Stripe 如何构建“小助手”——每周通过 Slack 反应推送 1300 个拉取请求的 AI 编码代理 | 史蒂夫·卡利斯基（Stripe 工程师）

在Stripe，我们每周大约有1300个拉取请求，除了审查之外，没有任何人工协助。

At Stripe, we're landing about 1,300 PRs that have no human assistance besides review per week.

我们工作的起点很多，可能是在Google文档里规划新功能，或者收到一个Jira工单，也可能是在Slack里讨论事情。

A lot of where our work begins is it could be in a Google Doc as we're planning a new feature, or maybe a Jira ticket comes in, or we're talking about something in Slack.

我可以点击一个表情符号，然后菜单会尝试用Stripe可用的所有工具一次性解决那个提示。

I can click an emoji, and then the menu will sort of attempt to one shot resolving that prompt using all the tools that are available at Stripe.

在大型组织中，从好点子到落地，中间会有很多摩擦。

When you're in larger organizations, there's so much friction that can come between a good idea and getting it into the world.

我不仅能有一个这样的，还能同时运行很多个，在隔离的环境中并行，同时进行独立的修改。

Not only can I have one of these, but I could have many, many of these running in parallel in isolated environments, making isolated changes all at the same time?

你怎么完成所有这些代码审查的？

How are you getting all this code review done?

无论是Steve写的文本，还是Steve的机器人写的，你仍然需要那个CI环境来提供信心，确保被修改的代码是安全的，并且在部署时采用蓝绿部署，以便可以回滚。

Whether the text has been written by Steve or the text has been written by Steve's robot, you still want that CI environment that's providing confidence that the code that's being changed is safe and that as it rolls out, we're having blue green deployments so you can roll back to.

所有这些都至关重要，与代码的作者是谁无关。

All that is super critical independent of the nature of the authoring of it.

不管这些笔记本电脑配置多高，你开三四个工作树之后，风扇就开始像飞机起飞一样响。

No matter how juiced these laptops are, you get three or four work trees in and, like, it starts to sound like an airplane taking off.

这真不行。

It's no good.

所以我觉得在这种多线程的智能体工程工作中，云端环境和虚拟环境对提升开发速度特别重要。

And so I do think on this multithreading agentic engineering work, cloud environments and virtual environments are so important to unlock velocity.

欢迎回到《How I AI》节目。

Welcome back to How I AI.

我是克莱尔·沃，产品负责人，也是AI爱好者，我的使命是帮你更好地使用这些新工具。

I'm Claire Vo, product leader and AI obsessive here on a mission to help you build better with these new tools.

今天，我们有Stripe的软件工程师史蒂夫·卡利斯基，他会向我们展示Stripe团队如何利用各种智能体来开展工程工作。

Today, we have Steve Kaliski, a software engineer at Stripe, and he's gonna show us how the Stripe team deploys a bunch of opinions to do their engineering work.

我们还会看到一个智能体花五美元多一点点，用Claude Code来策划一个生日派对。

We'll also watch an agent spend a little bit over $5 to plan a birthday party all in Claude code.

我们开始吧。

Let's get to it.

本期节目由 Optimizely 赞助播出。

This episode is brought to you by Optimizely.

大多数营销团队不缺创意，缺的是时间。

Most marketing teams aren't short on ideas, but what they are short on is time.

而 Optimizely Opal 正好能帮你把时间省出来。

And that's exactly what Optimizely Opal gives you back.

它用 AI 智能体来处理真实的营销工作流，比如创建内容、检查合规性、生成实验变体、个性化用户体验、分析页面做 GEO，甚至处理审批和报告这类任务。

With AI agents that handle real marketing workflows, you know, like creating content and checking compliance, generating experiment variations, personalizing user experiences, analyzing pages for GEO, even tasks like approvals and reporting.

它是面向营销和数字团队的 AI 智能体编排平台，能无缝接入你已经在用的工具，处理那些枯燥的琐事，并确保一切符合品牌调性。

It's your AI agent orchestration platform for marketing and digital teams, plugging seamlessly into the tools you already use, handling the boring busy work, and keeping everything on brand.

这样，营销人员就能腾出更多时间去做真正该做的事。

That leaves marketers with more time to do your actual job.

想看看 Opal 能为你的团队自动化哪些工作，可以免费报名参加 Optimizely 的企业级智能体 AI 工作坊。

See what Opel can automate for your team by signing up for a free enterprise agentic AI workshop with Optimizely.

更多信息请访问 optimizely.com/howiai。

Find out more at optimizely.com/howiai.

现场参加的话，你会免费得到一副Ray-Ban Meta AI眼镜。

Attend live and you'll get a free pair of Ray Ban Meta AI glasses.

Steve，能请你来《How I AI》我特别激动，因为我在时间线上看到了Stripe的小黄人。

Steve, I'm so excited to have you on How I AI because I saw the Stripe minions on the timeline.

第一，品牌做得太棒了。

And one, exceptional branding.

别告我们啊。

Don't sue us.

第二，我特别喜欢你们Stripe团队和你一起，不只是做了一个智能体，而是在整个公司里都搞了一堆小黄人，能帮忙做开发工作。

And two, I just love the idea that you and your colleagues in the team at Stripe have created not just one agent, but minions all across the company that can help with development work.

我也特别期待你给我们展示一下，这在你日常工作中是怎么帮到你的。

And I'm so excited for you to show us how that helps you in your day to day here.

所以，欢迎来到《How I AI》。

So welcome to How I AI.

谢谢你的邀请。

Thank you for having me.

那你跟我说说，小黄人（minions）在Stripe对你个人以及整个Stripe团队产生了什么影响？

So tell me, what has been the effect that minions have had on you personally at Stripe and at the Stripe team as a whole?

好的。

Sure.

就我个人来说，举个例子，我不记得上次在文本编辑器里开始工作是什么时候了。

So, you know, for me personally, think sort of anecdotally, I don't remember the last time I I started work in the text editor.

我确实经常用到它。

So I do end up there often.

但我发现，我们很多工作的起点，可能是在Google文档里规划新功能，或者收到一个Jira工单，或者在Slack里讨论事情。

But, you know, what I found is that, you a lot of where our work begins is, you know, it it could be in a in a Google Doc as we're planning a new feature, or maybe a Jira ticket comes in, or we're talking about something in Slack.

这些更像是开始工作的自然入口。

And those are sort of, like, the more natural entry points to starting work.

然后，当真正需要动手干活或做最后调整的时候，你才会进入文本编辑器。

And then you end up in a text editor when it's time to, you know, actually do the work or make the final tweak.

这感觉非常自然。

And it just felt very natural.

而且我觉得，尤其是开始工作的那种启动成本感觉低了很多。

And I think in particular, the sort of, like, activation energy of starting work feels a lot lower.

所以，如果你在Slack的讨论串里，可能有一条用户反馈，而且是很简单的那种，比如……

So if, you know, you're in a Slack thread and maybe there's a a piece of user feedback and it's something simple like a...

你知道，我们需要更新文档，或者可能是更重要的事情，我们只是想做个原型。

You know, we have update the docs or or maybe it's something more consequential and we just wanna build a prototype.

我可以点一个表情符号，然后工作就开始了。

I can click an emoji and, like, the work begins.

而且很多时候工作也完成了。

And often the work finishes too.

我们……

We...

在Stripe，我们每周大约有1300个PR是……

Stripe, we're landing about 1,300 PRs that are...

除了审查之外，完全没有人协助的。

Have no human assistance besides review per week.

但至少，开始写代码、看到测试通过，或者测试失败，这些都不需要我亲自参与。

But at the minimum, the activation energy of, like, starting to write code, seeing tests pass, maybe a test fails, occurs without me even, you know, participating.

然后我可以加入进去，调整一下，让它们……

And then I can jump in, and I can tweak, and I can kinda, like, have them...

那种势头，那种生成性的动力，我可以中途加入。

That momentum sort of sort of, like, generative momentum, you know, that I I can hop in halfway through.

我觉得这最神奇的地方在于——我不会说Stripe是大公司，但你们确实有不少员工，业务也很大——我很喜欢这个“激活能量降低”的概念。

What I think is magical about this, and I I won't call Stripe a big company, but you do have a decent amount of employees and very, very large business, is I love that concept of activation energy going lower.

因为在更大的组织里，从好点子到落地，中间会有很多摩擦。

Because when you're in larger organizations, there's so much friction that can come between a good idea and getting it into the world.

这不是恶意。

And it's not malintent.

就是……

It's...

没人会说，哎呀。

Nobody is like, oh, man.

我真的很想把这个过程放慢下来。

I really wanna slow this process down.

要么就是，你知道的，功能上的问题。

It's either, you know, functional.

我没有技术方面的专业知识，没法真正从这里走到那里。

I don't have access to a technical area of expertise to actually get from here to there.

要么是操作上的问题。

It's operational.

我不知道怎么组织人、怎么有效沟通，才能把下一步做完。

I don't know how to organize people and communicate effectively to get the next step done.

或者就是大家每天各忙各的，被隔离在自己的工作里，想不到新的工作方式。

Or it's just kind of like people get siloed in their day to day and don't think of new ways to get work done.

对我来说，AI最让人大开眼界的一点就是，所有这些障碍都几乎消失了，因为协调成本可以降下来。

And one of the things that has been so revelatory about AI for me personally is, like, all that just kinda goes to zero because coordination cost can go down.

执行成本也可以降下来。

Execution cost can go down.

沟通成本也能降下来。

Communication cost can go down.

你就能更贴近工作本身，我觉得这才是我们真正在乎的有趣部分。

You just get closer to the work, which I think is the fun part we all really care about.

那给我演示一下，你实际上是怎么激活一个“小兵”的。

So show me how you actually activate a minion.

然后，我们稍微跳过一点。

And, you know, we skip this a little bit.

什么是“小兵”？

What a minion is?

简单介绍一下“小兵”吧。

The quick spiel of a minion.

在AI出现之前，我作为工程师，想对Stripe做点修改的时候。

When I, as an engineer, sort of in pre AI time, you know, want to make a modification to Stripe.

Stripe是个庞大的代码库，有很多服务。

Well, Stripe is a huge code base with tons of services.

它没法只在我的电脑上运行。

It it can't run on my computer alone.

所以Stripe在开发工具方面一直投入很大，有托管开发环境，我可以快速启动，里面代码和运行的服务都准备好了，我可以通过SSH登录进去做修改。

So Stripe already has a long history of of, you know, investing in great developer tooling, having hosted development environments that I can spin up, that, you know, have all the code already there and services running, and I can SSH in and and make modifications.

而且我们还有大量相关的CI工具。

And we have a ton of great CI tooling around that.

所以……

So that...

这就是背景。

That's the context.

你知道，这些我们都有。

You know, we have all that.

minion的想法是，我可以配置一个这样的环境，用提示词初始化，然后minion会尝试利用Stripe所有可用的工具一次性解决这个提示词。

The idea with the minion is that I can provision one of those environments, seed it with a prompt, and then the minion will sort of, you know, attempt to one shot resolving that prompt using all the tools that are available at Stripe.

包括我们所有的内部文档、内部CI、测试数据等等。

So all of our internal documentation, our internal CI, our, you know, test data, so on and so forth.

它会反复尝试，直到解决那个问题。

And it will loop through that in an attempt to, you know, solve that problem.

那我们就直接开始，看看一个典型的体验是什么样的。

So let's go ahead and and jump in and see what sort of a prototypical experience might look like.

我现在在一个Slack频道里。

So I'm in a Slack channel.

这个频道叫Steve Kaliski robots dash Claire。

It's called Steve Kaliski robots dash Claire.

我其实有一个Steve Kaliski robots频道，里面有76个人，但我确实有每一个……

I actually have a Steve Kaliski robots channel that has 76 humans in it, but I do have every...

一开始只有我和我的机器人，现在有一些观众在围观。

It it started off as just me and my robots, and now there's some sort of, you know, like, audience observing.

但假设，也许我是……

But let's imagine that, you know, maybe I'm...

我……

I...

我在想一个新功能的想法，或者想改进我们现有的文档。

I'm thinking of a new feature idea or I wanna improve documentation that we have.

我们很快就要发布新版本了，我想把文档做得更丰富一些。

So we have a a launch coming up soon, and I I wanna sort of embellish the documentation.

所以我会说，我在 stripe.com/payments/machine 上有个关于文档的好主意。

So I'll say, I have this cool idea for docs at stripe.com/payments/machine.

这是我们新的机器对机器支付项目，我们稍后在通话中会看到。

This is our new machine to machine payment work, which we'll look at later in our call.

我想确保着陆页真的吸引人，并提供一个好的代码示例，让大家能快速上手。

And I want to, you know, make sure the landing page really sticks and gives a good code example of how to get started quickly.

所以可能有人发了这样的消息，或者通过工单进来的，或者不管来源是什么。

So maybe someone posted a message like that or it came in through a ticket or or whatever the origin may be.

我现在要做的就是添加一个反应，就是创建 minion 支付服务器。

All I have to do now is, you know, add a reaction, which is create minion pay servers.

这是 Stripe 内部的一个特定仓库。

This is a particular repository within Stripe.

开发代理那边会先回复一个“稍等，正在处理”的消息。

We get the one sec cooking from the dev box agent.

然后我们在这里收到回复，说你的支付服务器小助手，也就是那个仓库，已经创建了新分支，落地页的代码示例也做好了。

And then we get a reply in here saying, your minion for pay server, it's the repository, for a new branch that's created, landing page code example has been created.

然后它会启动我们的文档服务，这样我最后就能预览了。

And it's gonna kick off our doc service, so I can eventually preview it.

现在，我要点击“跟随操作”。

Now, I'm gonna click follow along.

所以现在它正在做的是，配置我之前提到的那个开发环境。

So right now, what it's doing is it's provisioning that development environment I was talking about earlier.

这部分是……

So this is...

这部分不是新功能。

This part isn't new.

它很棒，但并不是新东西。

It is excellent, but it's not new.

基本上，它会在云端启动一个实例。

And basically, it's gonna spin up a instance in the cloud.

它会应用我和代理在Stripe内进行编码所需的所有配置。

It's gonna apply all the configuration that's required for both me and the agent to do coding within Stripe.

所以这只需要几秒钟。

So this will just take a few seconds.

它会检出那个仓库的新分支，配置本地数据库，应用我的git配置。

It's gonna check out that repository with a new branch, configure the local database, apply my git config.

它会设置一个VS Code服务器，这样我就可以通过网页或本地连接它。

It's gonna set up a VS Code server so I could connect to it just through the web or or locally.

还有一些扩展。

As well as some extensions.

Minions真正棒的地方在于，显然，有代理循环在进行代码修改。

So what's really, like, great about Minions is, you know, obviously, there's the the agent loop that's, you know, making the code modifications.

但它建立在开发者生产力团队所做的大量出色工作之上，这些工作让获取一个完美运行的Stripe开发环境变得容易，这意味着，你知道，我不仅能有这样一个环境，还能同时运行许多个这样的环境，在隔离的环境中并行进行隔离的修改。

But it's built on top of, like, a ton of incredible work that our developer productivity has done around just making it easy to get, like, a a perfectly operating Stripe development environment for coding, which means that, you know, not only can I have one of these, but I could have, you know, many, many of these running in parallel in isolated, you know, environments, making isolated changes all at the same time?

所以，你看，那个一键表情符号，同时用几条消息就能搞定，真的很棒。

So, you know, that little one click emoji, could have done that with a few messages at the same time, which is really great.

所以呢。

So

这里我想提一点，我们之前请了LaunchDarkly的朋友Zach来上节目。

One thing I wanna call it here is we had my friend Zach from LaunchDarkly on.

对。

And Yeah.

他说了一句话，就是，你看。

One thing he said was, look.

对开发者好的，对智能体也好。

What's good for the developer is good for the agent.

所以，没错。

So Yep.

这就是一个良性循环：如果你为人类工程师投资开发者体验，你的智能体也会从中受益。

There's this virtuous loop of if you have or do invest in developer experience for your human engineers, your agents will benefit off of that.

反过来，如果你为你的智能体工程师投资开发者体验或智能体体验，你的开发团队也会从中受益。

And in turn, if you invest in developer experience or agent experience for your agent engineers, your development team benefits from that.

所以我总是跟人说，工程团队，我们一直在问，能不能在路线图上多给DX一点时间？

And so I always tell people, you know, engineering team, we've always asked, like, can we just give a little bit more time on the road map to DX?

拜托拜托。

Like, pretty please.

能不能在这里投点资源？

Can we invest here?

我觉得如果你把它跟AI项目挂钩，那就是搞定这些好事的秘密途径。

And I think if you attach it to an AI initiative, that's, like, the secret way to get some of that that good stuff done.

我是说，想象一下你的……

I mean, imagine your...

你知道，有些代码库很小，但Stripe就很大。

You know, some code bases are small, you know, Stripe is huge.

你知道，想象你第一天上班，没有文档，没有工具，而且还有……

You know, imagine you show up day one, and there's there's no documentation, and there's no tools, and there's...

然后他们就说，祝你好运。

And they just say, good luck.

换谁都会觉得难。

Like, anyone would have trouble.

就算你把智能体扔进去，很可能整个代码库会把上下文窗口撑爆。

And even if you threw the agent at it, it's very likely that the the context window would be blown by the whole code base.

光是扫描一遍去理解所有细节，几乎不可能，或者成本高得离谱。

Just scanning through to understand all the intricacies would be, like, impossible or extremely expensive.

所以，如果在Stripe当工程师，90%的常见工作都有一条非常明确的路径，那就会让……

So, you know, if there's a very blessed path for 90% of the common activities in being a, you know, engineer at Stripe, that makes it...

那也会让智能体成功的概率变得非常高。

That makes the propensity that the agent succeeds really high too.

所以，想象一下我们想改一个API，我们确实经常这么干，一年要改几百上千次。

So, you know, imagine we wanted to make an API change, which we do, know, you hundreds or or thousands of times a year.

我们有非常好的文档，讲怎么添加新字段、新方法或新资源，小助手可以照着读，然后照着执行。

We have really good documentation on, you know, how to add a new field or a new method or a new resource that the minion would read and would execute against.

然后，你知道，它一次成功的概率非常高。

And then, you know, the the propensity of it would one shot is very high.

所以，对开发者来说，好的文档对智能体同样重要，你说得对。

So good docs for developers are equally important for the agent to your point.

所以我们现在已经从启动开发环境过渡到了第一次智能体运行。

So we've now transitioned from, you know, booting up the development environment to now we're in the the first agent run.

所以我们这里有我在Slack里发的那个提示。

So we have that prompt that I posted in Slack here.

现在它要做的是启动一个Goose实例，基本上就是那个用来跑完整个流程的框架。

And now what it's gonna do is boot up an instance of Goose that's basically the harness that's, you know, gonna run through all this.

我们确实和Block团队聊过一期关于Goose的节目，就是那个搭建起来的开源智能体框架。

We did have an episode with the block team about Goose, this this open source agent harness that got set up.

我想提醒一下那些没在看视频、只在听的朋友，我特别喜欢你的系统提示。

And I wanna call out one thing for folks that are not watching and are listening, which is I love your system prompt.

太复杂了。

So sophisticated.

它说，完整实现这个任务，冒号，然后随便你写什么。对。

It says, implement this task completely, colon, and then just whatever you put Yeah.

用Python。

In in Python.

错误。

Mistakes.

你忘了说不要出错。

You forgot no mistakes.

但我觉得，很多人真的以为他们必须把最初的提示词设计得非常复杂。

But, you know, I I think people really think they have to overarchitect their initial prompt.

而且我认为，如果你有一个好的工具框架，即使提示词很随意，也能帮你得到不错的结果。

And I think if you have a great harness, it can go a long way to extracting out a successful outcome from a pretty loose prompt.

完全同意。

Totally.

而且我们……

And we...

这在某种程度上都是一场实验。

A lot of this is an experiment in some way.

你知道，随着新模型的推出，我们也在构建新工具，这其中有某种动态性。

You know, as new models come out and, you know, we build new tools, like, there is this sort of dynamic nature to it.

而且我们……

And we...

我们构建了很多有趣的机器人，它们能帮助编写提示词。

We've built a lot of interesting, you know, bots that help write the prompt.

所以，也许首先它会执行搜索代码库、查看所有拉取请求或谷歌文档等任务，或者别的什么。

So, you know, maybe first it will do the task of searching through the code base or looking at all the pull requests or Google Docs or whatever it may be.

我觉得那是Stripe。

I I think that's Stripe.

大多数可能有MCP服务器的系统都已经有了。

Most things that could have an MCP server have an MCP server.

所以我们能够与很多内部数据进行交互。

So we're able to interact with a lot of the internal data we have.

然后它能生成一个提示词，我可以直接粘贴到这里，或者分配给代理。

And then it can make a prompt that I could then paste in here or I get assigned to to to the agent.

所以，这就是为什么我想要那个公开频道，因为我们不再结对编程了，但我们结对写提示词。

So that's sort of, like, you know, part of why I wanted that public channel we were looking at is, you know, working on the CF that we don't pair program anymore, but we, you know, pair prompt.

这种活动可以和其他工程师、其他数据源或者其他代理一起进行，对吧，来搞清楚我们能不能正确地向代理解释怎么做。

And that that activity could be with other engineers or other data sources or or other agents too, right, to, like, figure out if we can, you know, properly explain to the agent, you know, how to do it correctly.

不管怎样，现在它正在做的是，它拿了我给它的链接，那是公开文档。

In any case, you know, what it's doing now is it's taking the link I gave it, which is to public documentation.

它会搜索代码库，用我们的一些代码搜索工具来定位那个特定更改应该放在哪里。

It's gonna search through the code base and and use some of our, you know, code searching tools to to locate where that change particular should go.

它会执行一整套工具。

It's gonna execute a whole sequence of tools.

随着时间推移，当它弄清楚应该在代码库的哪个位置工作，以及应该做什么修改时，它们最终会提交这些更改，并生成一个拉取请求，供我和我的同事审查。

And over time, as it figures out where in the code base it should work, what the modification should be, they'll ultimately commit those and make those available in a pull request that that me and my fellow colleagues can review.

我对此有几个问题，因为我们看到了一些例子，人们构建自己的云代理，然后有点……

I have a couple couple questions on this because we've seen a few examples of folks building their own cloud agents and and kind of...

我很好奇，为什么选择Goose，而不是自己开发或者用更商业化的解决方案？

And I'm curious, you know, why why Goose, you know, versus doing something on your own or doing sort of a a more commercial solution?

我想知道你们内部有没有讨论过，还是说……

I'm curious if there was an internal discussion or how this...

还是说这其实是自然发生的，因为一个工程师用起来效果不错。

Or did this happen organically because it worked for one engineer.

我很好奇你们是怎么想到在开发工具上叠加Minions这个想法的。

Curious how you kinda seeded the idea of of Minions on top of your development tools.

当然可以。

Sure.

其实我们也让Stripe的工程师广泛使用Claude Code、Cursor这些工具。

So, you know, we also make Claude code and and Cursor and tools like that widely available to engineers at Stripe.

所以我觉得我们的总体想法是，我们想加速开发，这样就能更快地为用户推出新功能。

So, you know, I think our general sentiment is, we want to accelerate development so we can build new features for our users.

而且以后还会有新模型、新工具出现，我们也想构建……

And there are gonna be new models coming out, new tools, and we wanna build...

我们尽可能多地推广这些工具。

Proliferate those as much as we can.

具体到 Minion 这个案例，

In the particular case of Minion, the...

它非常……

It's very...

我不想说非常特定，但它确实非常贴合 Stripe 开发者的体验和开发环境。

I don't wanna say very specific, but it's very specific to, like, the Stripe developer experience in the Stripe developer environment.

我们很早就开始尝试 Goose 了。

And we have been experimenting with Goose early on.

而且我觉得在这个具体案例中，我们也对它进行了分叉，做了一些修改。

And I think in this particular case, we forked it to make some modifications as well.

我们真正想要的是一个基础框架和循环，用来承载我们自己的所有工具和软件。

And really what we were looking is like sort of a a base harness and loop to apply all of our own tools and software to.

所以我们花了很多时间在提供好用的工具上，并且确保 minions 走的路径能紧密配合 Stripe 最常见的开发工作流。

So we we spent a lot we spent a lot of time on, like, making good tools available and making sure that the sort of routes that the minions go through, you know, work closely with, like, the most common Stripe developer workflows.

所以，这就像是商业方案和定制方案之间的区别。

So, like, it's, you know, sort of like commercial versus, you know, custom things.

比如，有些东西是Stripe的代码库特有的，作为Stripe的开发者，我们构建东西的方式也很特别，所以对我们来说，自己构建和部署这些东西反而更容易。

Like, there are things that are very specific about Stripe's code base and being a developer of Stripe and the way we build things that it it was just sort of easier for us to to build and deploy that.

但商业方案也很棒，我们大量使用它们。

But the commercial solutions are great, we use those extensively.

即使在后面的演示中，我也可以展示，比如，我可以打开VS Code网页版，手动编辑这里的代码。

Even later on this demo, I can sort of show, like, I can, you know, for example, I can pop into VS Code web where I could manually edit some of code that's going on here as well.

我也可以启动Claude，体验典型的Claude功能，包括所有Stripe的MCP工具……

But I can also boot up Claude, I can have sort of the typical Claude experience with all the Stripe MCP tool...

Stripe内部的MCP工具也都可以用。

Internal Stripe MCP tools available as well.

所以，没有哪个单一工具能包打天下，但我觉得在Stripe，端到端的开发体验主要还是靠Minions。

So, you know, there's there's no singular tool to rule them all, but I think the, like, overall end to end development story at Stripe is still time minions.

所以你可以看到，我现在在dead box和Claude里。

So you can see I'm in that dead box and Claude now.

是的。

Yep.

好的。

Cool.

我还有一个问题，然后还有一个观察，我想确保听众们不会错过。

I have one other question and then and then an observation I wanna make sure that that the listeners don't miss.

所以我的第一个问题是，你知道，Stripe 是一个资源非常充足的工程组织。

So my first question is, you know, Stripe is a very well resourced, I would say, engineering organization.

所以我猜你们有一个专门的团队，不仅负责开发工具，还负责管理 minions，把它当作一个内部产品来运营。

So I'm presuming you have a team dedicated to working on not just your dev tools, but as well as as minions and managing that as an internal product itself.

这个团队是不是作为一个独立的团队来建立的，专门专注于内部开发者体验？

Has that team been sort of built as a standalone team that's focused specifically on internal developer experience?

是这样运作的吗？

Is that how it works?

我们……

We...

从我记事起，我们就有一个开发者生产力团队。

We've had a developer productivity team for as long as I can remember.

我想想……

I think about...

我现在在Stripe已经六年半了。

I'm gonna strip six and a half years now.

这个团队专注于我使用的所有工具，并让它们变得更有用。

And, you know, that team is focused on all the tools that I engage with and and making them more useful.

这包括从我们如何使用Git和版本管理，到我们的技术中心和配置，再到我们的开发环境，以及整个流程如何串联起来。

So that's all the way from, you know, how how we interact with with Git and Versus Management to our tech centers and our configurations there to our development environment and and how that whole story pieces together.

而且，我们……

And, you know, we we...

作为Stripe的产品工程师，我非常关心我们的外部用户，希望他们在Stripe能取得成功。

You know, just as, you know, as a product engineer on Stripe, I I I care deeply about our external users and and them being successful at Stripe.

那个团队同样关心Stripe的工程师能否成功，能否快速构建东西。

That team cares equivalently about engineers at Stripe being successful and and being able to build things quickly.

而且我觉得在过去几年里，人工智能更是加速了这一点。

And I think that's been even more accelerated by by AI in the last couple years.

另外，我还想提一个观察，因为我觉得你一开始有点一笔带过了，但这对于真正想用AI来疯狂写代码的人来说非常重要。

And then one other observation I wanna make, because I think you glossed over it a little bit at the beginning, but it's so important for folks that really wanna go ham on coding with AI.

当然。

Sure.

那就是……

Which is...

你看。

Look.

我们所有工程师都有一台重达八百万磅的MacBook Pro。

All of us engineers have a MacBook Pro that weighs 8,000,000 pounds.

没错。

Sure.

那玩意儿可真是有点分量。

That can can do some damage.

我的那台，如果有人想知道的话，外号叫“大块头”。

Mine, for anybody who wants to know, its nickname is big boy.

所以每当我需要孩子们

So whenever I need my kids

好的。

Okay.

去拿我的编程笔记本时，我就会说，能把大块头拿给我吗？

To get my my coding laptop, I say, can you bring me big boy?

因为它……

Because it's...

我管背着两台它叫“旧金山负重行军”。

I call it San Francisco rucking when I carry two of them in my backpack.

天哪。

Oh my god.

但是，你知道，不管这些笔记本配置多高，你同时跑三四个工作分支，它就开始像飞机起飞一样响。

But, you know, no matter how juiced these these laptops are, you get, like, three or four work trees in all running and, like, it starts to sound like an airplane taking off.

这不行。

It's no good.

所以我觉得，在这种多线程的智能体工程工作里，云环境和虚拟环境对提升开发速度特别重要。

And so I do think on this sort of, like, multithreading agentic engineering work, cloud environments and virtual environments are so important to unlock velocity.

而在这方面，我还没看到多少大型工程团队在这些环境上投入足够资源，来真正发挥AI辅助编程或者智能体的潜力。

And that's one place where I haven't seen enough large engineering teams invest in those environments to really unleash the power of, either AI assisted coding for their software engineers or agents in general.

所以如果有CTO或者工程副总裁在听的话，如果你们想在接下来一年里真正推动增长，把这块搞定会非常有用。

So if there are any CTOs, VPs of engineering listening, you if were to invest in something to really unlock growth in the next year, getting that situation locked up would be really good.

因为，我经常听到很多人说，哦，我可以用Claude写所有代码。

Because, again, I hear so many people being like, oh, I can Claude code everything.

我可以花...

I can spend...

你知道，我可以用Codex做任何事。

You know, I can codex anything.

我可以快速创建所有这些工作树。

I can spin up all these worktrees.

我没事。

I'm fine.

然后我就问，你这些都是在本地跑的吗？

And I'm like, are you running all these local?

你到底在干什么？

Like, what are you what are you doing?

所以我想提醒大家的是，别忽略了你自己电脑的性能限制，它能支持多少并行任务，尤其是在像 Stripe 这样复杂的代码库里。

And so that's one thing I just want people to not miss is the limitations of your actual machine on how multithreaded you could be, especially in a complex code base like Stripes.

完全同意。

Totally.

而且，你知道，我手机上有 Slack。

And, you know, I have Slack on my phone.

所以我在上班路上就能启动其中一个“小助手”，比如在地铁上刷 Slack 的时候。

So I can even kick off one of these minions on the way to work, right, as I'm sort of going through Slack on the subway.

然后等我到了公司，我就能中途加入进去。

And then, you know, by the time I'm there, I I I can jump in halfway through.

我觉得这里有点夸张的说法是，想象一下如果一家公司的所有工程师都只能在一个代码库上工作……

I think that that maybe, like, the hyperbolic thing here is, like, imagine if all engineers at a company could only, like, work on...

没有Git。

Didn't have Git.

我们都得协调着一起在同一个代码库上工作。

We all had to, like, coordinate working on the one code base together.

那简直太疯狂了。

Like, that that would be crazy.

而这里的类比是，想象一下如果我被限制在……

And, you know, the equivalency here is, like, imagine if I'm bounded by...

你知道，我的代理只能使用我电脑上现有的资源来工作。

Know, you my agents are bounded by just what's available and can work on my computer.

如果十倍效率的做法是能同时运行十个代理，但又不能依赖我，就像大家都在买Mac mini一样。

If the the the the the 10 x thing to do is, you know, be able to have 10 of them run-in parallel, but also not be contingent on my, like it's like everyone's buying a Mac a Mac mini.

这样它就不会休眠。

So it doesn't fall asleep.

就是，光是为了让电脑不睡着，就有一整个产业了。

It's like, there's a there's a whole business around just the computer not falling asleep.

我真的是……

I legitimately...

首先，我楼上有四台Mac mini，其中一台基本上就是一台不关盖的笔记本电脑。

First of I have, four Mac minis upstairs, and one of them is just basically a laptop that doesn't close.

就是，我把它当……

Like, I use it

对。

Sure.

当一台不关盖的笔记本用，这真的大大提升了我做事的速度。

As a laptop that does not shut, and it's really unlocked my my velocity.

所以……

So...

好的。

Okay.

感谢你陪我们聊了这么多关于虚拟环境和本地主机的事情。

We thank you for going on this side quest about virtual environments and local host and all those things.

我是创始人，所以我知道大多数人创业并不是因为喜欢处理工资单或应付合规问题。

I'm a founder, so I know most people don't start companies because they love running payroll or managing compliance.

但在你招了第一个员工和准备下一轮融资之间，你总会陷入人力资源、IT和其他杂事的泥潭。

But somewhere between hiring your first employee and raising your next round, you end up in the weeds with HR, IT, and all that other stuff.

这正是Rippling要解决的问题。

That's what Rippling was built to solve.

Rippling是一个统一的平台，让初创公司从第一天起就能在一个系统里管理人力资源、工资、IT和财务。

Rippling is a unified platform that lets startups run HR, payroll, IT, and finance in one system from day one.

Rippling的初创公司套件用完全互联的平台取代了那些互不同步的零散工具。

The Rippling startup stack replaces disconnected tools that don't sync with a fully connected platform.

超过15000家初创公司，包括Cursor、Clay和Sierra，都信任Rippling来快速扩展，而无需增加额外的运营和人力资源人员，这样像你这样的创始人就能继续专注于建设。

Over 15,000 startups, including Cursor, Clay, and Sierra trust Rippling to scale fast without adding additional ops and HR head count so founders like you can keep building.

现在，获得风险投资的初创公司可以免费使用六个月的Rippling初创公司套件。

Right now, venture backed startups can get six months of Rippling startup stack for free.

前往 rippling.com/howiai，今天就注册吧。

Head to rippling.com/howiai and sign up today.

就是 rippling.com/howiai，今天注册就能免费使用六个月。

That's rippling.com/howiai to sign up for six months free today.

专注于你正在打造的事业，剩下的交给 Rippling。

Focus on what you're building and leave the rest to Rippling.

好的。

Okay.

所以你现在正在运行这个。

So you are now running this.

你将要……

You're going to...

我……

I...

你一开始说过只有一次机会。

It's it's you said one shot at the beginning.

确实，你是想用一个提示词，而不是每次回复都能得到你想要的结果，但它会进入那个框架里。

Really, you're trying to take one one prompt, and not a single reply gets you gets you what you want, but it goes into the harness.

它会自己循环运行，调用需要的工具，最终你作为最终用户会得到一个回复，那就是成功的实施方案。

It goes through its own loop, hits the tools it needs, and ultimately, you as the end user get one response back, which is here's the successful implementation.

没错。

Exactly.

所以我们已经能看到它在识别相关文件了。

So we can already see that's identifying the relevant files.

它还在自己跟踪待办事项。

It's keeping track of its own to dos.

就像，这是我们给它设定好的重点。

Like, that's something that we've codified in it to to focus on.

它在做修改。

It's making changes.

它在准备提交等等这些。

It's, you know, preparing the the commit and so on and so forth.

最后，等它跑完，我们就会看到结果，就是它完成了。

And and ultimately, in in sort of, like, taking out the oven, we'll see a response at the end of just like, it finished.

你可以直接去看那个拉取请求，然后正常的人工审查环节就继续了。

You know, like, you can go ahead and look at the pull request, and the the the sort of normal human review part continues.

所以

So

我们快速聊聊这个。

Let's talk about that really quickly.

你提到每周有1300个由代理发起的拉取请求，差不多是这个数吧。

You said 1,300 code or agent initiated PRs per week, something like that.

然后人类参与代码审查。

And then humans are involved in code review.

这么多代码审查，你们是怎么完成的？

How are you getting all this code review done?

嗯，你可以这么说，如果我省下主动写代码的时间，我就能把时间重新集中在审查正在写的代码上，或者和用户合作等等。

Well, you can make the argument that, you know, if I'm spending less time actively writing code, I can, you know, know, recenter my time on reviewing the code that's being written or working with users and so on and so forth.

所以我觉得这是很重要的一部分。

So I I think that's a big part of it.

我觉得另一方面，又回到了那个CI环境。

I I think the the other side of it, like, it comes back to that CI environment.

所以要有很好的测试覆盖率，要有能模拟产品端到端交互的合成测试。

So having really good test coverage, having synthetics that run to, you know, simulate end to end interactions with your product.

这些都能帮助你在审查代码时建立信心。

Those all help inspire confidence in the code you're reviewing.

如果没有这些，尤其是在一个庞大的代码库中，要审查代码并对其正确性有高度信心，会非常困难。

So absent those, like, it'd be really difficult to look at code, especially in a huge code base, and have high confidence that it works.

所以，再说一次，不管代码是Steve写的还是Steve的机器人写的，你仍然需要那个CI环境来提供信心，确保被修改的代码是安全的，并且在发布过程中，有蓝绿部署这样的机制可以回滚，所有这些都至关重要，不管代码是谁写的。

So, you know, again, whether the the text has been written by Steve or the text has been written by Steve's robot, you still want that CI environment that's, you know, providing confidence that the the code that's being changed is safe and that as it rolls out, you know, you're having sort of blue green deployments so you can roll back to, like, all that is super critical independent of of of the, you know, the nature of the authoring of it.

我确实相信，如果编码变得更容易，而编码历来是产品开发的瓶颈，那么瓶颈就会转移到其他领域。

I I do I do believe, like, if coding becomes easier and coding historically has been the bottleneck in product development, it's just gonna shift to other areas.

所以，如果编码实际上变得免费，那么审查将会变得非常具有挑战性。

So if, like, coding in effect becomes free, the review is gonna be really challenging.

或者一开始想出足够多的点子可能是个大问题，或者怎么分发这些点子。

Or getting enough ideas in the first place could be a big problem or distributing them.

所以我觉得注意力就会转移到其他方面去。

So I I think the attention is just gonna, you know, move around to to other areas.

好。

Great.

然后在我们进入你的下一个工作流程之前，还有一个问题，我特别期待那个流程，剧透一下，就是除了工程师之外，还有人在用 Minions 吗？

And then one other question before we go on to your next workflow, which I am so excited about, spoiler alert, is are more than engineers using Minions?

你有没有看到产品经理、设计师也加入进来？

Are you seeing product managers, designers come in?

这个在公司里、跨部门之间是怎么推广的？

How is this going across the company and across functions?

我觉得，你知道，我喜欢 Slack 这个例子的部分原因是整个公司都在用 Slack。

I I think, you know, part of why I like the Slack example is the entire company is in Slack.

而且，你知道，从激活能量的角度来看，嗯...

And, you know, to a point of activation energy, you know, eve...

即使你电脑上有文本编辑器，我把文档什么的都给你，但对于不是工程师的人来说，这可能会非常困难或者让人望而却步，或者不管是什么情况。

Even if, like, you had the text editor on your computer, and I and I gave you the docs and and and whatever it may be, you know, to to someone who's not an engineer, it could be really challenging or intimidating or or whatever it may be.

而且，你知道，对于……

And, you know, for...

无论你只是想要一个概念验证，还是打算修改文档，或者不管是什么，你大概可以用纯文本写出你想要实现的事情。

Whether you just want, like, a proof of concept or you're gonna make a docs change or or whatever it may be, like, you can you can probably write out in plain text the the thing you want to occur.

你可能在写产品简报，或者给出设计反馈。

You you might be writing the product brief or you might be giving design feedback.

实际上，你在某种程度上就是在写提示词。

Like, you're you're in effect just writing a prompt at some point.

所以，能够直接点击一个表情符号，或者标记机器人来启动 Minion，我们正努力让更多非工程师使用它。

So being able to just click an emoji or or or, you know, tag the the robot to spin off minion, we're we're trying to see more nonengineer usage there.

是的。

Yep.

太棒了。

Amazing.

好的。

Okay.

那我们进入下一个工作流程，这个我真的很期待。

So let's go to our next workflow, which I am Yes.

超级兴奋。

Psyched.

作为一个楼下堆了一堆Mac mini的人，我对这个特别感兴趣。

As somebody with a stack of Mac minis downstairs, I am excited about.

在Stripe，我们其实从几个角度来思考AI。

So, you know, at at Stripe, we're, you know, we're thinking about AI in a a few ways.

刚才展示的demo，是我们如何利用AI来加速内部产品开发和工程效率。

So the demo we just showed is how we're thinking about using AI internally to accelerate our product development and and engineering.

第二个角度，是我们如何支持那些在自己的产品中运用AI的企业，以及如何帮助他们的商业模式。

The second way is, you know, thinking about how we're supporting all these businesses that are, you know, leveraging AI in their own products and how we can support their business models.

比如通过按用量计费，我们刚刚发布了LM token计费产品的测试版。

And, you know, that's with things like usage based billing, we we just announced our our beta of our LM token billing product.