Learning on the Shop floor 在车间学习
Years ago I wrote about my 多年前我写过关于我的. I dropped out of school at 16 and went to work at a Siemens subsidiary, where the most interesting people sat in the basement and used Delphi instead of the corporate-mandated Rosie SQL (both pretty much lost to time and progress). I learned to be a programmer by watching them. By making them coffee. By hanging around long enough that their judgment seeped into mine.。我16岁辍学后去西门子子公司工作，最有趣的人坐在地下室，用 Delphi 而不是公司强制使用的 Rosie SQL（两者都已被时代淘汰）。我通过观察他们学习编程。给他们冲咖啡。长时间待在他们身边，让他们的判断潜移默化到我身上。
I have been thinking about that experience a lot in the last year, because we built something at Shopify that runs on the same principle.过去一年我一直在思考这段经历，因为我们在 Shopify 构建了一个基于同一原理的系统。
She's called River. River is an AI agent that lives in our company's Slack. You talk to her the same way you would talk to a teammate: by mentioning River in a Slack channel. She can read code, run tests, write code, open pull requests, query our data warehouse, look at production traces, and a lot more. We use this constantly.它叫 River。River 是一个居住在我们公司 Slack 里的 AI 代理。你可以像和同事一样通过在 Slack 频道中提及 River 与她对话。她可以阅读代码、运行测试、编写代码、打开拉取请求、查询我们的数据仓库、查看生产追踪等。我们经常使用它。
In the last 30 days, 在过去的 30 天里，5,938 Shopify employees worked with River across 4,450 different Slack channels5,938 名 Shopify 员工在 4,450 个不同的 Slack 频道中使用了 River. It opened 。它打开了1,870 pull requests in the last week alone仅在上周就有 1,870 个拉取请求 in our main monorepo. About one in eight pull requests merged into our codebase last week was authored by River, reviewed by us. 在我们的主 monorepo 中。上周约每八个合并到代码库的拉取请求中，就有一个是由 River 编写、我们审阅的。
There are a lot of coding agents in the world right now. What makes River special is a constraint: 现在世界上有很多编码代理。让 River 与众不同的是一个约束：She only works in the open.她只在公开环境中工作。
A constraint that became a feature这个约束成为了特性
When we started building River, the obvious thing to do was let people use her in private. That is how many other AI assistants work. ChatGPT is a private window. Claude is a private window. Cursor is between you and the IDE.当我们开始构建 River 时，显而易见的做法是让人们在私密环境中使用它。这也是许多其他 AI 助手的工作方式。ChatGPT 是私密窗口。Claude 是私密窗口。Cursor 则介于你和 IDE 之间。
We made the opposite decision. River lives in slack, our company chat. River does not respond to direct messages. She politely declines and suggests to create a public channel for you and her to start working in. I myself work with river in 我们做了相反的决定。River 生活在 Slack——我们的公司聊天工具中。River 不会回复私信。她会礼貌地拒绝，并建议创建一个公开频道让你们一起工作。我本人在 channel and many followed this pattern. Every conversation is therefore searchable. Anyone at Shopify can jump in. In my own channel, there are over 100 people who, react to threads, add color and add context, pick up the torch, help with the reviews, remind me how rusty I am, and importantly, learn from watching.频道中使用 River，许多人也遵循了这种模式。因此每一次对话都是可搜索的。Shopify 的任何人都可以加入。在我的频道里，有超过 100 个人会对线程做出回应、补充细节、提供上下文、接力、帮助审查、提醒我生疏的地方，最重要的是，通过观察学习。
This was odd at first. People are used to private workspaces with their tools. Asking for help feels different when the whole company can see the question. But something happened that we hoped for but did not fully predict the impact of:起初这很奇怪。人们习惯于在私密工作空间里使用工具。当整个公司都能看到你的提问时，求助的感觉会不一样。但出现了一件我们期待却未完全预料到影响的事：
People started learning from each other.人们开始相互学习。
A support engineer in 一位支持工程师在 would watch a backend engineer in another channel get River to find the right log query, and the next day she would do the same thing. A new hire would scroll back through 的频道里看到另一个频道的后端工程师让 River 找到正确的日志查询，第二天她也会这么做。新入职的员工会翻回 to see how senior people scope a request before they ever sent their first one.查看资深人员在发送第一条请求前是如何定义需求的。
As so often with German, there is a word for the kind of environment: 正如德语中常有的那样，有一个词可以形容这种环境：Lehrwerkstatt.Lehrwerkstatt。 Literally: 字面意思是：A teaching workshop教学工作坊. The whole shop floor is the classroom. You learn by being near the work. Being a constant learner is one of the core values of the firm. 。整个车间就是教室。你通过靠近工作来学习。持续学习是公司核心价值之一。
Shopify wants to be a Lehrwerkstatt at scale and River has now gotten us closer to this ideal than ever. It’s Shopify 想要成为大规模的 Lehrwerkstatt，而 River 让我们比以往更接近这一理想。它是osmosis learning,渗透式学习， because it does not require a curriculum, a training plan, or a manager. It just requires everyone's work to be visible to the maximum extent possible. Everyone learns from each other. 因为它不需要课程、培训计划或经理，只需要让每个人的工作尽可能可见。每个人都能相互学习。
I'm genuinely excited by this- somewhat accidental- discovery and thought I'd share. 我对这次——有点偶然——的发现感到由衷兴奋，想与大家分享。
Why this matters more, not less, with AI为什么这在 AI 时代更重要，而不是更不重要
A common worry about AI is that it will make people stop thinking. Why would a junior developer learn to debug if the agent does it for them? Why would they read the codebase if they can just ask?人们常担心 AI 会让人停止思考。初级开发者如果让代理帮他们调试，还会去学习调试吗？如果可以直接提问，他们还会去阅读代码库吗？
I think the worry is real but the framing is wrong. The risk is not that AI does the work. The risk is that AI does the work and we never learn from it我认为这种担忧是有道理的，但框架错了。风险不在于 AI 完成工作，而在于 AI 完成工作而我们从中没有学习。. If every interaction with an agent happens in a private window, the only person who learns anything is the person at the keyboard. Everyone else is locked out of the apprenticeship.如果每一次与代理的交互都发生在私密窗口，唯一能学到东西的只有键盘前的那个人。其他人被排除在学徒制之外。
When people work together with their agents in public, the opposite happens. The best prompt patterns spread, knowledge spreads. The clever way one developer investigated a Slack permissions bug becomes the template for how everyone else investigates. The skill someone wrote to teach River about the company's checkout data warehouse gets reused by twelve other teams. River herself learns: every channel can pre-load the zones, skills, and instructions its team needs, written by the people closest to the work. River also has a memory that is constantly learning and un-learning critical information about the company and the best way to do work.当人们在公开环境中与代理协作时，情况恰恰相反。最佳的提示模式会传播，知识会扩散。某位开发者巧妙地调查 Slack 权限 bug 的方式会成为所有人调查的模板。某人写给 River 的关于公司结账数据仓库的技能会被另外十二个团队复用。River 本身也在学习：每个频道可以预加载该团队所需的领域、技能和指令，由最贴近工作的人员编写。River 还有一个不断学习和忘记公司关键信息以及最佳工作方式的记忆。
The agent does not replace the apprentice, nor does it replace the mentor. The agent makes the 代理并不取代学徒，也不取代导师。代理让whole company整个公司 an apprentice because everyone is constantly watching the most experienced people work alongside it.成为学徒，因为每个人都在不断观察最有经验的人与它一起工作。
This is also why the merge rate keeps climbing. We did not retrain a model. We did not switch models. An improvement from 36% to 77% over two months came from people watching River work, noticing where it got stuck, and writing down what it should have known and helping make River itself a better teammate. Every team's accumulated taste flows into the agent. The agent gets better at being Shopify.这也是合并率持续上升的原因。我们没有重新训练模型，也没有更换模型。两个月内从 36% 提升到 77% 的改进，来源于人们观察 River 工作、发现它卡住的地方、记录它本该知道的内容并帮助让 River 成为更好的队友。每个团队累计的品味都会流入代理。代理变得更像 Shopify。
The company moves at the speed of its slowest secret公司以最慢的秘密的速度前进
When I think about why this matters, it comes back to something I have believed for a long time: the speed of an organization is determined by the speed of its lowest-bandwidth communication channel and rhythm. Meetings are slow. Email is slow. Private DMs are slow. Maybe not for the individuals involved in them, but for the organization. The information and decisions that come from them never fully diffuse into the rest of the organization without huge additional communication effort.当我思考这为何重要时，我回想起长期坚持的信念：组织的速度取决于其最低带宽的沟通渠道和节奏。会议慢，邮件慢，私信慢。也许对参与者个人来说不慢，但对组织而言是慢的。来自这些渠道的信息和决策若没有巨大的额外沟通努力，永远无法完全扩散到组织的其他部分。
A public conversation between humans or with a competent agent is none of those things. It is fast, it is searchable, it is teachable, and it compounds. The next person who has the same question does not have to ask it.人与人之间或与有能力的代理之间的公开对话则不同。它快速、可搜索、可教学，并且会产生叠加效应。下一个有相同问题的人不必再提问。
I do not think the future of work is humans being replaced by agents. I wrote a piece in 2018 called 我不认为未来的工作是人类被代理取代。我在 2018 年写过一篇文章《, about how chess got more popular, not less, after computers learned to play. The same lesson applies here. The right model is not human ，讨论了在计算机学会下棋后，棋类并没有变得不受欢迎，反而更受欢迎。同样的教训也适用于这里。正确的模型不是人类or machine. It is the apprentice and the master, both watching each other learn, both getting better on the shop floor.机器，而是学徒与师傅，彼此观看对方学习，在车间共同进步。
That is what River is. This is our Lehrwerkstatt.这就是 River。这是我们的 Lehrwerkstatt。
想发布自己的文章？想发布自己的文章？
升级为 Premium升级为 Premium