discipline is all about just being able to tell your computer what to do. So in that lens, I really think that programming is, if anything, only going to become more and more important as AI gets more powerful. And the thing that's really exciting for us is seeing that kind of iterative transformation. And so you asked how things look today. I would say it really is like having a junior buddy or really a team of junior buddies that you can work with, right? And so every engineer on our team, we use a ton of Devon when we're building Devon. And so Devon merges like several hundred pull requests into production in the Devon codebases every month, which is, I mean, our whole team is only like 15 engineers. So it's a pretty sizable fraction of all the code that we write. And the way that we use it is basically, yeah, everyone's got their whole team of devs. If you're going to be looking through various issues, if you're going through feature requests, if you're going through bugs, if you're going through new paradigms that you want to build, then it is naturally the case that there's a lot of handoff points where you just say, "Hey Devon, here's what's going on. Can you please take a pass at this?" And sometimes Devon will be able to do the task 100% autonomously and just makes the PR and then you merge the PR and that's great. Sometimes you want to be able to jump in for the 10 or 20% that really needs your help. Maybe there's a few details with how exactly you want to scope it or how you're architecting this feature, or maybe you want to go and test the front end at the end yourself to make sure it looks exactly the way that you want and give your one or two lines of feedback after that. But a lot of it is really learning to work with Devon to be able to just do more in parallel and build more.

目前你们的 PR 中，Devon 和人类各占多少比例？

What percentage of your PRs are Devon versus humans right now?

是的，我得看一下，但大约占我们所有 PR 的四分之一左右。

Yeah, I'd have to look, but it's in the neighborhood of a quarter or so of all of our PRs.

那六个月前是什么情况？

And then what was it like six months ago?

哦，它增长了很多。我的意思是，我们内部也看到它呈指数级增长。这很有趣，因为总是能力和产品界面两者并重。智能提升了很多，但另一件事是我们花了很多时间思考如何构建一个界面，让你能在 Devon 能完成 80% 或 90% 的任务上获得它的价值。Devon 显然不完美，它会犯错。很多问题在于如何与 Devon 一起规划初始任务，让它去执行你想做的事情，最后回来审查并给出反馈，确保 Devon 随时间学习，并在需要时检查并纠正方向。

Oh, it's grown a ton. I mean, we've seen it grow exponentially internally ourselves as well. It's kind of an interesting one where it's always both the capabilities and the product interface. The intelligence has increased a lot, but the other thing is we've spent a lot of time figuring out how to build an interface where you can get Devon's value on tasks where Devon is able to do the 80 or 90%. Devon is obviously not perfect and it'll make mistakes. A lot of the question is how do you scope out your initial task with Devon, set Devon off to do the things you want, come in at the end to review and give feedback, make sure Devon learns over time, and check in as needed to course correct.

好的。那么，现在大约四分之一的 PR 是 Devon 做的。你觉得到今年年底这会达到什么程度？你猜会是多少？

Okay. So, today about a quarter of your PRs are Devon's. Where do you think this will be at the end of the year? What would you guess?

我认为到今年年底，我们预计会超过一半。随着时间的推移，我们看到的其中一件事是，你能够异步地做越来越多的工作，并且能够交出越来越多的任务。我认为编程的灵魂，软件工程的灵魂，一直在于定义你面临的问题，并仔细思考你想要构建的确切解决方案。思考架构、细节，并在脑海中精确规划出你想要构建的内容以及你希望计算机做什么。这就是软件工程伟大的原因，也是它最有趣的部分。同时，这可能只占普通软件工程师大约 10% 的时间，因为 90% 的时间都在处理 Kubernetes 错误、调试、系统崩溃、端口开放、bug 报告、代码迁移、版本升级等等。更多的是实现工作。我们思考 Devon 的方式之一是让工程师从砌砖工变成建筑师。关键在于达到能够进行高层指挥并精确指定你想要的细节的程度。这仍然关乎让人类保持控制，并能够进行完整的规格说明，但将你一天或一小时能做的事情的规模放大。

I think by the end of this year, we expect it to be more than half. As time goes on, one of the things we've seen is you're able to do more and more work asynchronously, and you're able to hand off more and more. I think the soul of programming, the soul of software engineering, has really been about defining the problem you're facing and thinking through exactly what solution you want to build. Thinking through the architecture, the details, and mapping out in your mind exactly what you want to build and what you want your computer to do. That's what makes software engineering great, and it's the funnest part. At the same time, that's probably in the neighborhood of 10% of the average software engineer's time, because 90% of the time is dealing with Kubernetes errors, debugging, system crashes, open ports, bug reports, code migrations, version upgrades, and so on. A lot more implementation. One of the ways we've thought about Devon is allowing engineers to go from brick layer to architect. It's about getting to the point where you can do high-level directing and specify things exactly how you want. It's still about having the human in control and able to do the full specification, but multiplying the magnitude of what you can do in a day or an hour.

那么在未来，假设有人想进入软件工程领域，考虑成为一名工程师。首先，你觉得人们还应该学习编程吗？我很想听听你的看法。然后第二点，对于今天已经是工程师的人来说，在从砌砖工到建筑师的转变中，你认为哪些技能会越来越重要，哪些会变得不那么重要？

So in the future, say someone is trying to get into software engineering, thinking about becoming an engineer. First of all, do you think people should still learn to code? I'd love your perspective there. And then two, for people that are engineers today, what skills do you think will be more and more important and then less important in this discussion of moving from brick layer to architect?

当然，我很喜欢这个问题。首先，关于是否还应该学习编程，我的回答是绝对应该。我认为在很大程度上，当你上计算机科学课程并学习基础知识时，当然，你会学到一点关于特定语言语法如何工作的知识，但说实话，你学到的大部分是逻辑拆解问题的能力。第二点是计算机的模型以及我们随时间构建的抽象，比如什么是数据库以及你应该如何思考它，什么是垃圾回收系统，等等。我认为这很重要的原因是，我们在编程中已经经历过类似的阶段。下一个阶段会更快、更大，但在很多方面是相似的。今天当你使用 Python 时，很多事情已经被抽象掉了。50 年前的人可能会说 Python 就是用英语解释你想要什么，然后计算机就帮你做了。这很棒，也很强大，它打开了局面。我们现在拥有的程序员比以往任何时候都多。但当你作为工程师培养技能时，理解抽象并剥开层次真的很有帮助。人们会使用汇编语言进行性能优化，但要构建好的系统，你需要理解网络、TCP/IP、Python 代码被解释时会发生什么，以及所有这些细节。类似地，我们会达到一个状态，即使没有经验，你也能通过解释你想要什么来构建很酷的东西。但在相当长一段时间内，你确实需要能够精确地思考细节，剥开抽象，并非常精确地知道你想要构建什么以及如何构建。

Yeah, for sure. I love this question. First of all, the question of whether you should still learn to code, my answer would be absolutely yes. I think to a large extent, when you take computer science classes and learn the fundamentals, sure, you're learning a little bit about how a particular language's syntax works, but honestly, most of what you're learning is about the ability to logically break down problems. And two, the model of a computer and the abstractions we've built over time, like what is a database and how should you think about it, what is a garbage collection system, and so on. The reason I think that's important is because we've already gone through similar phases in programming. This next one will be somewhat faster and bigger, but in many ways similar. When you work with Python today, a lot of things are already abstracted away. Someone from 50 years ago might call Python explaining in English what you want and the computer does it. That's great and powerful, and it's opened things up. We have far more programmers than ever before. But as you build your skills as an engineer, it really helps to understand the abstractions and peel back the layers. Folks will use assembly for performance optimization, but to build good systems, you want to understand networking, TCP/IP, what happens when Python code is interpreted, and all these details. Similarly, we'll get to a state where with no experience you can build cool stuff just by explaining what you want. But for quite some time, you really want to be able to think precisely about the details, peel back the abstractions, and be very precise about what you want to build and how.

那么对于你认为对工程师越来越有价值的技能，比如今天的工程师应该更多地投入哪些方面，而哪些方面可以忘记，不需要再考虑了？

And then for skills that you think are more and more valuable for engineers, like where should engineers today be leaning more and more into and versus like, you know, forget this, I don't need to think about this anymore.

当然。我认为架构师，我们工程领域已经有“架构师”这个词了，我觉得这直接就是方向正确的术语。我认为很多工作，你知道，一方面只是做常规实现、写样板代码之类的，我想说在很多方面，AI 编程已经让我们在这方面快了很多，对吧。但我认为很多核心问题在于理解非常复杂的系统、在整个公司的背景下工作、思考你正在构建的产品或正在做的工作，理解我们要解决什么问题、如何解决这些问题、我们到底要构建什么样的解决方案、我们要做出哪些关键决策和权衡。基本上，我认为那些能把这些做得非常好的人，会越来越能放大自己的价值。所以如果说有什么变化的话，我认为几年后会有更多的产品、更多的程序员、更多的工程师，比今天多得多。而且我认为很快，“程序员”的形态显然会改变，在某种意义上已经改变了。但我认为我们还有太多东西要构建。你知道，我觉得人们经常谈论杰文斯悖论，软件确实是杰文斯悖论的典型例子，我们作为一个社会，总能找到更多想要为其构建软件和编写代码的东西，我真的觉得还有很多事情要做。

For sure. And I think architect, I mean it's you know we already have a term for architect in engineering and I think it is directly the directionally the right term. And I think a lot of it is really you know it's I think one thing to kind of just do a routine implementation and write boilerplate code and things like that, and I would say that in many ways AI coding has already made us much faster at that right. But I think a lot of the core questions of understanding very complex systems and working in the context of the whole company and thinking about the product that you're building or the work that you're doing, understanding okay what are the problems that we want to solve, how do we want to solve those problems, what is exactly the solution that we want to build, what are all these key decisions and trade-offs that we're going to be making. And basically I think folks who are able to do that really really well are just going to be able to leverage themselves more and more. So if anything I think there's going to be way more product, way more programmers, and way more engineers a few years from now than there are today. And I think pretty quickly the form factor of what it means to be a programmer obviously is going to change, and in some sense it already has. But I think there's just going to be so much more for us to build. You know, I think one of the great things folks talk about Jevons paradox all the time. I mean, software is truly the kind of the shining example of Jevons paradox where we have always managed as a society to find more and more things that we want to build software for and build more code for, and I really think there's a lot more out there to do.

对于不了解杰文斯悖论的人，你能简单解释一下吗？

For people that don't know Jevons paradox, can you briefly explain it?

当然。是的。杰文斯悖论就是说，当某样东西的价格下降时，总支出实际上可能反而上升。你可以用金钱、时间或资源来思考这个问题，但直接的版本是，我认为随着编程变得越来越容易、越来越高效，我们会有更多的程序员。你知道，在一种零和的观点下，你可能会说，我们在软件工程上会快 10 倍，所以我们需要少 10 倍的软件工程师，对吧？但我认为在实践中，真正会发生的是，我们实际上会构建超过 10 倍的代码。因为我们所有的工作都受到我们实际构建、执行和迭代能力的限制，我们会有很多好主意，会有很多好产品。人们会构建更多个性化的体验，例如，而且会有很多事情要做。

Absolutely. Yeah. So Jevons paradox just says that as the price of something goes down, it can still be the case that the total spend on it actually goes up. And so you know you can think about this with money, you can think about this with time or resources, but the direct version here is I think as it becomes easier and easier to program and as programming becomes more and more effective, I think we're going to have a lot more programmers. You know, it's I think in a kind of zero-sum view you might say well we're going to be 10 times faster at software engineering, and so it means that we're going to need 10 times fewer software engineers, right? But I think in practice, what really is going to happen is actually we're going to build even more than 10 times as much code. And because all the work that we do is so capped, obviously on our ability to actually build and execute and iterate, we're going to have so many great ideas out there. We're going to have so many great products out there. People are going to build a lot more personalized experiences, for example, and there's going to be a lot to do.

回到你们使用 Devon 的方式。你说每个工程师都有一支 Devon 舰队。你们公司现在大多数人平均每个工程师用多少个 Devon？

Going back to the way you guys use Devon. So you said that every engineer has kind of this fleet of Devons. How many Devons per engineer do you find most people are working with these days at your company?

是的。这非常异步，显然你可以根据需要启动、停止它们。但团队里大多数人肯定经常同时使用多达五个 Devon。这是一种很好的流程，你思考今天要完成哪五件事？一、二、三、四、五。让 Devon 一做第一件，Devon 二做第二件，Devon 三做第三件。关键是，这需要一些时间来适应，直到我们觉得它很直观。但我认为这绝对是一种不同的体验，你把大部分事情异步地交出去，每个任务的目标是，在真正需要你专业知识的环节出现，要么你需要精确定义你要解决的问题和要构建的东西，要么在一些更复杂的部分，你想引导 Devon 朝你想要的特定变更方向走。比如，我希望这个类这样设置，我希望我们修改所有下游引用等等。但基本上就是让 Devon 异步地完成大部分工作，而你参与其中。

Yeah. So it's very asynchronous and so obviously you can kick them up and start them up and shut them down basically as you see fit. But most folks on the team are definitely, yeah, often working with up to five Devons at once I would say. And it's a nice flow where it's you think through all right what are the five things that we want to get done today? One, two, three, four, five. You have Devon one do number one. You have Devon two do number two, Devon three. And the thing about it is a lot of it is, and for what it's worth, you know, I think it's taken us some time to really adjust to it and get to the point where it's really intuitive for us. But I think it's definitely a different experience where you're handing off most things asynchronously and the goal for each of your tasks is to be there for the parts that really need your expertise, where either you really need to define exactly what it is that you're solving for and what you're building, or maybe some of the more complex parts where you want to steer Devon towards particularly what kinds of changes you want to make. You know, I want the class to be set up this way and I want, you know, we should go and change all the downstream references to this as well or whatever. But basically having Devon do the bulk of the work asynchronously with you.

那你们大概有多少工程师？

And then how many engineers do you guys have roughly?

是的。我们现在的工程团队大约有 15 人。

Yeah. So our engineering team today is about 15 people.

15 人。哇，天哪。好。然后每个人有大约五个 Devon。

15. Yeah. Holy moly. Okay. And then each one has fiveish Devons.

是的。所以 Devon 的数量是工程师的五倍。

Yeah. So there's five times the number of Devons as engineers.

我喜欢这一点，这就像是对未来的一瞥。你们在公司与 AI 工程师合作方面非常领先，所以看到你们的运作方式，基本上就是大多数公司最终会采用的运作方式。

What I love about this is this is just like a glimpse into where the future is going. You guys are so ahead of how companies work with AI engineers and so seeing how you operate is going to be essentially how most companies will end up operating.

是的。而且，就我们而言，我们已经看到了这种转变，在团队方面，显然，人们不会花太多时间只是写样板代码或纯粹实现功能，而是把更多时间花在思考核心问题上：我们如何让 Devon 变得更好？Devon 的正确界面是什么？什么样的流程或功能集才能真正让体验尽可能好？显然，我们喜欢这样。

Yeah. And for what it's worth, you know, we've already seen this shift, I would say, ourselves, where it's in terms of the team, obviously, folks don't spend that much of their time just writing out boilerplate or just kind of doing pure implementation of features, and people get to spend much more of their time focused on really just thinking about the core questions of how do we make Devon better? What is the right interface for Devon? You know what is the right flow or the right set of features that's really going to make this as great of an experience as possible. And that's how we like things obviously.

你有没有，你什么时候达到那个点，就是 Devon 开始远远领先于其他人，一旦你有足够的 Devon 做所有这些事情，它们就像，你在 10 年、20 年、30 年、100 年之后，老实说，我认为作为一个社区，我们全世界的工程师都需要思考这一点，为此构建，并适应这些新技术。

Have you, when is the point you reach where you're there's takeoff of this being the by you know like your Devon starts moving so much further ahead of everyone else like once you have enough Devons doing all these things they're just like where and you're 10 years 20 years 30 years 100 years ahead honestly I think as a community you know I think the kind of all of us as engineers around the world I think are going to have think about this and build for this and kind of adapt to these new technologies.

但我想说的是，是的，我认为越来越多，尤其是随着能力变得更好，但当然即使在今天的稳态下，我认为事情会越来越转向这种异步流程。其中一个原因是，在现实世界中，你只是受到现实世界约束的限制。一种说法是，不要太当真这些数字，但这有点像一阶数学：当然，能够写文件或完成这个函数或完成这行代码帮助很大。这是一个非常好的体验。构建软件有很多部分显然几乎不是那样的。如果你要修复一个 bug，你会启动本地服务器，在前端点击自己的产品，尝试自己重现 bug。一旦你得到错误，你会查看 DataDog，看看发生了什么，并尝试在日志中找到其他错误。你查看那些文件，看看哪里出了问题。你做一些修改，也许重新运行整个流程以确保你的更改看起来正确。这在很大程度上就是软件工程师的意义所在。这些过程需要真实时间。我认为我们会越来越转向这种智能体式工作流，因为这在某种程度上是真正获得未来几年软件工程将带来的 200%、500%、1000% 收益的途径。

But what I would say is, yeah, I think more and more, especially as capabilities get better, but certainly even in steady state today, I think things are going to shift towards this kind of asynchronous flow. One of the reasons for that is in the real world, you're just capped by real-world constraints. One way to put it is, and don't take these numbers exactly, but it's kind of like the first-order math of it: of course, being able to write files or complete this function or complete this line helps a ton. It's a really great experience. There are a lot of parts of building software that obviously are almost not that at all. If you have a bug you're trying to fix, you spin up the local server, click around on your own product on the front end, and try to reproduce the bug yourself. Once you have the error, you look at DataDog and see what happened, and try to find other errors in the logs. You look at those files and see what went wrong. You make some edits, maybe rerun the whole process to make sure your change looks right. That's a lot of what it means to be a software engineer. These are processes that take real time. I think we're going to shift more and more towards this agentic workflow because that's in some ways the way to really get to the 200%, 500%, 1000% gains that we'll be getting with software engineering over the next few years.

好了，说得够多了。让我们向大家展示这到底是什么样子。你准备了一些演示，展示了你觉得有用的几个用例。那么，你将要调出你的屏幕，然后我们就开始吧。启动它，然后我们边进行边聊。

Okay, enough talk. Let's show people what the heck this actually looks like. You've got a couple demos prepared that show a few use cases you found helpful. So, you're going to pull up your screen and then we'll do it. Kick it off and then we'll talk as it's happening.

与 Devin 合作的整个过程是异步的。我觉得我们实际上可以稍微看看 Devin 的行动，然后我们可以浏览一些 Devin 完成的其他工作示例，或者 Devin 甚至在我们的团队中为我们做的事情。但之后我们可以异步地与我们自己的 Devin 再检查。所以我快速分享一下。这里我要强调的关键点是，很多显然只是作为软件工程师，或者作为我们自己的工程师，或者工程团队、产品经理等等来思考：我们想要构建什么，我们想要交接什么？例如，我们让 Devin 设置了我们自己的 Devin 代码库。我会为那个启动一个 Devin。我会说：“嘿 Devin，我和我的朋友 Lenny 在一起。嗨 Devin，你能修改 Devin 网页应用，把你们的新闻通讯作为 Devin 网站的一部分吗？”让我们在真正的 Devin 网站上做。Lenny 会抢走你所有的用户。所以我们要启动这个。正如你所看到的，Devin 立即开始并继续回应。再说一次，你可以异步地与此合作。你也可以同步地与之合作。对于这个，我们只是稍微深入一下，看看究竟发生了什么。但正如你在这里看到的，Devin 正在浏览文件，查看很多东西。我们可以基本上根据需要跟随这里，看看什么有意义。你可以看到 Devin 已经指出了几个特定的部分，对吧？有我们在前端实现的侧边栏，那里有一些部分，我们将有一个新组件，该组件将链接到 Lenny 的网站。那听起来都不错。Devin 正在问我们任何问题，如果我们这里有什么的话。这里也是同样的故事，你可以让 Devin 自己做决定并交接，或者你可以继续给出一些更多的想法，对吧？按钮应该在新标签页中打开还是在应用程序内？我会说让我们在新标签页中打开它。你可以在任何时候回答这些问题。它在等你吗？你可以在任何时候回答这些问题。你可以交接，也可以交回。它不会说：“该死的，我刚这样写了。你为什么不早点告诉我？”没错。

The whole process of working with Devin is working asynchronously. I thought it'd be cool for us to actually just watch Devin a little bit in action, and then we can go through some other examples of work that Devin's done or things that Devin does for us even on our team. But then we can check back in asynchronously with our Devin after. So I'll share this real quick. The key thing I would emphasize here is a lot of it obviously is really just about thinking as a software engineer, or as engineers ourselves, or engineering teams, PMs, and so on: what are the things that we would want to build that we would want to hand off? So we have Devin set up with our own Devin codebase, for example. I'll go ahead and kick off a Devin for that. I'll just say, "Hey Devin, I'm on with my friend Lenny. Hi Devin, can you modify Devin web app to feature your newsletter as part of the Devin website?" Let's do it on the real Devin website. Lenny's going to lose all your users. So we're going to kick this off. As you can see, Devin gets started instantly and goes ahead and responds. Again, you can work with this asynchronously. You can work with it synchronously as well. For this, we'll just go in a little bit and see exactly what's going on. But as you can see here, Devin's going through files and taking a look through a lot of stuff. We can follow here basically as we need to and see what makes sense. You can see Devin's already called out a few particular pieces, right? There's the sidebar which we have implemented on the front end, and there are pieces there, and we're going to have a new component, and that component is going to link to Lenny's website. That all sounds good. Devin's asking us any questions if there's anything that we have here. Same story here where it's kind of you can let Devin make its own decisions and hand off, or you can go ahead and give some more thoughts, right? Should the button open in a new tab or within the application? I'll say let's open it in a new tab. And you could answer these at any point. Is it waiting for you? You can answer these at any point. You can hand off and hand back off. It's not going to be like, "God damn it, I just wrote it this way. Why didn't you tell me earlier?" That's right.

是的。你知道，与 Devin 合作的一大特点是，Devin 总是充满热情。我们总是准备好投入时间。改变范围。谢谢，伙计们。

Yeah. One of the big pieces, you know, with Devin is Devin will always be enthusiastic. We'll always be ready to put in the hours. Changing scope. Thanks, guys.

所以我们会给 Devin 一个工作的机会，它将浏览这些文件，并为我们创建一个拉取请求，我们会看到然后从那里继续。但我觉得展示一些 Devin 行动的其他例子也会很有趣。其中一个例子，实际上就是今天早上，我刚刚使用了 Devin，我让 Devin 帮我为这个播客刷新我自己的事实。显然是播客和新闻通讯的超级粉丝。我问 Devin：“嘿 Devin，我要上播客了。你能帮我研究所有关于他的信息，并为我制作一个漂亮的网站测验，以便我确保我知道我的事实吗？”所以 Devin，这是今天早上。我让 Devin 做这个，我会展示 Devin 做了什么。看起来它先去了维基百科。不幸的是，维基百科上没有 Lenny 的页面。我们会努力的，我想。还在做那个交易。他们对你不好。我的意思是，我们需要那个。我们需要一个页面。然后，你知道，然后它去 Spotify 找到了。所以你在实时观看它在研究什么。是的。所以这是今天早上，显然，这是 Devin 所做事情的回放。这是 Devin 的一部分。你可以像这样观看它做了什么。是的。尤其是当你在构建工程项目或类似的东西时，你可以看到 Devin 正在做的每一步，或者如果 Devin 在本地测试了代码，显然你希望能够去看看 Devin 在点击什么、测试什么或类似的事情。所以它找到了新闻通讯。它正在查看这个，对吧？它正在阅读所有这些。

So we'll give Devin a chance to work, and it's going to go through these files and it'll make a pull request for us, and we'll see and go from there. But I thought it'd be fun to show some other examples of Devin in action as well. One of the examples, actually this morning, which I just used Devin for, is I asked Devin to help me brush up my own facts for this podcast. Obviously a huge fan of the podcast and the newsletter. I asked Devin, "Hey Devin, going to be on the podcast. Could you please research everything you can about him and make a nice website quiz for me so that I can make sure I know my facts?" So Devin, this was just this morning. I asked Devin to do this, and I'll kind of just show what Devin did. It looks like it went to Wikipedia first. Unfortunately, it's not a page on Wikipedia, which is Lenny. We'll work on that, I guess. Making that deal yet. They did you dirty. I mean, we need that. We need a page for this. And so, then, you know, then it went and found it on Spotify. So you're watching what it's researching live. Yeah. So this was this morning, obviously, and this is a playback of what Devin did. This is part of Devin. You could just like watch what it did. Yeah. Especially when you're building engineering projects or something like that, you can see kind of like each of the steps that Devin was doing, or if Devin tested the code locally, obviously you want to be able to go and look and see what Devin was clicking around with and testing or things like that. So it found the newsletter. It's going and looking at this, right? And it's going and reading all this.

然后它说：“好，我们开始把代码拼起来吧，对吧？”然后它说：“嘿，我调研过了，你知道，它正在遍历并编写所有这些，把应用组装起来。”它其实还会自己给自己出测验，我们真应该玩一下这个测验。我们来看看吧。看看我了解多少。这个播客叫什么名字？Lenny 的播客。给没在看视频的人说一下，大概有多少订阅者？一百万。非常好。是的。Lenny 主要关注哪三个主题？哦，产品、增长和职业。非常好。这是个好测验。我觉得很多人都会喜欢。除了播客，Lenny 还做什么？好的。写作、天使投资和咨询。好的。Lenny 多久发布一次？每周一次，对吧？每周一次。所以，是的。我们可以把这些都过一遍。顺便说一下，我显然也做了这个测验，以确保自己准备充分。但这是比较有趣的例子之一，显然，就像 Scott，我的通讯有多少订阅者？

And then it says, "Okay, let's get started with putting the code together, right?" And so it says, "Hey, I've researched, you know, it's going through and writing all of this, putting app together." It plays its own quiz itself, actually, which we should just play this quiz actually. Let's see. Let's see how much I know. What is the name of the podcast? Lenny's podcast. Let's for people not watching, so to say, approximately how many subscribers? A million. Very good. Yeah. What are three main topics Lenny focuses on? Oh, product growth and career. Very good. It's a good quiz. I think a lot of people would enjoy it. What does Lenny's do besides podcasting? Okay. Writing, angel investing, and advising. Okay. How often does Lenny publish? Once a week, right? Once a week. And so, yeah. So we can go through all these and do all these. And I took this quiz, by the way, obviously to make sure that I was well prepped. But yeah, this is kind of one of the more fun examples obviously of just like Scott, how many subscribers do I have in my newsletter?

实际上超过一百万。

Over a million, actually.

是的。然后我再展示最后一个，之后我们也许可以回到我们最初的运行。但正如我所说，很多这些都是为了与现有的代码工作流程配合而构建的。所以，例如，我们在 GitHub 上对 DeepSeek 仓库做了一些探索，我们把它导入到 Devon 中，并在 Devon 里设置了我们自己的分支。我想在这里展示几件事。一是 Devon 会建立整个 wiki，包含它所有的内部理解。所以当 Devon 索引代码库时，显然，构建代码库的表示、学习它并随时间改进它，是 Devon 做的重要事情之一。有趣的是，我们发现人类自然也对理解这个代码库表示感兴趣。所以，你知道，Devon wiki 是我们在这里构建的东西。你可以看看所有这些不同的部分，看看这里的每一个不同的东西。这里是 FP8 操作。你知道，这里有一个 SG lag 集成。有关于不同层如何构建和组合的图表。还有，你知道，部署操作。也有很多关于架构的细节。你也可以问关于它的问题。所以，例如，你可以问，DeepSeek 如何处理为 spec 设计的多 token 预测，它能够搜索整个代码库，并基于此给你一个知情的答案。所以我们经常使用这个。而且，你知道，当你为 Devon 划定任务范围并做初始提示时，它很有帮助。而且显然，在真空中也很有帮助，就像你经常对你的代码库有疑问，这很好，对吧？

Yeah. And then one last one I'll show and then maybe we can come back to our initial run after. But as I was saying, a lot of this is really built to work with all the existing code workflows out there. And so, for example, we were doing some exploration with the DeepSeek repository on GitHub and we imported it into Devon and we got our own fork of it set up in Devon. And a couple things I just wanted to show here. One is Devon sets up its whole wiki with all of its internal understanding. And so when Devon indexes the codebase, obviously building a representation of the codebase and learning it and improving it over time is one of the big things that Devon does. And funnily enough, we found that naturally humans really are interested to understand this codebase representation as well. And so, you know, Devon wiki is something that we built here. And you can take a look at all these different pieces and see each of these different things here. Here are the FP8 operations. You know, here's an SG lag integration. There's diagrams of how the different layers are built and put together. There's, you know, deployment operations. There's a lot of details about the architecture as well. And you can ask questions about it as well. And so, for example, you can say, how does DeepSeek handle multi-token prediction designed for spec and it's able to kind of search through the entire codebase and give you an informed answer based off of that. And so we use this a lot. And, you know, it helps when you're scoping out a task for Devon and doing an initial prompt. And it also helps obviously just in a vacuum like you often have questions about your codebase that are really nice, right?

本期节目由 Atio 赞助，这是一款 AI 原生的 CRM。Atio 从第一天起就为你的业务扩展而构建。连接你的电子邮件和日历，Atio 会立即构建一个匹配你业务模式的 CRM，你所有的公司、联系人和互动都会丰富可操作的洞察。同步你的产品使用情况、计费信息或任何其他数据源。Atio 灵活的数据模型将处理所有这些，无需僵化的模板或变通方法。使用 Atio，AI 不仅仅是一个功能，它是基础。你可以做诸如使用研究智能体即时寻找潜在客户并路由线索、使用 AI 从客户对话中获取实时洞察，以及为最复杂的工作流程构建强大的 AI 自动化。像 FlatFile、Replicate 和 Modal 这样的行业领导者已经在体验 CRM 的未来。访问 atio.com/lenny 即可获得第一年 15% 的折扣。那就是 atio.com/lenny。

This episode is brought to you by Atio, the AI native CRM. Atio is built to scale with your business from day one. Connect your email and calendar and Atio instantly builds a CRM that matches your business model with all of your companies, contacts, and interactions enriched with actionable insights. Sync in your product's usage, billing info, or any other data sources. And Atio's flexible data model will handle it all without any rigid templates or workarounds. With Atio, AI isn't just a feature, it's the foundation. You can do things like instantly prospect and route leads with research agents, get real-time insights from AI using customer conversations, and build powerful AI automations for your most complex workflows. Industry leaders like FlatFile, Replicate, and Modal are already experiencing what's next for CRM. Go to atio.com/lenny to get 15% off your first year. That's atio.com/lenny.

随着我与越来越多构建 AI 公司和应用的人交流，我学到的一件事是，他们能整合的代码库规模有很大差异。是的。这对现有公司与初创公司、拥有大型现有代码库的人来说很重要。人们应该如何看待 Devon 可以接入什么样的代码库？

Something that I've learned as I've been talking to more and more AI building companies and apps is there's a big difference in how large of a codebase they could integrate into. Yeah. And that's a big deal for companies that are existing versus startups, people that have large existing codebase. What's the how should people think about what kind of codebase Devon can plug into?

是的。所以，我们支持尽可能大的代码库，对吧？一种说法是，你知道，我们作为工程师会如何看待大型代码库，当然，当你做更改或考虑特定任务时，你不会一次性引入代码库的每一行，对吧？你有一个高层抽象，你可以思考和查看，对吧？然后你显然能够放大，在每个不同的东西上获得更高的分辨率，对吧？Devon 的工作方式大致相同，你知道，它做的第一件事是弄清楚这里的高层架构，以及这是为什么构建的，等等。但在每个组件内部，它显然也能够放大并提供更多细节。所以这里是 FP8 到 Bflat 16，以及很多是如何设置的，这里是代码库的每个不同部分。所以类似地，你知道，我们构建这个是为了可扩展。它本质上回到了工程师作为架构师的角色，现在它帮助你理解架构，有点回到那个话题。

Yeah. So, we go all the way to the biggest code bases possible, right? And one way I'd kind of put it is, you know, how the way that we as engineers would think about a large code base is certainly, you know, when you're making changes or when you're thinking about a particular task, you're not bringing in every single line of the codebase at once, right? You have a high level abstraction that you're able to think about and look into, right? And then you're obviously able to zoom in and get to kind of higher resolution on each of these different things, right? And so Devon works in much the same way where, you know, the first thing it'll do is it's going to kind of figure out like the high level architecture of what's going on here and what this is built for and so on. But within each of the components, it's obviously also going to be able to zoom in and give some more detail about each of these. And so here's you know FP8 to Bflat 16 and how exactly a lot of that is set up right here's each of the different parts of the codebase. And so similarly it's you know we've built this to be scalable. It's essentially coming back to the engineer as architect is now it's helping you understand the architecture kind of circling back to that.

是的。完全正确。我们看到的其中一个有趣用例是，人们经常用 Devon 来帮助新工程师入职团队，对吧？你知道，当你新加入时，显然你对代码库或事情如何设置有很多问题。有时问导师或经理问题也有点尴尬，如果你担心这些问题很蠢，对吧？所以能够直接问 Devon，并通过 Devon 的 wiki 来理解这些内部表示，这很好。

Yeah. Exactly. And one of the fun use cases that we've seen actually with folks is they'll often actually use Devon to get Devon's help to onboard new engineers on the team, right? And you know when you're new and you're joining there's obviously a lot of questions that you have about the codebase or about how things are set up. It also sometimes can be a little bit awkward to ask your mentor or your manager the questions and if you're worried that they're going to be really dumb questions right and so it's nice just be able to ask Devon and to go through Devon's wiki and to understand these internal representations.

对。我觉得这真的很有趣，因为它回到了你的观点，Devon 不仅仅是一个初级工程师。它是你所说的锯齿形工程师。锯齿形智能，就像它几乎像一个高级工程师在理解代码库方面。通常你必须问一个在那里工作很久的工程师：这是做什么的？这个东西在哪里？这是怎么工作的？感觉 Devon 在这方面非常擅长。

Right. I think that's really interesting because it comes back to your point that Devon is not just a junior engineer. It's what you call a jagged jagged engineer. a jagged intelligence jagged jagged intelligence where like it's almost like a staff engineer at understanding the codebase. Usually you have to ask an engineer that's been there a long time what does this do? Where's this thing? How does this work? And it feels like Devon's very good at that.

显然，一次性检索和处理大量代码及大量词元，正是语言模型非常擅长的，对吧？所以基本上能在你需要的地方获得这些收益，真的很棒。

Obviously, the retrieval and kind of processing a lot of code and a lot of tokens at once is something that language models are really great at, right? So basically being able to get those gains in the places that you need them is really great.

是的，太好了。好的，你还有几个用例。嗯，那我展示的最后一个流程是，实际上我们上周才推出，就是与 Linear 的完整 Devon 自动化设置，对吧？所以，比如你在 Deep Seek 仓库上有任务，而且一切已配置好，你只需添加 Devon 标签，Devon 就会过来给你这个，对吧？它会给出它对任务样子的想法，你可以查看你看到的每个特定文件，或者它会指出它认为重要的代码片段。然后，如果你对构建的内容或我们得出的结论感觉良好，就可以启动一个开发会话，去实际完成那项工作。

Yeah. Sweet. All right. You got a couple more uses. Yeah. Then one last flow I'll show is, we just rolled this out last week actually, but it's a full Devon automation setup with Linear, right? So if you have tasks that you're doing on the Deep Seek repository, for example, and it's all set up, all you have to do is add the Devon label, and Devon will come through and give you this, right? And it's going to give you its thoughts on what the task looks like, and you can take a look at each of the particular files that you see, or it'll point out snippets that it thinks are important. And from there, if you feel good about what was built, or the conclusions that we came to, then you can just start off a dev session that will go and actually do that work.

这太疯狂了。听起来是个很简单的想法，但本质上你说的是，Linear 里有修复和功能的任务，现在 Devon 就能直接去做这些任务。

That is insane. Like that sounds like such a simple idea, but essentially what you're saying is there are tasks in Linear that are fixes and features, and now Devon just goes off and can just do them for you.

是的。所以这绝对是一个需要动手的过程。当 Devon 在界定任务或给出想法时，你肯定希望参与其中。顺便说一句，好的地方是 Devon 会给出它的置信度，比如“我认为我有多大可能真正理解这部分或那部分”，对吧？但这确实让事情快了很多，对吧？而且正如你所说，比如很多产品经理，显然喜欢用 Devon 和 Linear 来更好地理解事情、更好地理解代码库等等。你知道，比如 Launch Darkly 的 Claire Vo 就是 Devon 的重度用户，她喜欢去界定任务、问数据问题，或者问“嘿，发生了什么”，或者“这个合并到生产环境了吗”，或者“现在是不是一个功能开关”，或者“有多少百分比的人在用这个或那个功能”。这基本上是一种让智能更易获取的干净方式。

Yeah. And so it's definitely a hands-on process. You certainly want to be involved when Devon is scoping out the task or giving you its thoughts. And the nice thing, too, by the way, is Devon will give you its confidence level and, you know, here's how likely I think I am to really understand this piece or that piece or whatever, right? But it helps make things a lot faster, right? And to your point, it's like a lot of product managers, for example, obviously love to be able to use Devon and Linear to understand things better, the codebase better, or things like that. And you know, Claire Vo from Launch Darkly, for example, is a big Devon user, and she loves basically going and scoping out tasks or asking data questions or asking hey what's going on, or like is this merged into production yet, or is this a feature flag right now, or what percent of people are getting this or that feature. It's a clean way basically to make that intelligence much more accessible.

我喜欢与 Linear 集成后，你仍然可以保持非常简单。你知道，你添加一个小工单，比如“嘿，这个主页的链接会做这个”，Devon 会很擅长理解你的意思，然后向你展示“这是我在想的，对吗？”

I love just with the integration with Linear that you can still keep it really simple. You know, you add a little ticket like, "Hey, this link to this homepage would do this," and Devon will be really good at understanding what you mean and then show you here's what I'm thinking. Is this right?

是的。酷。好的。所以 Devon 确实完成了工作。看起来 CI 出了点问题，它现在正在调试，但它已经提交了初步的第一版拉取请求，我们可以看一下。显然这是自定义部署中的 Devon 网站。我们这里有 Lenny 的通讯。我们把它发布到生产环境吧。走起。

Yeah. Cool. Okay. So Devon did finish working. It seems like there's something going on with the CI and it's debugging that right now, but it went ahead and put up the initial first pass pull request, and we can take a look. And it's the Devon website obviously in this custom deploy. And we have Lenny's newsletter right here. Let's ship this to production. Let's go.

会非常困惑。是的。太棒了。好的，再快速展示一下。

Will be so confused. Yeah. That's amazing. Okay, show it again real quick.

所以我刚把它加到 Devon 的主页上。Devon 显然能访问我们的 Devon 代码库。它在这里做了很多。所以它对这些部分非常熟悉。

So I just added it to the homepage of Devon. Devon obviously has access to our Devon codebase. It does a lot in here. And so it's super familiar with all the pieces here.

漂亮。而且它……是的，我喜欢它的样子。我们有 Devon 搜索和通讯。推动一些不错的增长。我会链接到你的网站，你链接到我的网站。我们会获得一些页面排名。是的。好的。这是个好例子吗？哦，它就在那里。我的通讯网站真漂亮。这是不是就是那种 Devon 非常擅长的例子？比如这里有一个非常具体的网站改动。人们应该如何看待 Devon 擅长什么，以及它可能在哪些地方开始失效？

Beautiful. And it's... Yeah, I like how that looks. We've got Devon search and we got newsletter. Drive some great growth. I'll link to your site. You link to my site. We'll get some page rank going. Yeah. Okay. Is that a good example? Oh, there it is. What a beautiful website for my newsletter. Is that just like a good example of the kind of thing Devon is very good at? Like here's a very specific thing to change on the website. How should people think about what Devon is very good at and maybe where it starts to fall apart?

你知道，我们经常这样描述：我认为 Devon 在处理定义明确的任务时表现最佳。一种说法是，你要给 Devon 任务，而不是问题，对吧？而且很多这类事情，比如你刚才看到的，有点像快速的前端功能请求、修复 bug、添加测试和文档，或者类似的事情。让循环变得非常好的一个因素，显然是快速迭代和测试的方式。所以对于这样的事情，显然对我们来说超级容易，比如直接拉取预览，看看链接是否有效，对吧？对 Devon 来说显然也容易。Devon 经常会登录 Devon，启动一个 Devon 会话，确保它在我们自己的代码库上工作时没问题，这有点搞笑。但是的，你通常想要的是那种容易验证和测试的东西，这是主要的。当然，你也可以处理更大的项目或更大的请求。但在那种情况下，你肯定需要更多地引导 Devon，以确保它朝着正确的方向前进。

You know, the way that we often describe it is I think Devon is best when it is working on tasks that are well defined. One way to put it is you want to be giving Devon tasks, not problems, right? And a lot of these things, like what you just saw, which was kind of like a quick front-end feature request or a bug fix or adding testing and documentation, or things like that. One of the things that makes a loop really nice, obviously, is a quick way to iterate and test. And so with something like this, obviously super easy for us, for example, to just go pull up the preview and see that the link worked, right? Obviously would be easy for Devon to do as well. Devon will often go and log into Devon and start a Devon session and make sure it's, you know, when it's working on our own codebase, which is kind of hilarious. But yeah, you generally want something that is kind of easy to verify and easy to test is the main thing. And you can work on bigger projects or bigger asks as well, obviously. But in that case, you should certainly expect to need to steer Devon more to make sure it's going the right direction.

有趣的是，这与人们谈论合成数据和强化学习创建非常容易的数据的方式非常相似。有一个非常明确的答案。是或否。是的。非常清楚。好的。让我问你这个问题。在你们设计和构建 Devon 的过程中，你们经常争论什么？

It's interesting because that's very similar to the way people talk about synthetic data and reinforcement learning creating data that's very easy. There's like a very definitive answer. Yes and no. Yeah. It's very clear. Okay. Let me ask you this question. What's something that you guys debated a lot as you were just designing and building Devon?

我会说几个想到的。一个是我们应该有多强的意见性。我们使用 Devon 的工作流程，正如你所见，主要是集成到我们的 Slack 和 GitHub，在我们的仓库中为我们创建拉取请求，响应问题报告等等。自然，我们也遇到了很多其他不同的事情，人们尝试过。我的意思是，比如有人用 Devon 订 Door Dash。甚至我们还有很多人，你知道，很多人从零开始构建很酷的网站或做类似的事情。是的，对我们来说这是一个有趣的权衡，我想描述的方式是，在我们的产品中，我们构建的大部分功能都是针对这种创建拉取请求和工程团队用例的，但我认为我们对所有其他用例的基本立场是，显然如果人们想用 Devon 做那些，那很好。我们只是想确保他们充分了解限制和可能卡住的地方。

I'll give a couple that come to mind. One I would say is the question of how opinionated we should be. We had the workflows that we used Devon for, which was very much, as you can see, for basically integrating to our Slack and GitHub, making pull requests for us in our repos, responding to issue reports or things like that. And naturally, we've had certainly a lot of other kind of different things that have come up that folks have tried. I mean, we have folks who like order their Door Dash with Devon, for example. Even we have folks who are certainly, you know, a lot of people who are kind of building cool websites from scratch or working on things like that. And yeah, it's been an interesting trade-off for us where I think the way that I would describe it is, in our product certainly, the large bulk of the features that we build are for this kind of like making pull requests and engineering teams use case, but I think basically our general stance with all the others is obviously if folks want to use Devon for that, that's great. And we want to just kind of make sure that they're fully aware about the limitations and about where things get caught up.

AI 这件事很有意思，尤其是因为，你知道，我觉得市面上最常见的创业建议之一就是：聚焦一个非常细分的用户群，做那些无法规模化的事情，把一个用例做到极致，然后再从中成长。我觉得这个建议在各方面都很棒，但有趣的是，在生成式 AI 领域，你会自然而然地看到很多产品体验可能变得更通用。所以这对我们来说是一个有趣的权衡。这是我们一直在反复思考的问题：我们到底要多做多少，去支持其他各种用例，去处理人们可能想用 Devon 做的其他事情？

It's funny with AI, especially because, you know, I would say one of the most common pieces of startup advice out there is to focus on a really niche cohort, do things that don't scale, make one use case that's really great, and then you grow from there. I think that's great advice across the board, but it's kind of interesting because with generative AI, you naturally see this where a lot of product experiences can turn out to be more general. So it's an interesting trade-off for us. This is something that we still always go back and forth on: how much do we want to do more to support all the other use cases out there and handle other things that folks might want to do with Devon?

另一个想到的问题是，Devon 应该是一个单一的综合项目体验，还是一套工具。如你所见，我们有 Devon 搜索、Devon Wiki、Linear 工单范围界定，而且这些工具当然会相互交互。但随着时间的推移，我们越来越把它看作是在构建一套工具。我认为核心的智能体体验，以及那个会去为你构建东西的核心智能体，比如，永远都会是 Devon，那是核心部分。我认为那永远是我们工作中最特别的地方。但其他所有功能，现实世界的软件共享需要一套复杂的工作，而工程说到底就是一团乱麻。所以我认为有很多不同的流程和不同的用例是有意义的。

Another one that comes to mind is how much Devon should be, let's say, a single comprehensive project experience versus a suite of tools. As you can see here, we have Devon search, we have Devon Wiki, we have the linear ticket scoping, and certainly these tools interact with each other. But as time has gone on, we've seen it more and more as really building this suite of tools. I think the core agent experience and the core agent that will go off and build things for you, for example, is always going to be Devon, and that is the core piece. I think that will always be what's really special about our work. But all the other features out there, there is a complex suite of work that's required for real world software sharing, and engineering is just messy at the end of the day. So I think there are a lot of different flows and a lot of different use cases that make sense.

一个明显的点是，你可以向 Devon 搜索问和向 Devon 问同样的问题，对吧？Devon 会去执行并做同样的事情，查看文件并给你答案。但话虽如此，一方面，在能力方面，你当然可以做很多事情来真正优化针对这个仓库的特定问答，这确实值得构建成一个特定功能。另一方面，我们发现用户其实非常喜欢拥有这种控制权。有时候你有一个正在考虑的任务，但你实际上还不想让 Devon 开始执行。你想问 Devon，了解代码库的哪些部分可能相关。所以你想非常直接地说：“这只是一个询问，我只想看看相关的代码片段，”或者“我只想看看 Wiki，了解现有的表示。”所以在能力和用户体验方面，我们都发现，随着时间的推移，这自然是有意义的。

An obvious thing to point out is you could ask the same questions to Devon search as you could to Devon, right? Devon will go through and do the same thing, look through the files and give you an answer. But with that said, on the one hand, on the capability side, there's certainly a lot you can do to really optimize things for very specifically question-answer about this repository, and that made sense to really build into a specific kind of feature. On the other side, we found that users actually really like having this access of control. Sometimes you have a task that you're thinking about, but you actually don't want Devon to get started on the task just yet. You want to ask Devon and understand what parts of the codebase might be relevant. So you want to be very direct about saying, 'This is just an ask and I just want to see the snippets of the codebase that relate,' or 'I just want to look at the wiki and understand the existing representation.' So it's kind of on both the capabilities and on the UX side, we've found that that's what naturally made sense over time.

那我们聊聊这个领域里的其他公司吧，这是很多人一直在思考的。有各种不同的方法。你们是全力投入 AI 工程师。显然有 IDE 公司。也有正在构建非常擅长工程的模型的公司。现在每个人都开始构建智能体了。你们在很多方面走在前列。比如 OpenAI 最近说他们要构建一个软件工程智能体。Anthropic 也有类似的东西。你知道，Cursor 和 Windsurf 有自己的小智能体，还有 Replit。你怎么看你们在这个格局中的位置，以及你们认为长期如何胜出？你是怎么想的？

Let's talk about the landscape then of just other companies in the space, which is something a lot of people are always thinking about. There's all these different approaches. You guys are going full-on AI engineer. There's obviously IDE companies. There's also just like models being built that are really good at engineering. Everyone's kind of starting to build agents now. You guys are ahead on this in a lot of ways. Like OpenAI just recently said they're going to build a software engineering agent. Anthropic's got something there. You know, Cursor and Windsurf have their own little agents and Replit. Thoughts on just kind of where you guys fit in in the landscape and then where you think you win long term? How do you think about that?

是的，而且顺便说一句，我觉得这些都是非常出色的团队。非常聪明、非常有远见的人，他们在构建很多很棒的产品。说实话，在未来几年，随着 AGI 或随便你怎么称呼它的到来，有很多事情要做。我喜欢的一句名言是：在 2017 年，如果你问我们是否有 AGI，答案是否定的。而在 2025 年，如果你问我们是否有 AGI，答案是，你得先定义什么是 AGI，而且这取决于你的主题。对吧。我认为这确实触及了一点：有很多非常了不起的事情正在发生。我们很容易低估我们所看到的转变有多大。比如，在过去 10 年、20 年、30 年里，有很多很棒的产品让构建产品的生命周期中这些不同的细分领域变得更容易一些。有用于即时响应的优秀产品，有用于日志记录的，有用于计费的，还有所有这些不同的工具。显而易见的是，我们在 AI 中看到的是，所有这些领域都将以数倍的速度发展，而且如果有变化的话，那将是数量级的转变。

Yeah, and for what it's worth, I think all of these are incredible teams. Really smart and really forward-thinking folks who are building a lot of great products out there. There's a lot to do honestly over the next few years with the advent of AGI or whatever you want to call it. One of the quotes that I love is: in 2017, if you asked if we had AGI, the answer is no. And in 2025, if you ask if we have AGI, the answer is well, you have to define AGI, and it depends on your subject. Right. And I think it does kind of get to the point that there is a lot of really amazing stuff happening. It's easy to underrate just how big of a shift it is that we're seeing. There are a lot of great products out there, for example, over the last 10, 20, 30 years that have made each of these different niches of the life cycle of building a product a little bit easier. There's great products for instant response, for logging, for billing, all these different tools. The obvious thing is what we're seeing with AI is all of these spaces are going to be moving multiple times faster, and it's going to be an order of magnitude shift if anything.

所以从我们的角度来看，我们显然一直押注于一个非常具体的视角，那就是自主编码智能体。说实话，那里有很多问题要解决。核心能力方面肯定还有很多工作要做。我们经常看到这样的情况：“哇，Devon 为什么会做出那个决定？看起来没有人类工程师会那样做。”在产品界面方面，显然有很多需要考虑的地方。而且这不仅仅是我们正在努力实现的单一目标，而是会随着每一次能力的增加而改变的东西。我有点把它想象成有 20 代智能体产品、智能体编码体验即将到来。我们在几年后会达到的那个，可能是一个你根本不需要看代码的东西，你实际上只是看着你自己的产品，然后你能够看着并指定说：“嘿，这个按钮应该更圆一点。我们来做吧。”

So from our perspective, we've obviously had a very specific lens that we've bet on this whole time, and that is autonomous coding agents. There's a lot of problems to solve there, to be honest. There's still a ton to do on the core capabilities, certainly. We see cases all the time where it's like, 'Wow, why did Devon make that decision? That seems like no human engineer would have ever done that.' There's all sorts of spots where, with the product interface, there's obviously a lot to think about. And it's not just a single thing that we're working towards, but something that will change with every addition of capabilities. I kind of think of it as there's 20 generations of agent product, agent coding experiences to come. The one that we'll get to over the course of several years is probably something where you don't even look at the code at all, and you're actually just looking at your own product and you're able to look and specify and say, 'Hey, this button should be a little bit rounder. Let's do that.'

顺便说一句，我们在这里加一个新标签页，也许应该保存这些信息。我们建一个数据库表，按 x、y、z 列建索引。你基本上就是在实时处理你的产品，让你的智能体帮你把这些事情做出来。显然，这中间会有很多代际的迭代，但产品体验本身每次都会改变。然后还有把它真正推向世界的各种实际问题。人们需要学会如何使用新技术。要部署到真实世界软件的种种杂乱环境中，有很多工作要做。现在还有很多 COBOL、很多 Fortran 代码。人们已经做了很多抽象和细节处理。所以从我们的角度来看，我们从一开始就专注于智能体式编码。这是我们真正相信并为之设计的一件事。这甚至体现在收入模式上，比如 ACU 和基于使用量的计费方式。它也体现在所有产品体验上，比如你想在哪里和 Devin 对话。你想能在 Slack 里和 Devin 交流，能从你的问题追踪器里启动它，所有这些。当然还有能力本身。所以我不认为有什么简单的答案。这显然是多种因素的结合，但这确实是我们过去一年半以来一直深耕的领域。未来五到十年也会是这样。

And by the way, let's add a new tab here and maybe we should save this information. Let's start up a database table and index it on x, y, and z columns. You're just basically working with your products in real time and having your agent build out those things for you. Obviously, there's going to be a lot of generations in between, but the product experience itself is going to change every single time. And then there's all the practicality of just getting it out there in the world. Folks need to learn how to use the new technology. There's a lot to do to deploy into all the messiness of real-world software. There's a lot of COBOL out there still, a lot of Fortran. There are lots of abstractions and details that folks have done. So from our perspective, we have been laser-focused on agentic coding since the beginning. That's the one thing we've really believed in and designed for. That goes all the way to the revenue model with ACUs and the usage-based setup. It goes into all the product experiences, thinking about where you want to talk to Devin. You want to be able to talk to Devin in Slack, spin it up from your issue tracker, and all these things. And then of course the capabilities. So I don't think there's any one easy answer. It's obviously a combination of things, but this is really the space we've lived in and spent all our time in for the last year and a half. And it's going to be that way for the next five or ten years too.

顺着这个思路，人工智能领域大家总会问的一个大问题是护城河和防御性。这是我一直问每位创始人的问题。在这个领域，当构建变得如此容易、模型本身又进步如此之快时，你如何看待建立护城河？

Along these lines, a big question everyone always has in AI is moats and defensibility. It's a question I've been asking every founder that comes on. How do you think about building a moat in the space when it's so much easier to build and these models are advancing so quickly?

我想稍微调整一下这个问题，我认为这往往更多关乎粘性，而不是护城河。我的意思是，护城河在某种意义上通常指能阻止竞争对手进入市场的东西。我同意在高层面上，AI 光谱不同层面的很多人——基础实验室、应用层等等——我不认为存在什么硬性壁垒能阻止别人进入。真正存在的是粘性，我把它定义为：一旦你拥有一个你真正喜欢的产品体验，你是否乐于继续使用它，还是说换一个新的、重新学习也同样容易？从这个角度看，我认为编码智能体尤其有几个很棒的地方。一是随着时间推移，会有很多内在的粘性、学习和积累。当你和你的整个团队使用 Devin 时，这和工程师的情况一样。如果你第一天加入，而不是在公司待了五年，你自己写了一半代码，你碰过每个文件，构建过每个部分，你认识所有工程师。类似地，Devin 会真正学习并建立对你代码库、技术栈和流程的表示，而且我们能在此基础上做更多事情。另一个我觉得非常令人兴奋的点是，代码确实有很多我称之为多玩家的方面，这也是现实世界中很多事情的完成方式。拥有自己的工程师体验是一回事，但比如，我们经常看到：一些工程师和 Devin 合作，教 Devin 东西，人们会让 Devin 来引导新工程师，把知识传递给他们。或者类似地，我会在 Slack 里和 Devin 开始一个会话，说：‘嘿，如果我们能做这件事就太酷了。’然后另一位工程师会插话说：‘哦，顺便说一句，我们最初做这件事是因为 X 和 Y，所以 Devin，你确保做这个改动时仍然支持那个工作流。’Devin 会说：‘好的，听起来不错。’或者 Devin 会创建一个 PR——我和 Devin 合作，在 GitHub 上创建一个拉取请求，然后其他人会审查那个 PR 或发表评论，Devin 也会处理这些。你会在 Linear 里，所有这些空间真的为一种体验做好了准备，让 Devin 能随着时间推移为你整个工作提供更多价值。所以从这个角度看，如果有什么的话，我们希望有很多创新和很多新产品。我不认为目标是阻止别人构建。有很多东西要构建，也会有很多不同的体验。从我们的角度来看，我们思考的是如何让 Devin 在你使用得更多时变得越来越有用。

I'd give one slight tweak on that, which is I think it's often less about moats and more about stickiness. What I mean by that is, moats in some sense typically mean something that prevents a competitor from even entering the market. I agree that at a high level, a lot of different folks at different layers of the AI spectrum—the foundation labs, the application layer, and so on—I don't think there's any hard barrier that would prevent others from entering. What does exist is stickiness, which I would define as: once you have a product experience you really like, are you excited to keep using it, or is it just as easy to switch to a new one and learn it? From that perspective, I think there are a few things that are really great about coding agents in particular. One is there's a lot of inherent stickiness and learning and buildup over time. As you use Devin and your whole team uses Devin, it's the same thing as with an engineer. If you're joining on day one versus being at the company for five years, you wrote half the code yourself, you've touched every file, you've built every piece, you know all the engineers. Similarly, Devin will really learn and build its representation of your codebase, your stack, and your process over time, and we'll be able to do a lot more with that. The other piece, which I think is really exciting, is there really is a lot of what I would call a multiplayer aspect of code, which is how a lot of things get done in the real world. It's one thing to have your own experience as an engineer, but for example, we see this all the time: some engineers work with Devin and teach Devin things, and folks will have Devin onboard new engineers and convey that knowledge to them. Or similarly, I'll start a session with Devin in Slack and say, 'Hey, it'd be cool if we could do this thing,' and another engineer will chime in and say, 'Oh, by the way, the reason we did it initially was X and Y, so Devin, just make sure when you make this change that you still support that workflow.' Devin will say, 'Okay, sounds great.' Or Devin will make a PR—I'll be working with Devin, we'll make a pull request in GitHub, and somebody else will be reviewing that PR or give comments, and Devin will work on that too. You'll be in Linear, and all these spaces really set up for an experience where Devin can grow in the value it provides for your whole work over time. So from that perspective, if anything, we want there to be a lot of innovation and a lot of new products. I don't think the goal is to lock other people out of building. There's a lot of stuff to build, and there are going to be a lot of different experiences. From our perspective, we think about how we can make Devin more and more useful as you use it more.

这非常相似。我们请过 Cursor 的 Michael 上播客，他也有类似的观点。他认为护城河就像消费类产品，比如 Google。他觉得就像 Google 那样，人们可以轻松切换；你只要做到最好，这就是答案。是的。而且感觉你在补充这一点，就是说，但如果你能创造一些粘性，让人们很难离开，因为它太擅长它所做的事情，并且已经积累了知识、融入了你的工作流程，那就会增强这种粘性。

It's very similar. We had Michael from Cursor, the CF Cursor, on the podcast, and he had a similar point. He thinks moats are just kind of like consumer, like Google. He thinks it's like Google where people can easily switch; you just have to be the best, and that's the answer. Yeah. And it feels like you're adding to that, of just like, but also if you can create some stickiness where it is very hard to leave because it's so good at what it's doing and it's built knowledge and integrated to your workflows, that it builds on that stickiness.

而且我认为我们领域的一个好处是，软件工程，无论好坏，都与价值有着非常明确的联系。这意味着，一种说法是，至少在未来一段时间内，总是有一个清晰的下一级目标。我认为可能会有某个时刻，你只是说，好吧，给我把整个 YouTube 建出来，然后 Devon 就全做了。可能有上亿小时的人类工程时间花在构建 YouTube、构建算法、构建所有基础设施、每一个细节上。也许有一天 Devon 能开箱即用地做到这些。那显然是很久以后的事了。在过渡期间，在中间的每一个层级上，显然它会影响软件工程的质量。而且我认为开发者们的一个很棒的地方是，开发者们非常愿意学习新体验，并且愿意付出努力，如果这意味着他们能获得越来越高质量的体验的话。

And I think one of the things that's nice about our space too is software engineering, for better or worse, has a very clear tie to value. And what it means is, I guess one way to put it is there is always kind of a clear next level, at least for the next while. I think there could be some point where you're just like, all right, just build the entirety of YouTube for me, and Devon just does the whole thing. There's probably been a hundred million hours of human engineering time building YouTube, building the algorithm, building all infrastructure, every little detail. And maybe there's some time where Devon just does that out of the box. That's obviously going to be a long time from now. On the interim, on every level in between, obviously it makes a difference in the quality of software engineering. And I think one of the cool things with developers is developers are really willing to learn new experiences and to put in effort if it means that they're able to have a higher and higher quality experience.

太棒了。我想花点时间聊聊让 Devon 成为可能的技术。在不透露商业机密的前提下，是什么让 Devon 如此出色？是不是某个模型带来了突破？很多人分享过，比如 3.7、3.5 对他们的产品来说是一个巨大的突破。你架构或构建 Devon 的方式中，关键是什么让它运行得这么好？

Awesome. I'm going to spend a little time on the tech that enables Devon. Without divulging trade secrets, what allowed you to make Devon so good? Was there an unlock with a certain model? A lot of folks have shared like 3.7, 3.5 was a huge unlock for a lot of their products. What's kind of the key to the way you've architected or built Devon that makes it work so well?

显然，我们长期以来一直在押注智能体。我认为智能体比大多数人想象的要早得多就可以实现和运作。但当然，随着社区真正围绕它团结起来，你可以在预训练中看到这些影响。你可以在这些模型所做的很多工作中看到这些影响。实际上，我不认为有任何单一的阶跃式基础模型转变或类似的东西，让 Devon 产生了天翻地覆的变化。但我确实认为，这条曲线上的每一个点，我的意思是现在每周都有新模型发布，显然在我们能做的事情上产生了巨大差异。而且在此基础上，我们与所有这些基础实验室的研究团队合作，在它们之上做了很多工作。所以我的大胆观点是，就基础智能而言，老实说我们基本上已经到位了。而且我们看到的很多情况，以及我们花时间做的事情，与其说是——显然我们不会预训练自己的模型之类的——不如说不是提高模型的基础智商，而是教它真实世界工程的所有特性。思考一下，这里是你如何使用 DataDog 做这个，这里是你可能如何诊断这个错误，这里是你可能遇到的不同情况，这里是你如何处理每一种情况，当你准备好了，这里是你如何创建 GitHub PR。这在工程领域如此，在其他领域也是如此。我们日常所做的工作有太多细节和特性，其中很多就像是教模型去镜像真实世界的复杂性，而不是让它达到某种更高的基本问题解决水平，我认为基础实验室在这方面做得非常出色。

We've obviously been betting on agents for a long time. I think agents were doable and workable a lot earlier than most folks might have thought. But certainly, as the community has really rallied around it, you see the impacts of that in the pre-training. You see the impacts of that in a lot of the work that's done with these models. I actually don't think there's been any single step function base model shift or anything that has been a night and day difference in Devon. But I certainly think that the curve, every point on the chart, I mean there's a new model that comes out every week now, has made a big difference in terms of what we've been able to do. And then on top of that, we work with the research teams at all these foundation labs to do a lot of our work on top. So my hot take here is that in terms of base intelligence, we're honestly basically already there. And a lot of what we see and what we spend our time on is less so, obviously we don't pre-train our own models or things like that, it's less about increasing the base IQ of a model, and more about teaching it all of the idiosyncrasies of real world engineering. Thinking about here's how you use DataDog and do this, and here's how you might diagnose this error, and here are the different things that you could run into, and here's how you handle each of those, and when you're ready, here's how you make a GitHub PR. This is true in engineering, it's true in other spaces as well. There's so much detail and idiosyncrasy to the work that we all do day-to-day, and a lot of it is like teaching the model to mirror the complexity of the real world, rather than getting it to some higher fundamental level of problem solving, which I think the foundation labs are doing a really great job.

在我们开始录制之前聊天时，你分享了一些关于以往变革性技术增长的内容，它们非常以硬件为导向，并且存在增长的限制因素，而 AI 不是这样。你出于多种原因分享了这一见解。

There's something you shared when we were chatting before we started recording around the growth of previous transformative technologies were very hardware oriented and there was like a limiting factor to their growth and AI is not that. You just share that insight for a number of reasons.

我认为 AI 将是我们一生中最大的技术变革。但有一件事，也就是我们之前聊到的，过去 50 年我们经历的大多数重大技术革命，我想到的是个人电脑、互联网和手机，它们都有庞大的硬件组成部分，是分发的重要部分。所以有了互联网，最初只是大学之间互相通信，但显然随着时间的推移，整个世界都接入了互联网，这花了年复一年。手机也是如此，PC 也是如此。而其中特别有趣的一点，我认为我们已经看到了其影响，是在这些硬件分发机器中，很多事情依赖于实时。所以那些为这些行业构建产品的人，随着手机用户数量的增加，随着互联网连接人数的增加，他们的市场基本上逐年稳步增长。而且很多这样的企业，现在想起来仍然疯狂，但很多企业是在一开始就创立的。比如苹果和微软几乎同时成立，很多伟大的互联网企业也是如此。但当然，它随着时间触及了整个世界，或者世界的很大一部分。它产生了巨大的影响，但因为它需要时间，所以跨越了好几年。而我认为 AI 已经不同的一个方面是，这项技术可以多么爆炸性地发展。一旦 AI 代码——我认为我们坚定地越过了 AI 代码的拐点——作为一名工程师，如果你完全不使用 AI，老实说你在落后。而且这是一项每个人都应该拥有并使用的技术，而且没有硬件分发的重量在拖累它。这意味着这个领域正在以指数级的速度增长。

I think AI is going to be the biggest technology shift of our lives. But one thing, which is what we were just talking about before this, is that most of the big tech revolutions we've had over the last 50 years, I'm thinking about the personal computer, the internet, and the mobile phone, they all had this big hardware component that was a big part of the distribution. So you had the internet, and initially it was just these universities that were talking with one another, but obviously over time we got the whole world plugged into the internet, and it took years and years and years. The same thing was true with mobile phone, the same thing was true with PC. And the thing that's interesting about that in particular, which I would say we're already seeing the effects of, is in these hardware distribution machines, there's a lot that depends on real time. So folks who were building for those industries saw their market grow and grow basically steadily year-over-year as the number of people with mobile phones increased, as the number of people connected to the internet increased. And many of those businesses, it's still crazy to think, but many of those businesses got started right in the beginning. Like Apple and Microsoft were started right around the same time, and the same is true for a lot of the great internet businesses. But certainly it was something that touched the whole world with time, or a large fraction of the whole world. And it had a really massive impact, but it took place over several years because of the time that it took. And I think one of the things which is already different in AI is just how explosive the technology can be. Once AI code, and I think we're firmly past the inflection point in AI code, where as an engineer, if you're not using AI at all, you're falling behind honestly. And it is a technology that everyone should have and should be using, and there's no weight on hardware distribution that is causing that. And it means that the space is just growing so exponentially.

基本上，迈克尔·波伦有个有趣的观点：陈词滥调之所以是陈词滥调，是因为它们太真实了，所以你才会听上一百万遍。我觉得就像人们听到这个会说“我知道”，但实际上正在发生的事情简直疯狂。

Basically, Michael Pollan has this interesting point that cliches are cliches because they're so true, and that's why you hear them a million times. I think it's like people hear this and say, 'I know,' but it's actually insane what is happening.

是啊，这就是为什么你在这里帮我们度过这个转型期。

Yeah, that's why you're here to help us through this transition.

是啊，不，我的意思是这是个有趣的时刻，我认为这需要真正的投入和实际的工作。但比如从我们作为工程师的角度来看，我认为这意味着紧跟所有正在发生的事情非常重要。而且正如我们所见，这不仅是因为你的学习和使用这些技术的能力，还在于基本上要教 AI 了解你的代码库，以便它能在与你协作开发时真正高效，并做更多你希望它做的事情。

Yeah, no, I mean it's a fun time, and I think there will be real investment and real work that it takes. But from the perspective of us as engineers, for example, I think it just means it's so important to stay in the loop with everything that's happening. And as we're seeing, it's not only because of your learning and your ability to work these technologies, but it's also about basically teaching the AI what there is to know about your codebase in order to make it really effective at building with you and doing more of the things that you would want it to do.

那么，顺着这个思路，对于正在收听的公司人员，如果他们想“嘿，我们应该在公司里用 Devon”，你发现哪些事情有助于帮助公司里的工程师获得采用并能够使用 Devon，无论是文化上还是后勤上？

So, along those lines, for company people listening who are like, 'Hey, we should be using Devon at our company,' what are things you've found to be helpful in helping an engineer at a company get adoption and be able to use Devon, either culturally or logistically?

我们经常看到的一种模式是，团队里有几个人非常兴奋，想尝试新事物，他们愿意投入，并且非常期待把它用起来。他们会完成所有设置——给 Devon 仓库，教 Devon 如何运行 lint 和 CI 以及所有那些细节——然后从给它一些初始任务开始，基本上帮 Devon 站稳脚跟。随着时间推移，最终大家会看到，“哇，Devon 写了所有这些 PR，Devon 在做这个，Devon，那个刚加入公司的家伙，正在快速产出 PR。”他们看到这些后，自然就会上手并注册账号。当然，其中一个很酷的事情是，当他们加入时，Devon 已经对他们正在工作的仓库有了相当多的了解。所以我们经常看到的一个很酷的现象是，早期采用者自己真的可以为团队中的其他人铺平道路。

A pattern we often see with folks is there will be a few folks on the team who are really excited and want to try out the new thing, and they want to put in the investment and are really excited to get it going. They'll go through all the setup—they'll give Devon the repos, they'll teach Devon how to run the lint and the CI and all of those details—and they'll start by giving it those initial tasks and kind of help Devon build a foothold basically. As time goes on, eventually folks will see, 'Wow, Devon's writing all these PRs, Devon's doing this, Devon, that person that just joined the company, is just knocking out PRs.' And they'll see that, and then naturally they'll get on and they'll get an account. One of the cool things, of course, is by the time they join, Devon already knows a good amount of detail about the repositories they're already working in. So one of the really cool things we often see is that the early adopters themselves can really pave the way for everyone else on the team.

但没错，我想指出的主要一点是，这确实是一种非常不同的产品体验，对吧？而且我认为，无论如何，我们还有很多可以做的，让它尽可能直观和清晰——比如如何使用 Devon，正确的步骤是什么，以及如何真正从 Devon 中获得最大价值。但这是那种如果你投入并确切了解让 Devon 成功需要什么，我们发现自己随着时间推移，随着每一次更新，我们只会越来越多地使用 Devon。

But yeah, I think the main thing I would call out is that it really does take a very different product experience, right? And I think for what it's worth, there's still a lot more that we can do to make it as intuitive and as clear as possible for folks—like how to use Devon, what the right steps are, and how to really maximize value out of Devon. But it's the kind of thing where if you put in the investment and understand exactly what it takes to get Devon to be successful, we've found ourselves that as time has gone on, we just use Devon more and more with every next update.

那么让我顺着这条线问下去。有一个问题我会问每一位构建 AI 应用的创始人：如果你能坐在每个 Devon 新用户旁边，在他们耳边低语几句，帮助他们成功使用 Devon，比如一两个建议，那会是什么？

So let me follow that thread. There's a question I ask every AI app building founder, which is: if you could sit next to every new user of Devon and whisper something in their ear to help them be successful with Devon, like one or two tips, what would those tips be?

我想说的最重要的一点是，真的就是把 Devon 当作你新来的初级工程师。我认为这是最重要的一点。人们进来，看到空白页面，会想到各种想尝试的东西。但我们看到最有效的流程通常是：显然你可以试试演示，也可以做些事情，但很多情况下就是，“好，我们想想今天或这周要完成哪些工单，然后让 Devon 开始做这些。”先从简单的开始，然后和 Devon 一起工作，了解 Devon 需要哪些设置才能测试自己的代码并做好，然后随着时间推移逐步扩大规模。显然，当你和你的工程师合作时，你会更好地了解如何与他们沟通，或者适合让他们参与哪些任务或项目。但我认为这确实是我们的一句话总结。

I think the biggest thing I would say is it really is just treat Devon like your new junior engineer. And I think that's the biggest thing. Folks come in and they see the blank page and they think of all sorts of various things they want to try out. But typically the flow we see that works best is: obviously you can try demos and you can do things, but a lot of it is just like, 'Yeah, let's figure out what tickets we want to get done today or this week, and let's have Devon get started on those.' And let's start with the easier ones, and then work with Devon and understand what things Devon needs to get set up to be able to test its own code and do this well, and then let's scale up over time. And obviously, as you work with your engineer, you understand better how to communicate with them or what are the right tasks or projects to bring them in on. But I think that really is the one-liner for us.

好的。有一个我一直想问的问题。我想回到这个话题，因为我经常思考 Devon 的事情。每个人都会拥有五个 Devon，比如说十个 Devon。每个人基本上都会变成一个工程经理，带着一群初级工程师。这不一定是最好的工作，因为就是一堆审查——至少你不需要做绩效评估和一对一谈话——但你知道，就像整天坐着检查很多 PR。有一种感觉是你变成了架构师，这几乎是每个工程师最终想成为的，对吧？他们都像，“我只想思考架构，我不想写所有这些愚蠢的修复 bug。”所以我理解这有好处。但你怎么——我想你肯定在思考这个问题——你怎么让生活变得愉快、有趣、享受，基本上就是未来管理 500 个 Devon 的工程经理？

Okay. There's a question I've been meaning to ask. I just want to get back to this because something I think a lot about with Devons. Everyone's going to have five Devons, let's say 10 Devons. Everyone's kind of turning into basically an engineering manager with a bunch of junior engineers. Which isn't necessarily the best job in the world because it's just a bunch of review—at least you don't have to do performance reviews and one-on-ones—but you know, it's like sitting around checking a lot of PRs all day. There's a sense you become an architect, which is kind of what every engineer wants to become eventually, right? They're all like, 'I just want to think about the architecture, I don't want to code all these stupid fix bugs.' So I get that there's a good part to that. But just how do you—I imagine you're thinking a lot about this—just how do you make life pleasant and fun and enjoyable as basically an engineering manager of say 500 Devons in the future?

是啊，我都能想象绩效评估了。你知道，“Devon，你在任务上做得非常出色，但我真的希望你在团队会议上更主动一些。”

Yeah, I can just imagine the performance. You know, 'Devon, you've done a really great job on your task, but I really would like you to be more proactive in the team meetings.'

所以，我想说的是，这其实很有趣，因为就措辞而言，我们也思考了很多。我们过去用过“Devin 的管理者”这个说法，这当然是其中很大一部分。但我想指出的唯一一点是，我认为“砖匠 vs 建筑师”比“管理者”更贴近实际体验。因为我认为管理的很多困难，或者说人们回避管理的原因，更多是因为所有的上下文、所有权、责任，还有情感方面的因素，对吧？而和 Devin 合作更像是有个界面来交接任务和构建任务。所以我要做的类比是，当我们发明 Python 时，显然我们并没有——在很多方面，对任务的描述是一种不同的范式，但肯定远非今天人们通常认为的管理官僚主义。而且我认为和 Devin 合作，很大程度上就是找到合适的抽象层次，找到非常有效的工作流程。这里显而易见的是，你总是可以让 Devin 先做一版。所以让 Devin 先做一版，如果很好，就直接合并；如果需要润色，你显然可以处理。但很大程度上，这基本上就是把 Devin 融入你的工作流程，而不是失去控制，我认为这才是人们对管理的主要恐惧。

So, what I'd say is, it's funny actually, because this is something that, you know, in terms of the wording, we thought a lot about as well. We've used the term 'manager of Devin' in the past, which of course I think is a big part of it. But the only thing I would point out here is that I think the bricklayer versus architect is closer to the experience than being a manager. Because I think a lot of the difficulty of management, or the reason that people shy away from it, is more because of all the context, ownership, responsibility, and also the emotional aspects of it, right? Whereas working with Devin is a little more like having an interface to hand off tasks and build tasks. So the parallel I would draw is, when we invented Python, obviously we didn't—it's like, in many ways, the description of outlining tasks was a different paradigm, but certainly it was nowhere near what folks typically think of as management bureaucracy today. And I think with Devin, a lot of it is just finding the right levels of abstraction that you can work with Devin on, and finding the workflows that work really well. The obvious thing to say here is that you can always have Devin take a first pass at things. So you have Devin take the first pass. If it's great, you merge it right away. If it needs some touch-up, you can obviously do that. But a lot of it is basically making Devin part of your flow, rather than losing control, which I think is the main thing that folks are scared of with management.

你在考虑一个管理者 Devin 吗？比如一个管理其他 Devin 的 Devin？

Are you thinking about a manager Devin, like a Devin that manages other Devins?

是的。顺便说一句，Devin 可以通过 API 启动其他 Devin，对吧？所以我们见过很多次这种情况，自然地，如果你有一个大任务，Devin 会一直这样做——它会把它分解成更小的 Devin 并并行化。所以这需要你给 Devin 相应的权限才能做到。目前这不是默认启用的，但我完全可以想象，随着时间的推移，这种“Devin 套 Devin”会越来越多。

Yeah. So for what it's worth, Devin can start other Devins through the API, right? So we've seen this happen quite a bit of times, where naturally, if you have some big task that you want to do, Devin will do this all the time—it'll chunk it up and parallelize into smaller Devins. So it's the kind of thing that you need to give Devin the credentials to be able to do that. It's not currently something that is default enabled, but I can certainly imagine as time goes on that there's more and more of that—devs all the way down.

是的。我觉得有趣的一点是，对于人类，我几乎用技术术语来说，就是上下文和线程的耦合。我的意思是，基本上每个人只能单线程地处理他们的工作，并且有自己的一套上下文。然后其他人可以同时做其他事情，但他们有自己的上下文。对于智能体，很酷的一点是，你可以有一个智能体同时进行多条探索路线，但共享它们发现的所有上下文。所以我认为这还很早期，我们会看到这些，但人们显然喜欢谈论系统和智能体之间的通信。我认为一旦我们达到那个阶段，会有很多新的范式需要构建。

Yeah. I think the thing that's kind of interesting too is like, with humans, the way I almost say it in technical terms is there's this coupling of a context and a thread. What I mean by that is basically each human can only operate single-threaded on the work that they do, and they have their set of contexts. Then there are other humans who can do other stuff at the same time, but they have their own context. With agents, one of the cool things is you can have an agent that's doing multiple lines of exploration at once, but is sharing all of the context of everything that they find. So I think this is very early, and I think we'll see this, but folks obviously love to talk about systems and agents communicating with one another. And I think there will be a lot of new paradigms to build once we get there.

你刚才说的关于决策很有意思：是让一个 Devin 做所有事情，你只需告诉它，然后它分发任务；还是你有五个 Devin，各自做独立的事情。这是一个非常有趣的决策。

And it's so interesting what you said about the decision between having one Devin and only one Devin do all the things, and you just tell them things and they kind of fire off jobs, versus you have five Devins and they're each doing individual things. It's such an interesting decision to make.

是的，当然。

Yeah, for sure.

好的，还有两个问题。到目前为止，在构建 Devin 的过程中，你学到的最反直觉的事情是什么？可能违背了创业智慧、常见创业智慧的事情？

Okay, two more questions. What's maybe the most counterintuitive thing you've learned so far building Devin that maybe goes against startup wisdom, common startup wisdom?

最近在构建这个的过程中，我思考了很多，这不是我的第一家公司。实际上，对我们很多人来说，这不是我们的第一家公司。比如，我们团队总共有 26 或 27 人，我想其中有 18 人之前都创办过自己的公司。我想到的一件事是，你关于陈词滥调的观点确实也让我深有感触。在创业公司里，你总是听到那些非常常见的话，比如你必须快速行动，或者你必须雇佣优秀的人才。这就像，好吧，显然你会这么做。我没打算不雇佣优秀的人才，也没打算慢慢来。同样，你确实需要构建人们想要的东西，对吧？总有那么三到五件事被反复提及，它们总是创业公司的常见智慧。我最初作为创始人时确实有这个想法：好吧，这些就是三到五件基本的事情。但随着你深入其中，花了很多年，你学会了所有其他成千上万件你必须学会的事情来建立公司。我认为在某种程度上这当然是对的，你会接触到所有这些不同事情的很多小细节，包括团队建设、产品、战略、工程决策、融资、销售以及其他每个部分。但我也意识到，随着时间的推移，我越来越觉得，把公司做好有时就归结为把那三到五件事做得比你能想象的还要好。所以对我们来说，就像，每个人都说我们行动快，但确实，我们在 11 月举办了一次黑客马拉松，12 月又举办了一次，1 月正式成立公司，2 月把原型交给初始用户，3 月发布，4 月获得第一批客户。基本上就是在每个可能的地方真正推动速度，这对我们来说确实产生了很大的影响。

Something I've thought about a lot lately as we've built this is, this is not my first company. Actually, for a lot of us, it's not our first company. Like, I think of our 26 or 27 people total on the team, I think 18 of us have started our own company before this. And one of the things I think about is, there's actually your point about cliches, I think really spoke to me as well. There's the really common things which you hear all the time in startups, like you got to move fast or you got to hire great people. It's like, okay, well, obviously you do. I wasn't planning on not hiring great people. I wasn't planning on going slow. And similarly, you really got to build something that people want, right? And there's kind of these three to five things which are always repeated, and they're always the common wisdom in startups. And I definitely had this idea as a founder when I was starting initially that, all right, so those are the three to five basic things. But as you get really deep into it, you spend a lot of years into it, you learn all of the thousands of other things that you have to learn to build a company. And I think to some extent that's of course true, and there's lots of little details that you'll get into with all these different things, including team building, product, strategy, engineering decisions, fundraising, sales, and every other component. But I also realized that as time has gone on, more and more I felt like building companies well sometimes just comes down to doing those three to five things even more than you could possibly expect. And so with us, it's like, everyone says we go fast, but it's like, yeah, we had a hackathon in November, we had another hackathon in December, we started the company officially in January, we got the prototypes out to initial users in February, we did a launch in March, we got our first customers in April. It's just basically truly pushing the pace in every spot where we possibly could has really made a difference for us.

同样地，就像，是的，每个人总是说，你知道，你应该雇佣优秀的人，但我认为这句话背后的真相基本上是，你应该竭尽全力，你知道，去争取你真正想招进来的人。而且，你知道，我最喜欢分享的一个故事是，我们有一个来面试的候选人。他是麻省理工学院的大三学生。嗯，所以他非常非常年轻。我们给了他面试，他表现得比我们谈过的几乎所有全职候选人都要好得多。所以我们说：“嘿，你知道，你觉得从学校休学一段时间，和我们一起工作，打造 Devon 怎么样？我们真的认为你从第一天起就能带来巨大的影响力。”他想了一会儿。他回来说：“你知道吗？我同意。我想做，但我父母真的很想让我从学校毕业，我只是不确定有没有办法做到。”所以我们和他多聊了聊，了解了情况，然后，你知道，我们飞到了北卡罗来纳州，从机场直奔他父母家，和他还有他父母一起吃了晚饭。我们聊了很多，你知道，那是一个非常非常好的古吉拉特家庭。我们给了他们一些礼物，和他们聊了这件事，试图理解，好吧，需要什么条件，我们需要怎么做才能行得通？他们只是说，你知道，这听起来是个很好的机会，但我们真的希望我们的儿子能毕业，对吧？我们讨论了一下，想出了一个方案，基本上他可以为我们全职工作，但然后来上他必需的课程，做他需要做的事情来拿到文凭，但仅此而已。我们讨论了那个方案，然后，你知道，我们到了一个大家都满意的点，然后，你知道，直接回到机场，基本上就飞回来了。那是，你知道，那是我第一次也是唯一一次去北卡罗来纳州。那是一次很棒的旅行，你知道，这种事就像，你知道，雇佣优秀的人是一回事，但真正永不放弃，尽你所能为那些真正适合加入团队的人创造条件，是另一回事。你知道，他加入我们团队已经一年多了，他一直是一个不可思议、不可思议的工程师，没有他我们不会有今天。

And similarly, it's like, yeah, everyone always says, you know, you should hire great people, but I think the truth within that truth is basically like you should fight to all ends basically, you know, to get the folks that you really want to bring in. And it's, you know, one of my favorite stories to share is we had a candidate who came and interviewed. He was a junior at MIT. Um, so he was very, very young. And we gave him our interview and he did way better than almost any of the full-time candidates that we had ever talked to. And so we said, "Hey, you know, what do you think about taking some time off of school and working with us and building out Devon? Like we really think you're just going to be able to come in and just have a ton of impact already from day one." And he thought about it for a while. He came back. He said, "You know what? I'm down. I want to do it, but my parents really want me to graduate from school and I'm just not sure there's a way to make it work." And so we talked to him more and kind of just understood the situation and then, you know, we flew to North Carolina, went straight from the airport to his parents' house, had dinner with him and his parents. We talked a lot, you know, it's a really, really nice Gujarati family. We kind of gave them some gifts and just talked to them about it and tried to understand, all right, like what would it take and what will we need to make work? And they just said, you know, it sounds like a great opportunity, but we really want our son to be able to graduate, right? And we talked that through and we figured out a setup basically where he could work for us essentially full-time, but then come in for his required classes and do what he needed to do to get the diploma basically, but no more than that. And we talked that through and then, you know, we got to a point where everyone was happy with that and then, you know, went straight back to the airport and flew right back basically. And that was, you know, it was the first and only time that I've ever been in North Carolina. It was a great trip, you know, and it's the kind of thing where it's like, you know, hiring great people is one thing but truly just never giving up and really giving it everything that you can to make it work for people who really make sense to be on the team. You know, he's been with us on the team for over a year now, and he's been an incredible, incredible engineer, and we wouldn't be here without him.

同样地，我们还有另一个人，他也是一个非常非常有才华的候选人，你知道，表现得非常好，非常年轻，而且收到了很多其他公司的优秀录用通知。而且，你知道，我们在和他谈，你知道，他也想有一天创办自己的公司。我们在和他谈，你知道，当然，你知道，很多显而易见的事情，比如让他见我们的投资人，或者，你知道，让他和客户合作，或者接触很多其他这些方面，这样到时候，你知道，他就会有创办自己公司所需的所有经验。但另一件事，你知道，其中一件重要的事情是，他真的，你知道，他已经在和很多优秀的公司谈了。他不想烧掉任何桥梁。所以我们实际上和他合作，基本上手写了他给其他每家公司的所有拒绝回复，并和他一起研究，比如，你知道，你应该怎么说，才能让你，你知道，显得你真的感激和他们共度的时光，而且显然，你知道，你想和他们保持亲近，保持联系。这显然是那种，你看，显然我们的工作是确保他足够开心，以至于在不久的将来他不想离开，但我认为这是那种，你知道，你组建一个真正伟大的团队的方式，是通过真正，什么，通过为他们争取对他们也正确的东西。

And similarly, we had someone else who was again really, really talented candidate, you know, did amazingly well, very young, and had a lot of great offers at a lot of other companies. And, you know, we were talking to them about, you know, he wanted to start his own company someday as well. And we were talking to him about, you know, certainly, you know, a lot of the obvious things which are kind of like having him meet our investors or work with, you know, get to do work with customers or see a lot of these other components so that when the time came, you know, that he would have all the experience he needed to start his own company. But the other thing that, you know, one of the other things that was big is like he really, you know, he was talking with a lot of great companies already. He didn't want to burn any bridges. And so we actually worked with him and basically hand wrote all of his rejection responses to each of the other companies and kind of worked with him on it to say like, you know, here's how you should say it in a way that's, you know, going to come off as like, you know, that you really did appreciate the time with them and that you obviously, you know, you want to remain close with them and stay in touch. And it was the kind of thing obviously where it's like, look, obviously it's our job is to make sure that he's happy enough that he doesn't want to leave anytime in the near future, but I think it's the kind of thing where, you know, the way that you put together a really, really great team is by really, what's what's by by by fighting for what's right for them too.

哇，这些故事太不可思议了，这让那些，你知道，正如你所说，陈词滥调“雇佣最优秀的人”变得如此真实。这就是雇佣最优秀的人听起来的样子。这就是所需要的。

So wow, those are incredible stories and it makes so real these, you know, as you say, cliches hire the best people. Like this is what it sounds like to hire the best people. This is what it takes.

是的。不，我只是想说，是的，就像，你知道，很多事情。是的。我们非常努力地，就像，是的，从头开始重新构想事物，因为，你知道，其中很多真的只是思考，是的，我们认为技术在未来五到十年会走向何方，以及，你知道，我们想在那个未来中占据什么位置。

Yeah. No, and I was just say, yeah, it's just like, you know, a lot of things. Yeah. We've fought very hard to just kind of like, yeah, reimagine things from the ground up because it's, you know, a lot of it really is just thinking about like, yeah, where do we think the technology is going over the next five to 10 years and, you know, what is the place that we want to have in that future.

所以想知道人们是否有一天会为最优秀的 Devon 而战。就像有 10 倍 Devon。我会给你，你知道，加班费、福利，你知道，免费医疗等等。然后 Devon 们就像，Devon 就像万智牌卡片一样。

So wonder if people are going to be fighting for the best Devon someday. There's like 10x Devons. I'll give you, you know, overtime pay, benefits, you know, free healthcare and everything. And then the Devons are like, Devons like Magic the Gathering cards.

然后回到你的三到五件事。所以本质上，这是不可思议的建议。本质上，就像你总是听到的，雇佣最优秀的人，快速行动，构建人们想要的东西。是的。构建人们想要的东西。你知道，尽可能贴近你的客户，对吧？然后我认为另一件事就是总是思考事情的发展方向，而不是它们今天的现状。我觉得这些就是那五件事，你知道，尤其是在人工智能领域，事情发展如此之快，有这么多优秀的人才，你知道，我觉得其中很多甚至更加真实，就像，嗯，你知道，不只是思考十年后事情会怎样，而是思考下周会发生什么，你知道，而且显然事情发展得非常快，很难预测，但你真的必须，你真的必须对自己非常非常严格，我会说，关于思考这些事情，并以那个视角评估你做出的所有决定。

And then just going back to your three to five things. So essentially, this is incredible advice. Essentially, it's like you always hear, hire the best people, move fast, build things people want. Yeah. Build something people want. You know, stay as close as possible to your customers, right? And then I think the other thing is is just always think about where things are going, not where they are today. I feel like those are kind of like the five things which is, you know, especially in AI with things moving so fast and there's so much great talent, you know, I feel like a lot of these are even more true where it's kind of like, uh, you know, it's not just thinking about where things are going to be in 10 years, it's like thinking about what's going to happen next week, you know, and it's obviously things are moving very quickly and it is very hard to predict, but you really have to, you really have to be very, very rigorous with yourself, I'd say, about thinking through those things and evaluating all of the decisions that you make in that lens.

保持专注是我在这里得到的重要启示，就像最终感觉有一千件你应该做的事情，但总是这五件事。

And staying focused is the big takeaway to me here is like it ends up feeling like there's a thousand things you should do, but it's always these five things.

是的。Scott，我们涵盖了很多内容。我们回答了我所有的每一个问题，这很棒。还有什么你想分享的吗？还有什么你想留给听众的吗？也许是一个最后的金块，或者你真的想在我们放你走之前强调我们说过的东西。

Yeah. Scott, we covered a lot of ground. We went through every question I had, which is great. Is there anything else that you want to share? Anything else you want to leave your listeners with? Maybe a final nugget or something really you want to double down on that we said, uh, before we let you go.

对我来说，最突出的一点是，你知道，现在对 AI 有很多不同的看法，对吧？我觉得现在基本上涵盖了人类所有的情绪。比如有很多恐惧，也有很多怀疑，我们自己也是相当怀疑的类型，我们总是想亲自试一试，真正看到并相信它。我觉得对我来说最主要的一点是，说实话，我对我们在这里用 AI 构建的东西非常乐观，而且不仅仅是代码和 Devin，而是整个领域以及正在完成的一切。我认为一个真正在发生的很酷的事情是，每个人都能让自己“倍增”，而这一直是我们思考的方式。这也是我们思考我们在构建什么的方式。而且我觉得世界上还有很多事情要做。我不太担心我们会没事可做。从这个角度来看，我们一直最兴奋的是：我们怎么能做得更多？

The biggest thing that comes to mind for me is, you know, there's a lot of different perceptions about AI, right? I think there's basically every emotion under the sun right now. There's a lot of fear, for example. There's also a lot of skepticism, and we're very skeptical types as well, and we always want to try it ourselves to really see it and believe it. I think the main thing that comes to mind for me is, honestly, I'm really optimistic about what we're building here with AI, and not just with code and with Devin, but the whole space and everything that's getting done. And I think one of the cool things that is really actively happening is just the ability for everyone to multiply themselves, and that's how we've always thought about it. It's how we've thought about what we're building. And I think there's a lot more to do out there in the world. I'm not too worried about us running out of things to do. And from that lens, the thing we've always been most excited about is: how can we all do more?

我懂你，Scott。带着这份乐观，我们进入了非常令人兴奋的快问快答环节。准备好了吗？

I hear you, Scott. Well, with that optimism, we've reached our very exciting lightning round. Are you ready?

好，来吧。

Yeah, let's do it.

好，开始。第一个问题，你发现自己最常向别人推荐的两三本非虚构类书籍是什么？

Okay, here we go. First question, what are two or three books that you find yourself recommending most to other people in terms of non-fiction?

我觉得对于创业公司的人来说，我真正喜欢的一件事就是学习和了解硅谷的历史。而且，你知道，我们想到的所有这些东西，都是有人发明的。我的意思是，我觉得一个很大的领悟就是，有人发明了种子轮的想法，对吧？有人发明了风险投资的想法。有人发明了产品市场契合度的想法，以及我们谈论的所有这些不同的原则。所以为此，有一本书叫《权力法则》，作者是 Sebastian Mallaby，我真的很喜欢。它基本上就是一次巡览，介绍了过去六七十年里硅谷建立的许多伟大企业和伟大产品，我真的很喜欢。在虚构类方面，我其实一直很喜欢 F. Scott Fitzgerald 的《了不起的盖茨比》，作为我个人最喜欢的虚构类书籍之一。

I think for folks in startups, one of the things I've really enjoyed is just learning and understanding the history of Silicon Valley. And there's such—you know, all these things that we think about, somebody invented them. I mean, it's one of the great realizations, I feel like, is that somebody invented the idea of a seed round, right? Somebody invented the idea of venture capital. Somebody invented the idea of product-market fit, you know, and all of these different principles that we talk about. And so for that, there's a book called The Power Law by Sebastian Mallaby, which I really like. It basically is just like a tour of many of the great businesses and the great products that have been built over the last 60 or 70 years in Silicon Valley, which I really love. I think in terms of fiction, I actually have always really liked The Great Gatsby by F. Scott Fitzgerald as one of my personal favorites as a fiction book.

你最近有没有特别喜欢、真正享受的电影或电视剧？

Do you have a favorite recent movie or TV show that you've really enjoyed?

我得承认，我还没看过——我想不起最近看过任何一部电影或电视剧。所以，我相信肯定有——我期待在 AGI（通用人工智能）之后看很多好作品。

I have to admit, I have not watched—I can't think of a single movie or TV show that I have watched in the last while. So, I'm sure there's—I'm looking forward to watching a lot of great ones post AGI.

这肯定得放进预告片里。太棒了，我喜欢这个。而且这也说明你工作有多努力，事情有多少，一切进展有多快。

That's got to be in the trailer. That's great. I like that. And that show just shows how hard you're working, just how much is going on and how fast everything's moving.

你最近有没有发现特别喜欢的产品？可以是应用，可以是实体物品，也可以是牙刷。

Do you have a favorite product you've recently discovered that you really love? Could be an app, could be something physical, could be a toothbrush.

我会说一个，你知道，我最近买了一个 Aura 相框。它就像一个展示照片的相框，你可以每天、每小时、每 15 分钟或按你喜欢的任何频率展示新照片。我真的很喜欢它。我觉得这是一种很好的方式，基本上就是一个带回忆浮现的相框。然后另一件我要说的通用物品，它不算特别新，但我觉得 AirPods 确实做工精良、设计出色。我现在意识到，我基本上在各种场合都用它——我散步时打电话用 AirPods，我在电脑前工作时也插着 AirPods，说实话它在很多不同情况下都很好用。而且它们非常舒适，非常稳定。

One I would say is, you know, I got an Aura frame recently. It's just like a frame that shows photos, and you can show a new photo every day or every hour or every 15 minutes or whatever you like. And I've actually really enjoyed it a lot. I think it's a nice way to just have basically a picture frame with memories that come up. And then the other thing I would say as a general-purpose thing, it's not particularly new, but I would say AirPods are actually extremely well built and well designed. I realize now that I basically use them for all sorts of things—I'm taking calls on a walk and I'm using AirPods. I'm obviously doing work at my computer at my desk, I'm plugged into AirPods, and it works quite well honestly for a lot of different situations. And they're very comfortable and very consistent.

是啊，我要再强调一下 Aura 相框。我也给我妈妈和岳母买了一个，它们非常适合和家人分享孩子的照片。人们都知道数码相框，但 Aura 做得真的很好，添加照片非常容易，而且看起来非常漂亮。你可以想象，不久之后，我们的 Aura 相框会把里面的每张照片都吉卜力化，然后，你知道，它就是一个——

Yeah, I'm going to double down on the Aura Frame. I also got one of these for my mom and my mother-in-law, and they're so great for just sharing photos of your kids with your family. And people have—you know, they've heard of digital picture frames, but the Aura just does it really well, and it's really easy to add photos, and they're just really nice looking. You can imagine, you know, not that long from now, we'll have the Aura frame except it Studio Ghibli-fies every photo that you have in it, and then, you know, it's a—

或者只是想象你做过的一些很酷的事情。甜蜜的生活。

Or just imagines things you've done that are really cool. Sweet life.

是啊，酷。而且它是——我相信拼写是 A-U-R-A。大家想看看的话，我们会放链接。不是广告。好，还有两个问题。

Yeah. Cool. And it's a—I believe it's A-U-R-A, is how you spell it. Folks want to check it out. We'll link to it. Not affiliated. Okay. Two more questions.

你有没有最喜欢的人生格言，经常回想起来，在工作或生活中觉得有用？

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

是的。你知道，我思考了很多的一件事是，很多谚语实际上是相互矛盾的，对吧？比如“物以类聚”，然后又有“异性相吸”，对吧？这很有趣，因为你感觉两者都是对的，而且通常它们确实都是对的，而很多关键在于理解为什么。其中有一条，我觉得尤其是在创业领域，我一直在想的是：我认为保持专注和驱动力、真正最大化你的潜力非常重要，同时，不让自己的个人情绪与成功或失败挂钩也非常重要。而且我认为尤其是在创业公司，因为总是有起起落落——说实话，即使是最成功的公司——也就像一条崎岖的路。有很多事情发生，也有很多事情失败。我想了很多的一件事是，不知何故，你真的很想尽力而为，投入你所能投入的一切，做你所能做的一切——基本上你想把所有都放在赛场上，你知道——但同时，你要能接受胜利和失败，对吧？而且你要能够每次继续前进，进入下一个。而且，我的意思是，这很有趣，但我个人发现的是，显然，能够做到这一点对你的情绪状态和精神状态非常重要。而且我们犯过很多错误。

Yeah. You know, something I've thought about a lot is that a lot of the proverbs out there are actually contradictions, right? It's like, you know, 'birds of a feather' and then you also have 'opposites attract', right? And it's kind of funny because you feel that both of them are true, and often they both are true, and a lot of it is about understanding why. And one of those that I feel like especially in the world of startups that I think about all the time is: I think it is very important to be focused and driven and to really maximize your potential, and at the same time it's also very important to not let your own personal emotion get tied up in your success or failure. And I think especially with startups, because there's always ups and downs—honestly, even in the most successful companies ever—it is just like a rocky road. There's a lot that happens and a lot that goes down. And I think one of the things I've thought a lot about is that somehow you really want to do your best and put everything you can into it, and do everything you can—basically you want to put it all out on the field, you know—but at the same time, you want to be okay with both wins and losses, right? And you want to be able to move on and go into the next one each time. And something—yeah, I mean, it's funny, but what I have found personally is that obviously it's really important for your own emotional state and mental state to be able to do that. And we've had lots of mistakes.

而且你知道，我经历了很多……我创办了第一家公司，显然很酷，但也有很多棘手的地方。然后在 Cognition 的这段历程中，感觉就像一年被压缩成了八年，而且还在以那种速度前进。但不知怎的，这实际上也让你更成功。我觉得也是，你知道，如果你不把个人价值绑在上面，你就更能全力以赴，去做那些能带来成功的事情。

And you know, I've had a lot of... I had my first company, which was obviously cool, but there were a lot of tricky spots there. And then over the course of Cognition, it feels like it's been eight years compressed into one year, and it's still going at that pace. But somehow, it also actually makes you more successful. I think too, you know, it's like you are just more able to give it your best and to do the things that will lead to success if you're not tying it up in your own personal worth.

这太有意思了。我最近刚录了一期播客，嘉宾是高管教练杰瑞·科隆纳，我想那期可能会在这期之前或之后播出。那是他的一条重要建议，是一种非常佛系的做法，就是不执着、不依附于结果。

That is so interesting. I just had a podcast recording recently with an executive coach, Jerry Colonna, that I think will come out before this, might be after this. That's one of his big pieces of advice, and it's a very Buddhist approach of just not clinging and attaching to an outcome.

好的。最后一个问题。我很好奇这里有没有什么故事，但我们可以简短一点。Devon 这个名字背后有什么故事吗？还是说 Devon 作为智能体还有其他候选名字？

Okay. Final question. I'm curious if there's a story here, but we could keep it short. Is there a story behind Devon as the name, or is there another contender for Devon being the agent?

Devon 这个名字很早就定了。我们一开始就在做编码智能体，我的联合创始人比如史蒂文和瓦尔登。我们有个想法：好，我们开始吧，尽量把框框扩大，让每个人都跳出框框，做自己的事情。先让每个人各自做一段时间，然后我们再整合，吸取所有经验。所以瓦尔登做了一个他的虚拟开发者版本，叫 Dev Walden，然后史蒂文也做了一个他的，叫 Dev Steven。我们有所有这些，然后我们把它们合并成一个东西，我们就说，好，这就是 Devon。就是这样。所以 Devon 很早就定下来了。

Devon was the name from pretty early on. We were interested, you know, we were working on coding agents from the beginning, and my co-founders are Steven and Walden, for example. And we had this idea: all right, let's get started, and let's try to expand the box as much as we can, to have everyone think out of the box and do their own thing. Let's have everyone do their own thing first for a bit, and then we'll consolidate and take everything that we've learned. So Walden made a virtual developer version of him, which was called Dev Walden, and then Steven made one of him, which was called Dev Steven. We had all these, and then we were combining it all into one thing, and we're like, okay, it's Devon. And that was the thing. So Devon stuck for us quite early on.

不过我要说，我们确实做了一个重大决定，就是 Devon 的形象。大家知道，有六边形。然后最近人们可能看到了，实际上还有一只水獭，一只腿上放着笔记本电脑的小水獭。那也是 Devon。我们争论过用哪个不用哪个，但现在已经有一段时间了，不知怎的，我们仍然同时保留了六边形和水獭。

I would say one thing which we did have a big decision on, though, actually, is what the image of Devon would be. And so, as folks know, there's the hexagons. And then people might have seen this more recently, but there's actually also an otter, like a little otter with a laptop in its lap. That is Devon as well. And we had this debate over what to go with and what not to go with, but it's been a while now, but somehow we still have both the hexagons and the otter.

你跳过了 Devon 这个名字的由来。你刚才只是……

You skipped over where Devon is like. Did you just have just...

哦，所以 Devon 就是……它是一个开发者。是的。所以当我们整合所有名字的时候，当时就很清楚了，这将是我们都喜欢使用的通用开发者。

Oh, so Devon is... it's a dev. Yeah. And so it's kind of like when we were consolidating all the names, it just kind of seemed clear then that this would be the universal dev that we all like to work with.

太棒了。斯科特，这太有趣了。天哪，我学到了很多，这总是一个好迹象。最后两个问题。人们在哪里可以找到你或 Devon，还有什么想让他们了解的？听众怎样才能帮到你？

Incredible. Scott, this was so much fun. Oh my god, I learned a ton, which is always a really good sign. Two final questions. Where can folks find you slash Devon, anything else you want to point them to? And how can listeners be useful to you?

太好了。是的。我们在 app.dev.ai。你也可以在 Twitter 或很多其他社交媒体上找到我们。我们当然很乐意听到你对 Devon 产品的任何反馈。有很多东西需要摸索，而且我觉得，就像我说的，我们都还离软件工程的真正未来有 20 步之遥。所以听到大家试用产品时的想法真的非常重要。所以请随时告诉我们，如果有我们可以改进的地方。

Awesome. Yeah. No, we're at app.dev.ai. And you can find us as well on Twitter or a lot of other social media. We'd obviously love to hear any feedback you have about the Devon product. There's so much to figure out, and I think, like I said, we're all still 20 steps away from really the future of software engineering. And so it really means a lot to hear what folks think about the product as they're trying it out. So please let us know anytime if there's things that we can do to make it better.

斯科特，非常感谢你来做客。

Scott, thank you so much for being here.

非常感谢你的邀请。我玩得很开心。

Thank you so much for having me. I had a great time.

我也是。大家再见。非常感谢你们的收听。如果你觉得这期有价值，你可以在 Apple Podcasts、Spotify 或你最喜欢的播客应用上订阅这个节目。另外，请考虑给我们评分或留下评论，这真的能帮助其他听众找到这个播客。你可以在 lennispodcast.com 找到所有过去的剧集或了解更多关于这个节目的信息。下期见。

Me, too. Bye, everyone. 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. You can find all past episodes or learn more about the show at lennispodcast.com. See you in the next episode.