GitHub 首席运营官解释为何 AI 尚未取代开发者

你好，我是迈克·泰勒。

Hi, I'm Mike Taylor.

我是Every公司的技术咨询主管，我最近和GitHub的首席运营官凯尔·戴格尔聊了聊，谈谈编码代理在前线到底发生了什么。

I'm the head of tech consulting at Every, and I sat down with Kyle Daigle, the COO of GitHub, and talked to him about what is happening on the front lines of coding agents.

现在GitHub每个月要处理一千七百万次轮询请求，增长速度非常快，这让他们站在了新经济浪潮的最前沿。

We have 17,000,000 poll requests coming in every month to GitHub now, it's growing exponentially and that puts them at the forefront of what's happening in this new economy.

我们讨论了这些变化对用户的影响，也谈到了对开源维护者带来的冲击，还聊了一个大家都特别关心的话题。

We talked about how that affects our users as well as how this affects open source maintainers and we covered a topic which is dear to everyone's hearts.

我该怎么防止每月200美元的编码代理订阅，突然涨到每月2000美元的使用限额？

How do I stop my $200 a month coding agent subscription from ballooning into a $2,000 a month usage limit?

这次采访我做了一点不一样的事——我告诉凯尔，我用AI模仿了他的声音来提前练习采访，结果他反过来给了我一个意想不到的回应。

In this interview, did something a little bit different, which is I told Kyle I had made an AI clone of him to practice the interview, and he revealed something surprising in return.

下面是我们的对话。

Here's the conversation.

Every是你在AI前沿保持领先的唯一订阅服务。

Every is the only subscription you need to stay at the edge of AI.

如果你关心紧跟最新模型和使用最新工具，那就得订阅Every，才能从海量信息中抓住真正有价值的内容。

If you care about being on top of the latest models and using the latest tools, you have to subscribe to Every to separate out the signal from the noise.

现在就去 every.to/subscribe 订阅吧。

Go to every.to/ subscribe today.

嗨，凯尔。

Hey, Kyle.

谢谢你在这次大会上抽时间跟我聊聊。

Thanks for spending some time with me at the conference.

是的。

Yeah.

当然。

Of course.

对。

Yeah.

前一天在会上见到你也很开心，没错。

It was good to meet you as well in the the day before, and Yep.

我觉得我们之前大概已经聊过这些问题了，但为了让更广泛的听众了解这里的情况，再梳理一遍还是有必要的。

I feel like we already kind of covered a few of these questions, but I think it'll be good to you know, for the wider audience so they can understand what's going on here.

没错。

Absolutely.

是的。

Yeah.

首先，我们之前谈到的一个很有意思的现象是，客户的群体结构正在发生变化，对吧？

So first thing I think is we were talking about is really interesting is that the demographics of the customer are changing, right?

以前根本不会用 GitHub 或任何开发者工具的人，现在也开始用了。

Like a lot of people who previously would maybe never use GitHub or never used developer products before are now using them.

那这对你们制定产品路线图产生了哪些影响？

So how has that changed the way that you decide the product road map?

是的。

Yeah.

是的。

Yeah.

我的意思是，对于 GitHub 来说，我们一直对‘开发者’这个概念有着非常宽泛的理解。

I mean, you know, I think, for GitHub in particular, we've always really had this really expansive view of what a developer is.

我刚开始写代码的时候，根本不会觉得自己是个开发者，那时候只是自己写点东西，后来还转了完全不同的职业方向。

I started as a developer before I would have ever called myself a dev, you know, where I was just like writing code, it was just for me and I went personally like a completely different career path.

我没上过计算机科学专业。

I didn't go to school for computer science.

我当时去的是艺术学校。

I was going to art school.

我写代码是为了赚艺术学校的学费，现在回头看，这个决定确实挺傻的。

I wrote code to pay for art school, which was a very silly, decision as an adult now, I guess.

但那时候，我就想着，我能和团队一起做出工具，让别人也能有同样的体验——比如只想做个给自己或家人用的 App，不管是当创业项目还是正经生意都行。

But, like, at then, you know, that sort of journey of just, like, I can create tools with the team and deliver them to people who can have that same experience of like, I just wanna build an app that's for me or for my family, maybe as a startup, maybe as a business.

我们当然也有非常专业的开发工具，大型企业都在用 GitHub。

We very much have like serious developer tools, know, like all the largest businesses are using GitHub.

但当我看 GitHub Copilot 这类应用时，我发现用 AI 的人里，每天同时跑好几个项目、开一堆智能代理会话的，其实也大有人在。

But when I look at something like the GitHub Copilot app, I see just as many, you know, developers that are using AI every day, running multiple projects, like all all kinds of agent sessions at the same time.

我还看到，GitHub 的法务团队、财务团队都在用 GitHub Copilot，今天我见客户时，他们也提到同样的情况。

And I see, you know, our legal team at GitHub using the GitHub Copilot app or the finance team, or I was meeting with a customer today and they were saying the same thing.

很多通常被业界称为知识工作者、并非专业开发的人，其实都在用这些工具为自己开发小应用或创建资源。

A lot of the folks that, you know, the industry would call knowledge workers or just, you know, non by trade developers are using these tools to build little apps or assets for them.

所以，虽然我们的重点始终是开发者，但我们希望让更多人能轻松尝试写代码，确保每个人都能顺利入门——比如通过 GitHub Copilot 这样的工具来接触编程。

And so while our focus is very much on developers, I think we wanna make it easier for people to choose to try to, you know, write some code, and make sure there's always an on ramp into, you know, writing some software now with, you know, things like the GitHub Copilot app.

对。

Yeah.

那你们怎么应对，或者怎么帮开发者应对这种额外的负担？现在 PR 数量太多了，我接触的开源维护者都说快被淹没了。

And then how do you deal with the or how do you help developers deal with the burden of all of that extra you know, there's a flood of PRs now, open source maintainers I talked to are, like, drowning.

你们觉得，到底该怎么做才能真正帮到他们？

How do you how do you like, what what needs to happen to help them?

对。

Yeah.

我觉得，对所有开发者来说，我们正在开发像 Copilot 代码审查这样的工具。

I mean, I think for all developers, we're, you know, building tools like the Copilot code review.

现在它已经具备自主行动能力了，能发现更多新颖的漏洞，你只要提个评论，代理就会自动接手并完成修改，如果你想的话。

It's now agentic, so it finds a lot more novel vulnerabilities and you can just, like, comment and the, you know, the agent will take that on and go implement the change if you want to.

我觉得代码审查这个环节其实被低估了，它真的是让 PR 更容易被审核的好方法。

So I think that code review step is, in some ways, like, overlooked as a really great way to get PRs to a place that are much more easily reviewed.

我觉得应用里的自主合并功能也是一个重要环节，我们在内部和社区里经常看到这种情况：你可能会对某处提出意见。

I think that the agentic merge in the app is another place where we see a lot of times internally and in the community, you know, you may comment on something.

你做了代码审查，几乎把所有问题都解决了，但最后还得手动完成一堆琐碎步骤才能把 PR 处理完。

You that might have a code review, and you might go through and get it almost all the way there, but then there's all those, like, manual steps just to finish processing the PR.

对的。

Mhmm.

相反，我可以直接设定好 GitHub Copilot 的权限，告诉它：好了，就按这个来。

Instead, I can go in and set exactly what I want to allow, you know, GitHub Copilot to do and say, okay.

现在去合并这个 PR，等 CI 跑完，再检查所有策略要求。

Now go merge this PR and wait for CI and wait for policies and all of that.

我觉得这才是关键所在。

I think that's a big part.

在开源方面，需求比较独特，因为你没法控制谁在提交代码，或者说以前一直都没法控制。

On the open source side, it's a unique set of needs because you don't control who's, you know, sending everything in, or you haven't really historically.

这也是我们一直关注的重点，就是给维护者提供更多工具，让他们来决定是否接受这些 PR。

And that's been really where we've been focusing is giving maintainers more tools to decide, well, do you want to accept all of these PRs?

比如，你希望接受来自哪些人的 PR？

Like, who do you want to accept them from?

还有，你需要做多少工作来证明你的贡献对这个项目是有意义的？

You know, how much work do you need to do to kind of prove that you're going to contribute something that is gonna be meaningful to this project?

这也是我们想为开源维护者提供工具，但真正让他们掌握控制权的地方。

And that's something that we wanna provide tools to open source maintainers, but really leave them in control.

每个社区都在选择略有不同的方式来解决问题。我们始终希望把决定权交给他们，给他们工具，让他们去赋能。但如果从中形成了某种标准，或者大多数人都在用某种做法，那我们就把它固定下来。

Every community is choosing a slightly different way to approach the problem and forget how we've always wanted to leave that in their hands, like give them tools and enable them, but if a standard comes out of that or most are using a certain practice, we'll lock that in.

但我们真的不想做第一个制定标准或提出某种做法的人。就像米歇尔·桥本分享过的他们用的担保机制那样。

But we don't really ever wanna be the first to create a standard or an approach like I think, you know, Michel Hashimoto shared like the vouch system that they use.

当时有人问我，为什么你不把这个推广给所有人呢？

And I was getting questions like, well, why aren't you roll this out to everybody?

但也有不少社区不想用这个系统，因为他们有自己的想法，觉得应该怎么运作。

But there's just as many communities that don't want to use that system because they have their own ideas of how it should work.

所以目前我们专注于为维护者搭建一套基础的控制工具。

And so for now, we're focusing on the the building blocks of controls for maintainers.

等大家慢慢摸索，维护者们也不断反馈，如果真有某种模式逐渐成型，我们再把它固定下来。

And then as we all are kinda learning together and as maintainers send feedback in, we'll, you know, we'll cement an entire system if one emerges.

是的。

Yeah.

我觉得你正处在第一线，亲眼看着这个全新的智能代理经济兴起，你不是在推特上说过吗？现在每月收到的合并请求比去年一整年还多。

And I feel like you have a front row seat to this, like, new agent economy where, you know, I think you said public on Twitter that you've had more pull requests submitted Yep.

每个月的提交量都比去年全年还高。

And, like, per month than you did all last year.

这些数据怎么突然就暴涨了？

Like, how how are those stats exploding?

是的。

Yeah.

我的意思是，我们在 GitHub 上看到的活跃度高多了。

I mean I mean, we're seeing way more activity on GitHub.

我们一直都在谈我们的用户，这么多年了，一直都在说这种增长。

You know, we've always been talking about our users, you know, for for many, many years and that growth.

但今年，明显看到开发者开始大量使用智能代理了。

But this year, we're seeing obviously the growth of developers having agents Yeah.

他们正在用这些代理来开发。

Building with them.

去年十月，在 GitHub Universe 上，我们提到过，全年 GitHub 上一共有十亿次提交。

And so last year, in October at GitHub universe, we shared, you know, there's a billion commits on GitHub for the full year.

照今年这个增速，全年可能会达到一百四十亿次——当然，实际不可能是线性增长。

We're on track to be 14,000,000,000 if the growth is linear this year, which it would not be.

你懂的。

You know?

三月份，光是智能代理创建的拉取请求就达到了一千七百万个。

In March, you know, there were 17,000,000 pull requests that were created by agents.

是的

Yeah.

这还只是由代理生成的拉取请求。

That's just the agent pull requests.

明白了。

Okay.

对。

Yeah.

因此，现在产生的代码量要多得多。

And so there's so much more code being created.

有时候大家都会觉得，哎，这些都只是垃圾代码。

And I think at times everyone goes like, oh, this is all just like slop.

全是些往上推的代码，没人会在意。

This is all just code that's being pushed up, but and no one cares.

但其实并不是这样。

It's not like really true.

是的。

Mhmm.

我们正逐渐走出早期采用的阶段。

We're all just actually getting to the point where we're no longer in the super early adoption.

我们肯定还没到顶峰，但确实在往上爬，你知道的，看看当不只是凯尔一个人在开发，而是凯尔加上一两个、直到无数个代理，都在使用我的技能、我的资源、我的上下文时，我们能做出什么。

We're definitely not at the peak, but we're climbing that hill, you know, to see what can we build when it's not just Kyle building, but it's Kyle and one, two, to end, you know, agents, that are using my skills, using, my resources, using my context, and so on and so forth.

所以我们正在大力投入，为下一波增长做准备，因为这看起来不是那种增长后就停滞的情况。

And so, like, we're, you know, investing heavily in preparing for the next wave of growth because it doesn't seem to be kind of like growing and plateauing.

它会持续增长，因为不管你在哪里开发，用什么工具，所有这些代码最终都会出现在 GitHub 上，那是你向世界分享的地方，也是你通过拉取请求协作的地方。

It's just gonna continue to grow because no matter where you're, you know, building or what tools you're using to build, all of that code ends up on GitHub, you know, and or that's where you're sharing it with the world, or that's where you're collaborating in a PR.

所以我们需要支持每个人的代理时刻，而不仅仅是 GitHub Copilot。

And so we need to be able to support everyone's, you know, sort of agent moment, and not just, you know, GitHub Copilot.

对。

Yeah.

对。

Yeah.

那商业模式会怎么变呢？

And, how does the business model change?

我觉得免费增值模式在以人为中心的世界里挺合理，毕竟我们睡觉的时候，这些智能体还在继续工作。

Because I think freemium makes sense, you know, in a human centered world where we go to bed, but the agents are still working while we're asleep now.

所以这会不会转向按使用量计费的方式？

So does that change like to usage based?

你是不是也觉得趋势会朝这个方向走？

Like, that you see that kind of where things are going?

是的。

Yeah.

我的意思是，我们目前还说不准最终会怎样。

I mean, like, I don't think we know, like, yet, ultimately.

现在的情况是，Kyle 可能用 GitHub 免费版，我们一直都有 API 调用限制这类机制。

Like, I think, know, we very much right now, like, Kyle's gonna have a license or Kyle's using githubdot,uh, .com for free, and we've always had, you know, API, rate limits, you know, and, like, things like that.

而这些限制，大概就是人们感受到智能体压力的地方。

And that's usually where folks are seeing this, you know, the agent back pressure, I think.

我觉得目标是，如果你希望做更多事，就像彼得·斯坦伯格说的，让一百五十个代理同时干活，那确实很棒。

I think the goal is that if you want to be able to do way more, if you want to be able to, like Peter Steinberger says, you know, 150 agents are doing everything all at once, you know, That's great.

我们希望让你能实现这一点，但同时，我也希望你能拥有出色的 GitHub 核心体验。

We want to be able to enable that to you, but at the same time, I want you to have a great core GitHub experience, you know.

至少，使用一定量的代理是其中必不可少的一部分。

And at the very least, there's some amount of agent, usage as part of that that, is necessary, you know.

这就像很久以前，我们有免费的公开仓库，但不提供免费的私有仓库。

I need similar to how we, way way way back, right, you'd have free public repos, but you didn't have free private repos.

是的。

Yes.

你明白吧？

You know?

后来我们就想，好吧。

And then we said, okay.

其实，个人有一些不想公开的代码是很合理的，所以我们提供了免费的私有仓库，让你能这么做。

Well, actually, like, it's fair for an individual to have some code that they don't wanna put out into the world, and we'll give you free private repos, you know, to allow you to do that.

GitHub 一直在随着行业和社区一起发展，但我们始终聚焦于一点：确保你这位开发者能获得成功所需的一切，同时也要和企业合作，帮助他们满足大规模运营的需求——这些需求通常和普通开发者不太一样。

So GitHub's always evolving as the industry and community does, but we're always sort of focused on, I need to make sure you, the dev, have what you need to be successful, you know, and then work with enterprises, you know, to make sure they have what they need to do at scale, which is usually a little bit different than what, you know, an individual dev is doing.

是的。

Yeah.

对。

Yeah.

而商业模式和定价策略，其实都指向了更广泛的微软生态。

And I guess the business model pricing kind of that all leads back into, like, the wider Microsoft orbit.

而且你现在身兼两职，对整个市场部门也有部分职责。

And because you have a dual role now, right, partial responsibility for the wider kind of marketing org.

所以你能跟我讲讲，这两方面的工作有什么变化？你是怎么分配优先级的？

So do you wanna talk me through how that's changed and how you prioritize between those two?

是的。

Yeah.

说实话，我在 GitHub 已经待了十三年了。

I mean, so, I've been at GitHub for a very long time, thirteen years.

我自己也是开发者，这么多年来一直带领工程团队。

And, you know, as a developer myself and leading engineering teams for a lot of that time.

我觉得 GitHub 一直最特别的地方，就是我们真的非常关注开发者。

And I think what's always been unique about GitHub is we really, really focus on the dev.

我们做的工具是给开发者用的，企业愿意买，这当然很棒，也确实很好。

Like, we're building tools for the developers and the fact that people, like enterprises are buying them is awesome, and that's definitely great.

但我们不是为采购方做产品，是为开发者做产品。

But we're not building for the buyers, we're building for the developers.

百分之百是这样。

And a 100%.

所以，作为 GitHub 的首席运营官，这一直是我关注的重点，我到现在也还在做。

And so, in this, and that's like been my focus as the COO of GitHub, which I continue to do.

现在我又兼任微软开发者业务的首席营销官，我的目标是纵观微软所有开发者工具和技术，确保我们提供的是真正贴合开发者体验的整体解决方案。

And then now as the chief marketing officer of developer for Microsoft, my goal is to look across all of Microsoft's tooling, their developer tools, their technology that they're bringing to developers, and making sure that we're bringing, like, holistic solutions that you can use that are, like, authentic, you know, to developer experiences.

像这次活动，我们今年的办法很不一样——首先地点选在旧金山，氛围跟传统会议厅完全不一样，重点就是：我能去听一场分享吗？我能亲手用这个工具吗？我不想被推销，我得能真正上手用，展台也得是这个感觉，等等。

And at events like this where, you know, we've taken a very different approach to build this year, We're in San Francisco first off, like, the vibe is a bit different than, you know, the conference hall, setup, you know, really focused on can I go to a session, can I use the thing, I don't wanna be pitched on a thing, I have to be able to use it, expo hall, kinda so on and so forth?

这正是把 GitHub 一直以来对开发者的专业、热情和专注，扩展到整个微软的更大范围。

It's really bringing that expertise and, you know, love and focus on the developer that GitHub's always had to have, even broader impact, throughout, all of Microsoft.

是的。

Yeah.

我刚才听你说，这是第一次有外部贡献者和外部演讲者参与 Build 吗？

And, did I hear you say that, this is the first build that you've had external contributors and speaker?

这次 Build 我们是有意安排了不少来自社区的演讲者。

Build that, I think, by intention, we focused on having, you know, speakers, you know, from the community Yeah.

尤其是在主要的演讲环节。

Like, in these primary sessions.

你懂的。

You know?

比如在主题演讲里，我们就请了彼得等一批人。

That includes, you know, in the keynote, we had a bunch of folks like Peter.

还有其他很多人的专场，也都安排上了。

You know, there's sessions, from, you know, and others as well.

我觉得这很重要。

I think that it's important.

软件开发其实是个团队运动。

Like, software development is a team sport.

总觉得一家公司、一个团队，哪怕加上GitHub和微软，就能解决所有问题，这想法太天真了。

You know, it seems silly to think that there's any one company, one group inclusive of GitHub and Microsoft and everyone that can just answer every single question.

软件可不是这么开发出来的。

That's not how software gets made.

我们都在用开源项目，实际上是在前人的成果上继续搭建。

You know, we're all at least using open source and we're building on the backs of these giant open source projects.

咱们该多邀请一些人来分享他们的故事，因为我坚信，这正是开发者们真正想要的。

Let's invite people in that can help tell their part of the story together because I deeply believe that that's what developers want.

我知道，这正是我想要的。

I know that's what I want.

你明白吧？

You know?

我知道我那些做开发的朋友也想要这样。

I know that's what my friends that are developers want.

当我们看这些活动、听大家的反馈时，他们特别兴奋能见到微软和GitHub的人，还能在活动中遇到这些外部视角。

And when we look at the events and we hear the feedback, they're excited to see people from Microsoft, from GitHub, and then, oh, I get to see this outside perspective at this event.

这真的很有意义。

It's really it's really meaningful.

对。

Yeah.

这说得通。

That makes sense.

而且这个市场竞争非常激烈。

And it's a very competitive market.

对吧？

Right?

当然。

Sure.

是的。

Yeah.

可能是最激烈的市场了。

Like, know, the the most competitive market probably.

也许是最后一个这么激烈的市场了，我不太确定。

Maybe the last competitive market, I'm not sure.

但问题是，变化这么快，你怎么在这一切中脱颖而出呢？

But but, like, how how do you differentiate in all of that given the the pace of change is so quick?

对。

Yeah.

我觉得我们一直坚持自己的根本，就是非常重视开发者的自主选择。

I mean, I think we continue to focus, like, on our roots, which shows, you know, we care a lot about, like, developer choice.

这一点从来都没变过。

It's always been true.

我们始终致力于为开发者打造工具，帮助他们实现想法。

We care about building for builders and enabling builders.

所以我觉得现在是个特别有意思的时刻，我们从一个拥有海量API、高度开放的时代，不知不觉进入了某种封闭生态，用户会自然产生一种归属感。

And so I think we're in a moment that's, like, really interesting because we've, went from an era of, you know, having a ton of APIs, all this access to a little bit of, like, an unintentional walled garden setup, you know, where, like, you get a kind of affinity.

有时候我会说，这就像个捕鼠器。

Sometimes I'll say it's like a little bit of a mousetrap.

是的。

Yeah.

你懂的。

You know?

然后你突然发现，哦，那边有个更有趣的东西，可你又得去学新工具、新平台，还得新开个账号。

And then you realize, oh, this thing's really interesting over here, and then I have to oh, I have to go learn a new thing or a new tool or a new account.

对我们来说，我们始终希望支持在GitHub上开发的开发者，让他们能轻松使用其他工具，我们会和所有伙伴合作，让这个过程尽可能简单。

And I think for us, we always want to enable developers that are building with GitHub to go use these other tools, and we'll partner with everyone to make that as simple as is possible.

虽然我也知道，还有其他公司在做类似的事，但我们的独特优势在于，能贯穿软件开发的全流程——不只是代码生成，也不只是协作和代码审查，而是覆盖每一个环节。

And while I think that, you know, there's, there's other folks that are doing, you know, similar things, I think the the ability to do that across the entirety of building software and not just the cogen side or not just, you know, the collaboration review side, but across everything is a real, you know, superpower of ours.

所以，我相信你们会看到我们在自身技术上持续投入。

And so, you know, I think you'll see us invest in our own tech.

我们之前提到过新的微软AI模型，我们会持续为开发者提供这些技术。

Like, we talked about the new, Microsoft AI models, you know, that we'll continue to bring to developers.

我们也会继续与Anthropic、OpenAI、Google，以及任何推出模型或编程代理的厂商合作。

We're also continuing to partner with Anthropic and OpenAI and Google and kinda anyone who's bringing a model to market or a coding agent to market.

无论你是想把你的技术接入我们，还是让我们通过GitHub和GitHub Copilot来集成，我们都会全力配合。

We'll partner with you and we'll both let you bring that to us or we'll omit it through GitHub and GitHub Copilot.

没错。

Yeah.

这种选择权至关重要，我们绝不会在这点上退让。

That choice is core, and that's something that I don't think we'll, ever, you know, like, back down on.

因为如果我们不这么做，开发者还是会自己选，只是会陷入另一个陷阱。

Because if we do, developers will still choose.

我们不希望软件世界变成这样。

They'll just be stuck in another kind of mousetrap.

我们不希望软件生态被局限在这样的框架里。

And we don't, you know, we don't want the world of software to be like that.

是的。

Yeah.

对。

Yeah.

没错。

Yeah.

另外，关于内部做决策的时候，最近有个新闻说云代码的许可证被取消了，你是怎么看这个的？

And, how do you like, when you're making the decisions internally, like there was a news cycle recently about how cloud code cancel licenses are being canceled.

你们在决定用自己开发的新模型，还是用 GitHub Copilot 桌面版，同时又让开发者去尝试其他工具的时候，是怎么权衡的？

Like how do you make the trade off between dogfooding your own products, like using the new models you made or using the GitHub Copilot app desktop app versus you know using like letting developers kind of experiment with other tools?

是的，我们其实都在用各种不同的工具，不然很容易脱离实际，只盯着自己做的东西。

Yeah, I mean we all are using a variety of tools because otherwise you lose track you know, or you kind of you're too interested in your own own work.

对。

Yeah.

就我来说，很多年了，我每天都在用 MacBook。

So like for me, you know, I've been a daily driver of a MacBook for for many many years.

我周末打游戏的时候用Windows电脑。

I use Windows PCs on the weekends when I play video games.

我接手这个工作后，手头有Mac、PC，还有一台运行Omar Linux的机器，这样我每个周六都能写代码，确保覆盖所有平台。

And I got this role and I have my Mac, I have my PC, and I have my Omar E Linux box so I can make sure that every weekend I code most Saturdays.

早上我会陪孩子参加体育活动。

I do my kids sports activities in the morning.

下午我就开始写代码，来回切换不同的设备，就是想亲身体验一下，到底是什么感觉。

And then in the afternoon I'm coding and I'm swapping between the boxes because I wanna understand like, okay, what's that experience?

GitHub Copilot这个应用，我只在Windows上用，因为我想确保使用Windows的开发者也能拥有出色的体验。

You know, the GitHub Copilot app, I only use on Windows because I wanna make sure that developers who are on Windows also deserve great apps.

用户可不只是那些用Mac的人。

You know, it's not just the audience that's on a Mac, for example.

我们整个团队都是这样，尤其是当我们考虑编码助手这类工具的时候。

And that's true across our teams, especially like when we're looking at, okay, what about the coding agents?

还有那些测试框架呢？

What about the harnesses?

那桌面应用呢？

What about the desktop apps?

内存管理又怎么说？

What about memory management?

还有方方面面，都得考虑吧？

What about, you know, everything?

我们团队有个特别好的文化，就是大胆尝试。

We have this really great culture of just, you know, experimentation.

每个人都在开发和使用这些工具。

Everyone is building and using these tools.

当然，我们主要精力还是放在自家工具上。

Well, obviously, we're putting most of our energy into our own tools.

这其实是个盲区，GitHub 过去就吃过这亏——当你把一件事做得不错，就会死死盯着，觉得这才是创业该有的专注，对吧？

It's such a blind spot that I think, it's happened to GitHub in the past where, like, when you're doing something and you're doing it well, you really laser focus and that's what every piece of startup energy says right?

就是低头猛干，拼命往前冲，但我觉得这样太狭隘了。

It's like look down and just keep moving and keep moving fast and I think that's myopic.

你知道吧，我虽然不可能每天用每一个工具，但只要一出新东西，我就想搞清楚它到底好在哪里。

You know, I think that while I can't spend every day using every tool, when something comes out, I wanna know why this is really great.

为什么大家用起来会觉得特别顺手？

Why are people having a great experience with this?

不只是为了理解，更是为了判断——就我们追求开发者自由这个目标而言，其实我不需要这个。

Not, not only so I can understand, but so I can figure out, okay, well, for our goals, for our goal of developer choice, I don't need this.

我想把精力集中在别的地方，但我也得明白，为什么开发者会选这些工具。

Like, I wanna focus over here, but I wanna know why a dev would pick these tools.

同样的道理也适用于我们的团队。

And the same thing goes, you know, the same thing goes for, our teams.

对。

Yeah.

那你怎么筛选呢？

And how do you how do you filter?

毕竟，很多想法都只是昙花一现。

Because, obviously, a lot of these ideas are relatively short lived.

对。

Yeah.

没错。

Yeah.

企业级产品的开发周期要长得多。

Enterprise the product development cycles are longer lived.

是的。

Yes.

你们是怎么决定的？

How do you decide?

对。

Yeah.

我觉得现在我们正处在这样一个阶段，重点是快速把握支持大量智能会话的能力。

I think that, like, right now we're in a moment where we're really looking at the, like, the short term in capturing the ability to have a multitude of, you know, agent sessions.

这个想法很明确，就是因为这个原因。

This idea of just, you know because that that that seems quite clear.

大家都在这么做。

You know, everyone's doing it.

我们该怎么把它固定下来？

How can we cement it?

但从长远来看，方向已经很清楚了。

But it seems clear on the longer term path.

模型会变得越来越强大。

Models are gonna continue to get better.

令牌的价格，也就是令牌经济，会成为决定大家选用哪个模型的关键因素。

The prices of tokens, like token economics is gonna be a bigger, bigger factor in, like, what models everyone's using.

我坚信，我们离在本地设备上运行超越小型语言模型的系统已经不远了，完全可以用来完成一些日常工作。

And I do strongly believe that we're not very far off from having serious, the serious ability to use, you know, something above a small language model on a local device, you know, to be able to do some of our work.

如果我假设在令牌使用上有这么多选择，那么从最初到现在，我一直觉得真正重要的是个性化、上下文、微调，或者说是记忆功能。

And so if I assume that I have all this optionality when it comes to tokens, you know, effectively, The thing that I think seems to be true from the beginning to open claw to now is this idea of, like, personalization or mine or context or fine tuning with context or, like, you know, memory.

这些想法自ChatGPT和GitHub Copilot问世以来，就一直成立。

All of these ideas seem to be a a truth that's been there since, you know, ChatGPT came out or GitHub Copilot came out.

是的

Yeah.

确实有一些实验，但整个行业还缺乏长远的规划和愿景。

And there's experiments, but not a long term, like, vision, I think, for this, like, across the industry.

所以我觉得，这正好说明了为什么你需要熟练地使用大量智能代理——如果你真在用代理，不可能只盯着一个在那儿干活。

So I think it's a good example of, like, where I need to get you to use agents incredibly well, a lot of them, because, like, if you're if you're into using agents, you're not just gonna be staring at a single agent working.

光靠单个代理，是没法带来长久的好体验的。

But that's not gonna give you a long term great experience.

真正让人觉得体验出色的，是你感觉代理在帮你完成思路，尤其是你根本不用亲手把思路写成代码告诉它。

Using an agent that you feel like is completing a thought for you, will give you that great experience, especially if you did not have to personally codify that thought to your agent.

别忘了，我经常会主动插入内容。

Always remember that I insert things.

这工作量可不小。

That's a lot of work.

对

Yeah.

百分之百。

A 100%.

它应该能本能地理解，或者再次通过训练后微调，甚至前沿调优，让模型深度了解我以及我使用工作的方式。

Like, it it should be able to intuit that or potentially again, like, you know, post trainer fine tune or frontier tune like a model on that deeply understands me and how I'm using the work.

是的。

Yeah.

我们目前就是这么看待这个问题的。

That is that is kinda how we're looking at it.

有时候是短期的，有时候我们得多次尝试，反复摸索，才能找到长期的解决方案。

It's like sometimes it's short term and sometimes we gotta take a bunch of bites of the apple or a bunch of attempts at the long term

对。

Yeah.

才能达成真正切实的成果，是的。

To get to something really tangible Yeah.

帮助我们向前推进。

To help us move forward.

不错。

Cool.

昨天我听人说了上百次‘爬山’这个词。

And I I heard the term hill climbing like a 100 times yesterday.

我特别支持这个方法，因为咱们用DSPY、自动研究这些工具做了不少实验。

And and I'm a big proponent of that because, you know, experiment experiments with DSPY, auto research, a few others.

你能说说，这怎么就成了重点吗？

Can you talk a little bit about how that's become a big focus?

对。

Yeah.

我觉得萨提亚、穆斯塔法还有负责Copilot团队的雅各布，都经常提到这个。

I mean, you know, I think, you know, Satya and Mustafa will talk about it, a fair bit and Jacob too, leading the Copilot group.

我们学到最关键的一点是，得把工具的实际使用作为核心，来优化底层模型——包括我们自己的模型——通过各种评估指标，比如用户给的点赞或点踩数据，还有他们到底接不接受、接受多少这些反馈。

The biggest thing that we've kind of learned is we need to use the, you know, the use of the tools as a core way to improve the underlying use of the models, our own models, etcetera, in just the evals that are necessary to ensure that we're actually improving from things like, you know, using the thumbs up, thumbs down data that comes in to using, like, whether you're accepting it and how much you're accepting.

所有这些数据量巨大，正是它们共同造就了那种不是只为你个人、而是为所有人量身打造的神奇体验。

You know, all of that data is enormous to create a that magical type of experience that's not you just for you, but for everyone.

所以每周我们都会讨论爬山结果，看数据、看改进情况，既要关注硬性指标，也要看软性反馈，因为有时候硬指标和评估标准显示进步了，但用户满意度却暴跌。

And so every week we're talking about the hill climbing results, know, we're looking at the data, we're looking at the improvement, we're looking at both the hard measures and the soft measures because sometimes the hard measures and evals and rubrics will show that we've made an improvement but like user sentiment will crash.

是的。

Yeah.

即使延迟和性能都没变。

You know, even with the same latency and performance.

对。

Yeah.

这其实就是过拟合了。

It's overfitting, basically.

完全正确。

A 100%.

所以我们得能非常快速地完成这个闭环。

And so, like, being able to being able to really do that loop incredibly quickly.

没错。

Mhmm.

然后我觉得主要目标是让每个人都拥有这样一台‘爬山优化’工具，不用像我们以前那样费劲地手动操作，尤其是对于企业用户，特别是那些正在使用 M365 的人。

And then I think the main goal is giving everyone one of these hill climbing machines and not have you have to do it the kinda hard way that we've all been doing it, but particularly if you're, you know, in an enterprise and you are using m three six five.

我们对这些数据非常了解，或者说，我们可以非常深入地了解这些数据，因为我们有各种资产、文档和聊天记录作为支撑。

We know so much about that data or we could know so much about that data because of all the assets, the documents, the chats.

所以，能够开启类似前沿调优的功能，并以 MAI Thinking One 作为基础模型来使用。

And so being able to, you know, turn on something like frontier tuning and using on the you know, using a MAI thinking one as the the base model.

它能带来实实在在的效果，而且不需要做那些额外的繁琐工作。说实话，我第一次听说这个的时候，心里想这简直就像个魔术戏法。

It shows real results without having to do all that extra work and it's been interesting because when I first heard about this, I'll be honest, like, I was like, this is like a magic parlor trick.

对。

Yeah.

你知道的，那样是行不通的。

You know, that is not gonna

根本没法运作。

be It can't really work.

没错。

Yeah.

你知道吧？

You know?

我觉得现实是，有时候优势就藏在那些看似不起眼的地方。

And I think the reality is that sometimes, where the alpha is.

就是那种感觉——这方法也太简单了，不可能真管用。

It's like where it feels like this is too simple to work.

对。

Mhmm.

你知道吧？

You know?

我们都拥有这么多数据，可到底该怎么用呢？

You know, we all have all this data and what are we gonna do with it?

以前我们得花老大劲儿才能让数据发挥作用，但现在有太多工具出现了，让我们可以直接用数据优化流程，持续改进，一步步往上爬。

We have to do all this effort to make it work, but I think so much has come down the pipe to allow us to just use the data and improve, look at the workflow and improve, and just keep doing the hill climbing.

所以我们老说这个，因为它不是什么颠覆性的突破，而就是持续的渐进式提升。

That's why I think we say it so much is that it's not these moonshots or like a hill climb.

其实就是不断迭代、提升，通过新的评估来优化，用新数据来训练，然后一直这样推进，直到我们能把自己那七款模型成功上线。对。

It is just climb, climb, improve, new eval improve, new data improve, and just keep going to get to the point where, you know, we're able to launch these models, seven models, for ourselves Yeah.

接着，让客户也能用上同样的，或者类似的工具来完成这些工作。

And then, you know, allow customers to use the same, or similar, you know, tooling to do it.

这就是让两百美元的订阅费不再涨到两千美元的答案吗？

Is that the answer to stopping the $200 subscription becoming a $2,000 subscription?

我是这么想的，从两百美元涨到两千美元，不仅仅是因为我们要把这些模型做得更懂用户，比如做前沿调优。我觉得更关键的是，我们怎么帮开发者自动选择最合适的模型，甚至在这一环节直接集成好模型。

I mean like I think that the you know $200 subscription to $2,000 is really gonna be not only, you know, making these models or, you know, frontier tuning these models so they way they know you better, but I also think it's really, really gonna be about how can we, particularly for developers, you know, help you automatically choose the models and potentially either have like a model in that step.

比如模型路由器和 A100。

Like the model router and A 100%.

没错。

Exactly.

就像那个，你知道的，根据任务意图自动选择模型的路由器，还有 GitHub。

Like the, you know, auto model router with task intent and GitHub.

微软 Foundry 也有一个模型路由器，能在 API 层面实现这个功能。

Microsoft Foundry has a model router as well that can do this, use sort of at an API level.

对。

Yep.

我们越能帮你明确自己的需求边界，比如你到底需要什么水平的性能，这个问题就越难解决。我愿意全力以赴，而不是只想坐在这儿，让系统帮我们自动选模型。毕竟很多时候，token成本高，是因为我们总在不断更换当天、当周甚至当小时的模型，而这些模型本身都非常昂贵。

The more and more that we can help you, you know, tell us a bit of, like, where your bars are, Like, is an incredibly hard problem and I'm willing to go all the way to the top or I don't I just kinda wanna sit, you know, here, and let us help choose the models because there's a lot of times where a lot of the reasons why tokens are expensive is because we're all going and choosing our model of the day or week or hour, you know, and those models are incredibly expensive.

我的思路一会儿陷入一个复杂问题，一会儿又跳到一个简单问题上。

But my train of thought is slipping in and out of a hard problem to an like a simple problem.

是的。

Yeah.

我个人觉得，我会让智能体帮我完成大量工作，但总还剩最后一步，是个小事儿，比如：‘哎，其实我不太想把这里的命名全改成那样。’

I I personally I feel like I'll you know I'll get an agent to do like an enormous amount of work and then there's always that last step that is like a smallish thing you know like oh I don't actually like change all the naming of this to this.

对。

Yeah.

这根本不需要。

And that's like Don't need

一个查找替换就够了。

a a find and replace.

你懂我的意思吧？

You know what I mean?

但我会真的为了省点令牌，特意从48或55切换到Haiku之类的模型吗？

But but am I gonna actually go and, like, oh, I wanna save tokens right now, so I'm gonna go off a four eight or five five and down to Haiku or something, you know?

大概不会。

Probably not.

是的。

Yeah.

但工具可以做到。

But the tools could.

对。

Yeah.

我觉得这对我们帮助会很大，尤其是在企业场景，甚至对个人开发者、做自动化、用Copilot SDK的人也都很有帮助。

And I think that will really help us, particularly in the enterprise, but even for, you know, individual developers and, you know, folks that are building, like, automations and using their Copilot, you know, SDK Yeah.

来驱动这一切。

To power that.

这也会对他们有帮助。

It'll it'll help them too.

我做了一件有点特别的事。

I did something a little bit weird.

希望你不觉得奇怪，但我做了一个你的AI版本，用来练习这次访谈。

I hope you don't find it creepy, but I I made a AI version of you to practice this interview.

真的吗？

No way.

是的。

Yeah.

到目前为止，它表现得相当准确，也希望你觉得这些问题还不错。

And it's actually been pretty spot on so far, and, hopefully, you think the questions have been good.

这些问题都很好。

And that's they've been great.

但现在我想看看AI版的凯尔是怎么说的。

But now I wanna see what AI Kyle said.

而且，对，就只是在终端里运行。

And, yeah, it's just in the terminal.

我没搞那么复杂，没做成视频那种形式，

I didn't have like I didn't go with the full whack and make a video thing,

但我现在

but But I'm

但我发现它特别有用。

but I found it immensely useful.

我就是想问问，你有没有看到别人在内部或外部做些什么奇怪的事情？

I like I just wanted to ask, like, what other, like, weird things are you seeing people do internally or externally?

天哪。

Oh, man.

真巧的是，你这么一说，我也在做类似的事——我同时用App和一个专门隔离的设备，完全不接触工作内容，就是为了分清界限。我花很多时间让它读我写的所有东西，毕竟这场访谈最后也会喂给它。

So, like, it's so funny that you say that because I, I do a very similar thing where, like, I I, I have both, like, via the app and then I have a claw that can't talk to work stuff, you know, just so I have, like, separation of estate, Where I spend a lot of time having it read everything I write and say, like this interview will get fed into it ultimately.

每天我都会收到一份通讯报告，不是说Kyle说了什么，而是‘Kyle，你老是这么说’。

And every day I get a like a comms report that's not like what Kyle said, but like, Kyle, you keep saying this.

是的

Yeah.

好的

Okay.

这不太清楚。

This isn't super clear.

对

Mhmm.

举个例子，你说话的方式，因为我发现自己写东西和说话时有一种特别的风格，特别喜欢用比喻。

Like, you know, based on how you speak, because I I find that I write and speak in, like, a very particular way that I, like, I wanna use a lot of metaphors.

所以它会给我一些清晰的比喻例子。

And so it'll just give me examples of metaphors that are clear.

是的

Yeah.

我发现，作为一个人，自我提升的这个循环确实挺有用的。

I find that the, like, self improvement loop as a human Mhmm.

这些智能体的能力会非常强大。

From these agents to be incredibly powerful.

以前我们跟GitHub的Hubot聊过，那时候说的聊天运维，我们总觉得，人更容易接受机器人给的尖锐反馈，而不是来自其他人的。

We used to talk about it way back with, Hubot at GitHub, like, chat ops, and we used to say, like, humans are way more willing to take critical feedback from robots than other humans.

确实是这样。

It's Yeah.

没那么有威胁感。

Less threatening.

百分之百对。

A 100%.

当我那个被我亲切地叫做Baxter的机械臂，告诉我某件事我做得有多糟的时候，嗯，

And so when my open claw that I affectionately named Baxter, you know, tells me how terrible I did in something, like Yeah.

我反而更愿意问：‘为什么这么说？’

I feel way better going, tell me why.

这样我写邮件、写脚本，或者审阅你给我的细节时，就能更好地吸收这些反馈。

And then ensure that when I'm writing emails, when I'm writing a script, or I'm reviewing details that you're giving me that feedback.