AI 在就业回流中的作用与工作的未来

你正在打造的业务、团队和运营方式，正是这个 AI 时代各公司尝试运营的最前沿。

The business you're building, the team you're building, the way you're operating is the very bleeding edge of how companies are trying to operate in this AI era.

我们有一位 AI 运营负责人。她一直在构建提示词和工作流，让我和团队里的其他人尽可能多地实现自动化。

We have a head of AI operations. She's just constantly like building prompts and building workflows so that I and everyone else on the team are just automating as much as possible.

关于 AI，有哪些你相信但大多数人并不相信的事情？

What are some things that you believe about AI that most people don't?

我讨厌那些“入门级工作被 AI 取代”的标题。每当我看到一个孩子用 ChatGPT，我就会想：“天哪，他们会比我共事过的任何人都快得多。”我们有个小伙子，他在两个月内取得了相当于一年的进步，因为每次我坐下来告诉他：“好，故事要这么讲，标题要这么想。”他就把这一切都记录下来，放进提示词里，从此再也不会犯同样的错误。

I hate the headlines that are like entry-level jobs are taken away by AI. Whenever I see a kid with ChatGPT, I'm like, "Holy, they're going to go so much faster than any other person that I've worked with." We have this guy, he made like a year's worth of progress in like two months because every time I sat down with him and told him, "Okay, here's how you tell a story. Here's how you think about a headline." Like, he recorded all of it, put it into a prompt, and he never made the same mistake twice.

有种感觉是，我们正走向一个不必编写任何代码的境地。比如，你的产品团队完全不写代码。

There's this sense we're getting to a place where you don't have to write any code. Like, you have a product team not writing code at all.

已经没有人再手动编码了。像我们这样的组织，那些处于前沿的人。我们现在做的事情，三年后所有人都会这样做。

No one is manually coding anymore. Organizations like ours, people who are playing at the edge. We're doing things that in like three years everybody else is going to be doing today.

我的嘉宾是 Dan Shipper。Dan 是 Every 的联合创始人兼 CEO，Every 是一家处于 AI 可能性最前沿的公司。他们仅有 15 名员工，却已经构建并发布了四款不同的产品。他们每天发布一份新闻通讯，还有一个咨询部门，帮助公司采用最新的 AI 最佳实践。在他们的产品团队中，工程师不手写一行代码，而是使用一系列智能体来帮助他们制定需求并构建产品。他们的编辑部门使用 AI 更快地发布更好的作品。他们甚至有一个人的全部工作就是帮助公司每位员工利用最新的 AI 工作流提高效率。在我们的对话中，Dan 分享了他们在内部使用的一系列策略，以提升员工的杠杆效应，他的个人 AI 工具栈，他发现的预测公司能否通过 AI 成功获得巨大生产力提升的一个指标，他如何以非常独特的方式构建公司，对 AI 未来的一系列预测，以及更多内容。如果你喜欢这个播客，别忘了在你最喜欢的播客应用或 YouTube 上订阅并关注。另外，如果你成为我新闻通讯的年度订阅者，你将免费获得一年的一堆惊人产品，包括 Superhuman、Linear、Notion、Perplexity、Bolt、Granola 等。请访问 lenny'snewsletter.com 并点击 bundle。接下来，有请 Dan Shipper。

My guest is Dan Shipper. Dan is the co-founder and CEO of Every, which is a company that is at the very bleeding edge of what is possible with AI. Their team of just 15 employees has built and shipped four different products. They publish a daily newsletter and they have a consulting arm that helps companies adopt the latest AI best practices. On their product team, their engineers don't handwrite a single line of code and instead use an arsenal of agents who help them craft requirements and build their products. Their editorial arm uses AI to publish better work faster. And they even have a person whose entire job is to help every employee at the company become more efficient using the latest AI workflows. In our conversation, Dan shares a bunch of tactics that they use internally to increase the leverage of their own employees, his personal AI tool stack, the one predictor that he's found for whether a company will successfully find huge productivity gains through AI, how he's building his company in a really unique way, a bunch of predictions for where AI is going, and so much more. If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube. And also, if you become an annual subscriber of my newsletter, you get a bunch of amazing products for free for one year, including Superhuman, Linear, Notion, Perplexity, Bolt, Granola, and more. Check it out at lenny'snewsletter.com and click bundle. With that, I bring you Dan Shipper.

本期节目由 CodeRabbit 赞助播出，这是一个 AI 代码审查平台，正在改变工程团队如何借助 AI 更快交付而不牺牲代码质量。代码审查至关重要但耗时。CodeRabbit 充当你的 AI 副驾驶，为每个拉取请求提供即时代码审查评论和潜在影响。除了标记问题，CodeRabbit 还提供一键修复建议，并允许你使用 GP 模式定义自定义代码质量规则，捕捉传统静态分析工具可能遗漏的细微问题。CodeRabbit 还直接在 IDE 中提供免费的 AI 代码审查。它适用于 VS Code、Cursor 和 Windsurf。到目前为止，CodeRabbit 已审查超过 1000 万个 PR，安装在 100 万个仓库中，并被超过 7 万个开源项目使用。使用代码 Lenny 在 coderabbit.ai 免费获得一整年的 CodeRabbit。网址是 coderabbit.ai。

This episode is brought to you by CodeRabbit, the AI code review platform, transforming how engineering teams ship faster with AI without sacrificing code quality. Code reviews are critical but time consuming. CodeRabbit acts as your AI co-pilot, providing instant code review comments and potential impacts of every pull request. Beyond just flagging issues, CodeRabbit provides one-click fix suggestions and lets you define custom code quality rules using GP patterns, catching subtle issues that traditional static analysis tools might miss. CodeRabbit also provides free AI code reviews directly in the IDE. It's available in VS Code, Cursor, and Windsurf. CodeRabbit has so far reviewed more than 10 million PRs, installed on 1 million repositories, and is used by over 70,000 open-source projects. Get CodeRabbit for free for an entire year at coderabbit.ai using code Lenny. That's coderabbit.ai.

今天的节目由 DX 赞助播出。如果你是工程领导者或平台团队成员，你的 CEO 最终必然会向你询问生产力指标。但衡量工程组织是困难的。我们都同意，像 PR 数量或提交次数这样的简单指标并不能说明全部情况。这就是 DX 的用武之地。DX 是由领先研究人员设计的工程智能解决方案，包括 DORA 和 SPACE 框架背后的研究人员。它将来自开发者工具的定量数据与来自开发者的定性反馈相结合，为你提供工程生产力及其影响因素的完整视图。了解为什么世界上一些最具标志性的公司，如 Etsy、Dropbox、Twilio、Vercel 和 Webflow，都依赖 DX。请访问 DX 网站 getdx.com/lenny。

Today's episode is brought to you by DX. If you're an engineering leader or on a platform team, at some point your CEO will inevitably ask you for productivity metrics. But measuring engineering organizations is hard. And we can all agree that simple metrics like the number of PRs or commits doesn't tell the full story. That's where DX comes in. DX is an engineering intelligence solution designed by leading researchers, including those behind the DORA and SPACE frameworks. It combines quantitative data from developer tools with qualitative feedback from developers to give you a complete view of engineering productivity and the factors affecting it. Learn why some of the world's most iconic companies like Etsy, Dropbox, Twilio, Vercel, and Webflow rely on DX. Visit DX's website at getdx.com/lenny.

Dan，非常感谢你来到这里，欢迎来到播客。

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

谢谢你邀请我。显然我一直是你的忠实粉丝，所以能来到这里是我的荣幸。

Thank you for having me. I've obviously been a huge fan for a long time and so it's an honor to be here.

这是我的荣幸，Dan。我觉得这期播客是命中注定的。我很高兴我们终于做到了。有太多我想聊的了。有太多我们可以聊的了。我想从一些热门观点开始会很有趣。我之所以想从这里开始，是因为我觉得你在思考 AI、用 AI 构建、使用 AI、评估 AI 上花的时间比我认识的任何人都多。所以我非常尊重你对未来走向的见解和观点。那么，让我问你这样一个问题，看看会聊到哪里。关于使用 AI 工具，有哪些你相信但大多数人并不相信的事情？

It's my honor, Dan. I feel like this is a podcast that was meant to be. I'm so happy we're finally doing this. There's so damn much that I want to talk about. There's so damn much we can talk about. I thought it'd be fun to start with just some hot takes. And the reason I want to start here is I feel like you spend more time thinking about AI, building with AI, using AI, evaluating AI than anyone else I know nearly. And so I really respect your insights and your perspectives on where things are going. So, let me just ask you this kind of question and see where this goes. What are some things that you believe about AI using AI tools that most people don't believe?

我要先说我最热门的观点，这也是我证据最少的观点。所以，我们就从那个开始吧。我还有其他更有理有据的观点可以给你，但这是我最热门的，那就是我认为 AI 可能是美国就业回流的最大推动力之一。所以，我认为每个人都担心它会让人失业。当然，它会改变你工作所需的技能，但我认为它实际上可能会让很多工作回流。它会通过两种方式实现。一是现在有很多昂贵的服务，富人和大公司正在为此付费，比如内部法律顾问或呼叫中心之类的。而廉价智能的作用是让这类服务对小公司和个人来说变得负担得起，从而刺激需求。它做的另一件事是让从事这些工作的人能够廉价地服务更多人。所以，如果你做客户服务，例如，它可能不会消除客户服务，但它可能让中西部通常会在呼叫中心工作的 10 个人服务数十万或数百万人。也许这有点夸张，但肯定比他们一直打电话时服务的人多得多。因此，美国公司雇佣美国本土员工会变得更具成本效益。而且我认为在很多情况下，美国人会更擅长使用这些 AI 工具来完成工作。

I'm going to go with my hottest take and this is the take that I have the least evidence for. So, let's just start with that. I have other more well-reasoned takes to give you, but this is my hottest one, which is I think that AI may be one of the biggest forces for reshoring American jobs. And so, I think everyone is worried about it unemploying people. And for sure, it will change the skills needed to do the jobs that you're doing, but I think it may actually reshore a lot of jobs. And it'll do that in two ways. One is there are a lot of expensive services that rich people and big companies are paying for right now. So like an in-house counsel or like a call center or whatever. And what cheap intelligence does is it makes those kinds of things affordable for small companies and individuals. So it stimulates demand. The other thing that it does is it allows people who are in those jobs to serve more people cheaply. So if you're a customer service, for example, it may not get rid of customer service, but it may allow 10 people in the Midwest who would normally be working at a call center to serve hundreds of thousands or millions of people. Maybe that's too much, but like a lot more people than they would ordinarily if they were the ones on the phone all the time. And so it becomes much more cost-effective for American companies to hire people in the US. And I think the people in the US are going to be better in a lot of cases at using these AI tools to do work.

所以我认为，让美国的工作由身处美国的人用 AI 来完成，实际上可能更高效，而且模型公司也在这里。所以有很多美国相关的事情在发生，你可以自己判断这是好事还是坏事，但我觉得在关于 AI 是否会取代工作的讨论中，这一点被忽略了。

So I think it may actually make it more effective to have those jobs in the US run by people sitting in the US who are using it to get work done, and also the model companies are here too. So there's a lot of American stuff happening, and you can decide whether or not you think that's a good thing, but I think it's quite lost in the conversation over whether AI will get rid of jobs.

我喜欢关于 AI 的乐观看法，这很棒。而且就像你说的，这对其他国家好不好还有待观察，但对美国是好的。还有什么？你还有什么其他大胆的观点？

I like optimistic takes about AI. So this is great. And to your point, TBD if this was good for other countries but good for the US. What else? What else you got? What other hot takes?

另一个大胆的观点，这个不太反主流，而是我觉得人们真的低估了 Claude Code 对非程序员的价值。而且我要扩展一下，不只是 Claude Code，谷歌也刚推出了 Gemini CLI 命令行界面，诸如此类。我来给听众解释一下 Claude Code 是什么。Claude Code 就是一个命令行界面，就是程序员用的那种黑色终端。你可以启动它，它能访问你的文件系统，知道怎么用各种终端命令，还能浏览网页等等。你给它一个任务，它就会自己运行 20 或 30 分钟，自主地、智能体式地完成任务。尤其是刚推出的 Claude Opus 4，这是 AI 自主工作能力的一次巨大飞跃，而且 Claude Code 还能生成多个子智能体并行处理任务，对程序员来说极其有用。比如 Anthropic 内部每个人整天都在用它，每个人都像吃了智能体药丸一样，有 15 个智能体在跑各种任务，简直疯狂。但非程序员不用它，因为用终端让人望而生畏。但你可以，比如下载你所有的会议记录，放在一个文件夹里，然后说：“好，我想让你读我所有的会议记录，告诉我一些我做的事，比如告诉我所有我微妙地回避冲突的时刻。”它会给自己写一个待办清单，它有一个小笔记本，可以逐个读取每个文件，然后写进笔记本，按清单执行，最后给你一个总结性的回答，而且是多轮交互的。所以它不是简单地把所有内容塞进上下文，就像你用 ChatGPT 或普通 Claude 聊天那样。它是真正地处理你给它的每一个文件。所以我认为它对于任何涉及处理大量文本的任务都极其强大。简单来说，你基本上在本地电脑上有一个智能体，可以读取你的本地文件并为你执行任务。

Another big hot take, and this is less contrarian and more like I think people are truly sleeping on it. I think people are truly sleeping on how good Claude Code is for non-coders. And I'll extend this to not just Claude Code, but Google just came out with the Gemini CLI command line interface. So things like that. And I'll tell you about, for people who are listening that don't know what Claude Code is. Claude Code is just a command line interface. So it's those black terminals that programmers use. It's a command line interface that you can boot up. It has access to your file system. It knows how to use any kind of terminal command and it knows how to browse the web, all that kind of stuff. You can give it something to do and it will go off and it will run for like 20 or 30 minutes and complete a task like autonomously, agentically. It's a, especially with Claude Opus 4 that just came out, it's like this gigantic leap forward in AI's ability to work by itself, and Claude Code can even spawn multiple sub-agents that do a bunch of tasks in parallel, and it's incredibly useful for programmers. Like everybody inside of Anthropic is using it all day every day. Like everyone's agent pill. They've got like 15 agents doing all this kind of stuff. It's crazy. But non-programmers don't use it because it's intimidating to use the terminal. But you can like download, for example, you can download all your meeting notes and put it in a folder and just be like, "Okay, I want you to read every single one of my meeting notes and tell me something that I do, for example, tell me all the time that I subtly avoided conflict." And it will write a little to-do list for itself. It can have like a little notebook. It can go and read each little thing and then write into its notebook, go down a to-do list and give you a summarized answer over multiple turns. So it's not just like stuffing everything into context, which is what you'd be doing with like a ChatGPT or a regular Claude chat. It's like actually processing every single file that you give it. And so I think it's incredibly powerful for any kind of task that involves processing a lot of text. So, as a simple way to think about this, you basically have an agent on your local computer that can read your local files and do your bidding.

是的，完全正确。而且它能长时间运行而不出岔子。

Yes, exactly. And it can do that for long amounts of time without going off the rails.

有意思。所以非技术人员需要克服一个小障碍，就是使用终端和输入命令，但一旦运行起来，你只需要用英语和它对话，让它做事就行了。

Interesting. And so there's like a small hurdle that non-technical people have to overcome, which is using their terminal and giving commands, but once they get it running, it's just you talk to it in English and ask it to do stuff.

完全正确。所以这里的核心观点就是，Claude Code 这个大多数人认为是为工程师设计的工具，其实是非技术人员最被低估的工具。

Exactly. So the hot take here is just Claude Code, which most people think is for engineers, is the most underrated tool for non-technical people.

是的，完全正确。

Yeah, exactly.

你觉得人们还能怎么用这个？这个会议记录的例子真的很酷，我能想到人们会这么用。你还见过或想到其他用法吗？

What are some other ways you imagine people seeing this? This meeting note example is really cool and I could see people using this. What else have you seen or think?

我经常做的一件事。我的工作很大一部分是写作，比如，我喜欢——我知道你会问我喜欢的书，所以我先透露一下——我喜欢《战争与和平》。我刚读了第三遍。

Something that I've done a lot. So I'm a writer for a lot of my job, and for example, I love, and I know you're going to ask me about books I love, so I'm going to give you a sneak peek, which is I love War and Peace. I just read it for the third time.

哇。

Wow.

它太长了，但太好看了。我觉得托尔斯泰是个天才作家。我想做的一件事是，我想让我的写作带有一些托尔斯泰的风格。我做到这一点的方式是，我觉得他非常擅长用那些微妙的句子，通过人物的行为来展现他们的想法和感受，比如他们的面部表情，或者他们声音的语调与眼神之间的不匹配，诸如此类。他简直是人类行为和心理学的大师。所以我直接把《战争与和平》下载到电脑上，因为它是公有领域，可以下载。然后我让 Claude 读《战争与和平》的前三章，提取出所有那些描写，然后为它自己制作一个指南，教它如何像托尔斯泰那样进行人物描写。你完全可以用普通的 Opus 命令做到这一点，但你不可能把整本《战争与和平》都塞进去。要让它做到这一点，需要更多的引导，而它几乎是自主完成的，不需要我真正干预。它最后还下载了俄语版和英语版的《战争与和平》，然后开始比较我喜欢的不同场景，告诉我翻译中可能遗漏的东西。所以你可以根据你关心的任何细分领域，深入到你想要的程度，无论多深、多怪、多书呆子气都行。同样，如果你有大量的客户访谈或客户数据需要处理，它对于从这样的大型数据集中找出信息也极其强大。

It's so long, but it's so good. I think Tolstoy is a brilliant writer. And one thing that I wanted to do was I was like, I want to inflect some of my writing with some of Tolstoy's style. And the way I did that is, I think he's incredible at these little subtle sentences where he shows you what a character is thinking and feeling just by how they behave, like how they move their face or the mismatch between the intonation in their voice and the expression in their eyes. Like all that kind of stuff. He's just like an incredible student of human behavior and psychology. And so I just downloaded War and Peace to my computer, which you can do because it's public domain. And then I had Claude read like the first three chapters of War and Peace and pull out all of those descriptions and then make a guide for itself for how to do character descriptions like Tolstoy. And you could totally do this with like a regular Opus command, but you couldn't put all of War and Peace into it. It would take a lot more handholding to get it to do this, and it just sort of did this by itself without me really intervening. It also ended up like downloading, I had it download a Russian version of War and Peace and the English version, and then start comparing different scenes that I love to tell me about things that I might have missed in the translations. So like you can get as deep and weird and nerdy for whatever subfield you care about as you want to. Same thing for like if you've got tons of customer interviews or tons of customer data you want to go through. It's like incredibly powerful for figuring stuff out from big data sets like that.

你其实启发了我去用这个。这不是你描述的用法，但也很酷。这听起来会很书呆子气。我现在在读《安娜·卡列尼娜》，也是托尔斯泰的。这是之前一位播客嘉宾推荐的，所以我想：“好吧，我得读读这个。”也很长。我在 Kindle 上读，我就想：“好吧，才读了 13%，我已经读了好几个月了。”

You actually inspired me to use this. This is not what you're describing, but it's also something that's very cool. This is going to sound so nerdy. I'm reading Anna Karenina right now, based on also Tolstoy. And this is recommended by a previous podcast guest, and so I was like, "All right, I got to read this." Also very long. I'm on my Kindle. I'm just like, "All right, 13% in. I've been reading for months."

大胆观点。我觉得《战争与和平》比《安娜·卡列尼娜》更好，尤其是对技术人来说。但两本都很好。

Hot take. I think War and Peace is better than Anna Karenina, especially for like a tech person. But they're both good.

好吧，这就对了。这就是我的一年。嗯，我看到你发过一条推文，讲这个我一直在用的用例，我很喜欢，就是在我读书的时候，让 ChatGPT 语音助手在旁边，然后直接问它问题，因为你实际上不需要把书喂给它。它知道整本书。而且 Anthropic 刚刚分享了这一点，我不知道是他们分享的还是有人在他们的法律简报中发现的，他们实际上买了大量的书并自己扫描了。是的。

Okay, there we go. There's my year. Um, I saw you tweet this use case that I love that I've been using, which is just while I'm reading, having ChatGPT voice sitting around and then just asking it questions because you don't actually have to feed it the book. It knows the whole book. And Anthropic just shared this. I don't know if they shared or someone found this in their legal briefings that they actually bought tons of books and scanned them themselves. Yeah.

这就是他们如何做到合理使用的。所以它有所有这些上下文。所以坐在那里问它，比如俄罗斯社会里的这个东西到底是什么，超级有趣。

Is how they did fair use. And so it has all this context. So just sitting there asking it like what the heck is this thing in Russian society is super fun.

这太棒了。所以这里的要点就是回到你的热点话题：你可以让一个智能体使用本地文件，在你的电脑上做各种酷炫的事情，而不是必须把它上传到项目或提示词里之类的。

So this is awesome. So the tip here is just coming back to your hot take: you can have an agent using local files and doing all kinds of cool stuff on your computer versus having to upload it into projects or into your prompts and things like that.

是的。

Yeah.

超级酷。所以我猜这里的赌注是，人们会发现这一点，并开始在日常中使用它。

Super cool. So I guess the bet here is that people are going to discover this and start using this just day-to-day.

我认为他们绝对会。而且我还认为，模型公司可能会开始让这变得更易用。比如我认为，从 Claude Code 和其他类似工具中产生的东西，会进入你使用的所有其他东西，无论是在网页上还是其他地方，最初的 AI 应用都是把聊天框粘贴到现有 UI 中。所以你知道，Copilot 在 IDE 里有自动补全。Cursor 有一个带小聊天的侧边栏。而 Claude Code 的不同之处在于，你从不看代码。它不是为手动编码设计的。它是为了让你说：“我希望你完成某件事。”然后它就去做了。我认为我们正接近一个点，对于几乎所有这些常见的应用，AI 将足够好，以至于我们可以在很大程度上摆脱那些界面，那些你需要深入查看它实际在做什么、与它的执行过程交织在一起的界面，而你更像是说：“我在委派任务。它会去完成。”

I think they absolutely will. And I also think probably the model companies are going to start making this more accessible. Like I think one of the things that will just come from Claude Code and other things like it into everything else you use, whether it's on the web or wherever, is the original AI apps were pasting a chat box into an existing UI. So you know you've got Copilot, it's got the autocomplete in the IDE. You've got Cursor, it's got a little sidebar with a little chat. And the difference with Claude Code is you never look at the code. It's not meant for coding by hand. It's meant for you to say, "I want you to get something done." And it goes and does it. And I think we're just getting to a point where for pretty much all of these, all the usual applications, AI is going to be good enough that we can get rid of the interfaces more or less where you're digging into all the things that it's actually doing and you're sort of interleaved with its execution, and you're more just like, "I'm delegating. It's gonna go do it."

是的。我请过 Cursor 的 CEO Michael Terrell 上播客，这就是他的宏大愿景：代码之后是什么。

Yeah. I had Cursor CEO Michael Terrell on the podcast, and this is his big vision: what comes after code.

完全正确，完全正确。而且我刚刚还请了 Base 44 的创始人上播客，他把这家公司以 8000 万美元卖给了 Wix。他分享说，在过去的三个月里，他没有碰过一行前端代码，全部用 Base 44，或者，抱歉，全部用 Cursor 和他使用的其他工具。所以这正在发生。

Exactly, exactly. And I also just had the founder of Base 44 on the podcast, who sold this company for 80 million bucks to Wix. And he shared that for the last three months, he hasn't touched a single line of front-end code, all Base 44 or, sorry, all Cursor and other tools he's using. So this is happening.

每家公司内部的人也是如此，没有人再手动编码了。

Same thing for people inside of every company, no one is manually coding anymore.

好的。在我们做之前，绝对需要谈谈这个。你还有什么其他热点想抛出来吗？

Okay. Definitely need to talk about that before we do. Any other hot takes that you want to throw out there?

我还有一个热点，就是我对 AGI 有一个定义。AGI 是出了名的难以定义，比如它成为通用人工智能意味着什么？图灵测试是一个，但我们在很多方面已经远远超越了图灵测试。所以我们没有一个好的定义。而我注意到的是，你可以通过你能给 AI 多长的“绳子”让它去工作，来判断它变得有多好。所以对于 Copilot，就像你可以用 Tab 补全，那只是开始。对于 ChatGPT，你问它一个问题，它返回一个回答，那可能比 Tab 补全稍微好一点。然后现在有了 Claude Opus 4 和 Gemini 之类的东西，它可以去做事，还有深度研究，它可以去做 20 或 30 分钟的工作。所以那条“绳子”变得越来越长，你需要干预的地方越来越少。我想到这个，它让我想起了 Winnicott，他是一位儿童心理学家。他写了一本书叫《游戏与现实》。他对成为成年人意味着什么的概念，从婴儿到儿童再到成年人的过程，是你刚出生时，你实际上与你的母亲或照顾者融合在一起。你和她或你和你的照顾者之间没有区别。而成长是一个逐渐在某些时刻被放下、你能承受被放下的过程。所以你学会了你和你的照顾者之间是有分离的。所以对于婴儿来说，就像不是每时每刻都黏在一起，你会被单独留下。也许就像你被单独留下哭出来。谁知道这对婴儿是不是正确的做法，有很多争议，但那是在教你，你和你妈妈或你和你爸爸之间是有分离的。不会总有人来抱你。而养育孩子就是知道他们什么时候准备好被放下一点，必须自己站起来。所以我认为人类发展也有同样的“绳子”。就像你会有越来越长的时间可以独自一人。所以我们还在 20 到 30 分钟的阶段，也许，我不知道。我猜你可能不能让一个幼儿独自待 20 到 30 分钟，但它比幼儿大一点。

I have one other hot take, which is I have a definition for AGI. So AGI is famously hard to define, like what does it mean for it to be artificial general intelligence? The Turing test was one, but we've pretty much blown past the Turing test in a lot of ways. So we have no good one. And so what I have noticed is that you can tell how much better AI is getting by how long a leash you can give it to do work. So with Copilot, it was like you can tab complete, and that was like the beginning. With ChatGPT, you ask it a question and it returns a response, and that's maybe slightly better than a tab complete. And then now with Claude Opus 4 and Gemini and all that kind of stuff, it can go off and work for, also with deep research, it can go off and work for like 20 or 30 minutes. So that leash is getting longer where you have to intervene. And I was thinking about this, and it reminded me of Winnicott, who's a child psychologist. He wrote this book called Playing and Reality. And his conceptualization for what it means to become an adult, what it means to go from being an infant to a child to an adult, is when you're first born, you're effectively fused with usually your mother, your caregiver. There's no difference between you and her or you and whoever your caregiver is. And growing up is this process of being gradually let down in certain moments where you can handle being let down. So you learn that there's a separation between you and your caregiver. So for infants, it's like instead of being fused at the hip for every hour of every day, you get left alone. Maybe it's like you get left alone to cry it out. Who knows if that's the right thing to do with infants, a lot of consternation there, but that's teaching you that there's a separation between you and your mom or you and your dad. There's not going to always be someone to pick you up. And raising a child is about knowing when they're ready to be let down a little bit and have to stand up on their own. So I think there's that same leash with human development. It's like you get longer and longer periods of time where you can be on your own. So we're still in the kind of 20 to 30 minutes is like maybe, I don't know. I guess you probably can't leave a toddler alone for 20 to 30 minutes, but it's a little bit older than a toddler.

也许 20 或 30 秒。

Maybe 20 or 30 seconds.

对于幼儿，你可以做到，比如你在同一个房间里，但不是每秒钟都和他们互动，有时可以持续 20 分钟。所以差不多就是这样。我认为 AGI 也有类似的“绳子”。所以我认为 AGI 的一个好定义是：什么时候让人们无限期地运行智能体在经济上变得有利可图？也就是说，它永远不会关闭。它是一个永远在运行的 Claude Code。它总是在做某事。你永远不会关闭它，你也不需要关闭它，因为你知道让它保持运行是值得的。它永远不会等着你说“好，下一件事”。当你像“好，下一件事”时，它总会回应你，但它基本上是在过自己的生活，就像一个青少年。而这对你来说是有利可图的。你宁愿让它那样做，也不愿让它只是等着你告诉它下一步做什么。我认为这是 AGI 的一个好定义。

You can with a toddler, it's like you can be in the same room but not interacting with them every single second, for 20 minutes sometimes. So it's around there. And I think there's a similar leash with AGI. And so I think a good definition of AGI is when does it become economically profitable for people to run agents indefinitely? So it just never turns off. It's a Claude Code that's always running. It's always doing something. You just never turn it off, and you don't need to because you know that it's worthwhile to keep it on. It's never waiting for you to be like, "Okay, next thing." It'll always respond to you when you're like, "Okay, next thing," but it's off essentially living its life like a teenager. And that is profitable for you. You'd rather have it do that than just wait for you to tell it what to do next. And I think that's a good definition of AGI.

而有利可图的部分也只是运行那个东西并拥有它的成本。

And the profitable piece is also just the cost of running that thing and having it.

部分是成本，部分是价值。显然，你可以稍微钻空子，比如“酷，我就让 Claude 永远循环运行。”但我说的不止这些，而是更广泛地采用始终工作的智能体。我喜欢“有利可图”这个标准，因为如果它花费一点钱，而门槛是盈利能力，那么它必须真正为你做有用的事情，你才会让它保持运行。

It's partly the cost and partly the value. And obviously you can game this a little bit and be like, "Cool, I'm just going to tell Claude to run in a loop forever." But I'm talking about more than that, a more widespread adoption of agents that work all the time. And I like the profitable thing because if it costs a little bit of money and the bar is profitability, then it has to actually be doing something useful for you to keep it on.

有趣的是，这也非常像高级员工和自主性的隐喻。本质上，他们越自主，你需要给出的指示就越少，你需要做的审查就越少，这也直接与他们的资历相关。

It's interesting how that also is very much the metaphor of a senior employee and autonomy. Essentially, the more autonomous they are, the less instruction you have to give, the less reviews you have to do, is also just directly correlated with how senior they are.

完全同意。

Totally.

好的，很好。还有其他类似的想法吗？

Okay, great. Anything else along these lines?

我的意思是，我有很多。我想我总体上讨厌那些标题，比如“它将取代工作”或“它将让三分之二的劳动力失业”。

I mean, I have plenty of them. I think I'm generally like, I hate the headlines that are like "It's going to replace jobs" or "It's gonna unemploy two-thirds of the workforce."

我觉得那不是真的。我讨厌那些标题，比如“你用 ChatGPT 时不用脑子”，或者另一个好标题是“医生单独、医生加 AI、还是只用 AI，哪个更好？AI 更好。因此医生会被淘汰。”我觉得这些都很蠢。所以对于医生加 AI 的例子，我认为重要的是要认识到使用 AI 是一项技能。如果你研究那些对 AI 没什么经验的医生，确实，你也许能制造出一种情况，只用 AI 更好，而且有时候确实会更好，但医生做决策和做事的情境太多了，很难拿一项研究就得出任何结论。尤其是当你面对一项发展如此迅速的技术，医生还不能被期待成为专家。但我猜五到十年后，情况会完全不一样。

Like I don't think that's true. I hate headlines that are like 'you don't use your brain when you use ChatGPT' or there's another good headline like 'doctors alone, doctors plus AI, or just AI—which one is better? AI is better. Therefore doctors are going to be outmoded.' All that stuff is, I think, pretty dumb. So for the doctors plus AI example, I think it's important to recognize that using AI is a skill. And so if you study doctors in a vacuum that don't really have a lot of experience with AI, yeah, you could probably create a situation such that it's better to just use an AI, and sometimes it is going to be better, but there's so many contexts that doctors need to make decisions and do things that it's really hard to take one study and make any sort of conclusion about that. And it's especially hard when you're dealing with a technology that's developing so rapidly that doctors can't really be expected to be experts at it yet. But I would guess in five or 10 years that will be totally and completely different.

对于学生例子，或者说“AI 让你脑子变懒”的例子，我认为重要的是要理解，在技术史上，你总是为了获得其他技能而放弃某些技能。比如，柏拉图 famously 对书写非常怀疑，因为他认为书写会损害记忆力，事实也确实如此。我们现在不如古人记得那么好，因为他们必须记住长篇史诗来娱乐彼此。但我认为，用稍微差一点的记忆力来换取书写是值得的。我认为 AI 也在发生类似的事情，是的，你可能会在某些任务上稍微不那么投入，但如果你用对了，你会在其他任务上更加投入，因为你有更大的能力。所以你可以设计一项研究说使用 AI 时大脑连接性下降，就像你可以设计一项研究说人们有书写技能时记忆力更差。但我不认为有人想回到一个没有人识字的世界。

For the student example, or the 'AI turns your brain off' example, I think it's really important to understand that in the history of technology, it has always been the case that you give up certain skills in order to get other ones. So, for example, Plato was famously very skeptical of writing because he thought it would harm your memory, and it did. We don't remember things quite as well as they did back in the day because they had to remember long epic poems to entertain each other. But I think writing is a worthwhile trade for having a slightly worse memory. And I think something similar is going on with AI, where yeah, you may be slightly less engaged in certain tasks, but if you use it right, you're going to be way more engaged in other tasks where you have much more power. And so you can construct a study that says brain connectivity goes down when you use AI, in the same way that you could construct a study that says people's memory is worse when they have writing skills. But I don't think anyone would want to go back to a world where no one was literate.

这太有趣了。有很多研究显示 AI 对学生的好处，比如在尼日利亚的研究，以及人们进步的速度。所以我认为你分享的这个背景非常重要，你会失去一些东西，但收获——希望是收获更大，而且到目前为止似乎确实如此。

That is super interesting. There's all these studies that are showing the benefits of AI to students, with these studies in Nigeria and just how fast people progress. So I think it's really important this context you're sharing, that you will lose some things but the gain—the hope is the gain is much higher, and so far it seems like it will be.

是的。我认为人们总是，尤其是在技术炒作周期或革命性范式转变的开始，很容易低估事情变化的速度。我经常用的例子是，我住在布鲁克林，我家街对面的裁缝不接受信用卡。信用卡已经存在很长时间了。所以即使是最好的情况，这类技术的采用也需要很长时间。而且我认为很容易低估人类知道如何应对的具体情境有多复杂。仅仅因为你能在测试中取得非常好的分数——这很不可思议，我爱 AI，它太不可思议了——但它并不能真正让你直观感受到，要取代你实际工作中或活动中的特定部分有多难。

Yeah. I think people always, especially at the beginning of a tech hype cycle or a revolution paradigm shift, it's always easy to underestimate how quickly things are going to change. And the example I always use is I live in Brooklyn and the tailor down the street from me doesn't accept credit cards. Credit cards have been around for a long time. So it takes a long time for technology like this to be adopted even in the best case. And I think it's really easy to underestimate how complex specific contexts are that humans know how to deal with. And just because you can get a really good score on a test, it's incredible—I love AI, it's so incredible—but it doesn't actually give you an intuition for how difficult it is to actually be replacing specific parts of work or activities that you would do.

我认为一个能让你有点直观感受的好例子是，大约一个月前，我花了一个周末做了个东西，叫“03 能预测我在会议上要说什么吗？”这是一个基准——CEO 基准。我这么做的原因是，OpenAI 测试模型强大程度的标准是，他们在内部代码库上测试。他们会问：“新模型在预测我们内部代码库中接下来会发生什么方面有多好？”因为这不在互联网上。所以这是一个很好的基准。于是我想：“嗯，我的会议记录也不在互联网上。我说的话很多都在网上，有些重叠，但这会很有趣。”所以我用这个基准测试了一系列前沿模型，就用我的 Granola 记录，结果它们表现很差。它们确实很差，但这并不是因为它们不聪明。

I think a really good thing to give you a little bit of an intuition for it is I built this thing over a weekend like a month ago that was 'Can 03 predict what I'm going to say in a meeting?' It's a benchmark—it's the CEO benchmark. And the reason I did that is because OpenAI's gold standard for testing how powerful a model is is they test it on their internal codebase. So they say, 'How good is the new model at predicting what comes next in our internal codebase?' because that's not anywhere out on the internet. So it's a really good benchmark for that. And so I was like, 'Well, my meeting transcripts aren't anywhere on the internet. A lot of what I say is on the internet, and there's some overlap, but it'd be kind of interesting.' And so I ran a bunch of the frontier models on this, on just my Granola transcripts, and they're pretty bad. They are pretty bad, and it's not because they're not smart.

现在有一股真正的推动力。Spotify 的 Toby 创造了“上下文工程”这个词，意思是在正确的时间把正确的上下文给模型，这至少占性能的一半。我认为这百分之百正确。这是我大约三年来一直在写的东西。当时我称之为“知识编排”。我认为上下文工程可能是个更好的术语，但这完全正确，而且这是一个非常非常难解决的问题。这不是一个一次性的问题，比如一个巨大的上下文包就完了。它会随着时间的推移变得更好，但一旦它擅长预测我在会议上接下来要说什么，我就会把它当作工具使用，这会改变我在会议上接下来要说的整个动态。所以这并不像看起来那么容易。

There's this real push now. Toby from Spotify coined this term 'context engineering,' which is like getting the context to the model—the right context at the right time—is at least half the performance. And I think that's 100% true. It's something that I've been writing about for like 3 years. At the time I called it 'knowledge orchestration.' I think context engineering is probably a better term, but it's totally true, and it's a very, very hard problem to solve. It's not just like a one-shot problem where it's like, you know, gigantic context bundle and we're done. It's going to get better over time, but the minute it gets good at predicting what I'm going to say next in a meeting, I'm just going to use it as a tool, and that's going to change the entire dynamic of what I say next in a meeting. So it's not as easy as it seems.

有趣。我想你可以从中构建一个 GPT，然后现在不用和 Dan 开会，直接和这个东西对话，它会做决定。

Interesting. I imagine you can build a GPT from that, and then instead of having a meeting with Dan now, just talk to this thing and he'll make decisions.

当然。我的意思是，我们确实做了一点。这和能准确预测我在会议上要说什么不一样，但我认为如果你是 CEO、创始人或经理，你会惊讶于你的工作中有多少是重复自己。这是这次 AI 革命最好的事情之一，就是你不需要重复自己。所以上个季度我们就是这样。我倾向于设定一两个季度目标，上个季度我们的一大目标就是“不要重复自己”。所以如果我能避免，我不想在会议上说同样的话两次。

Definitely. And I mean, we do this a little bit. It's not the same as being able to predict exactly what I'm going to say in a meeting, but I think if you're a CEO or founder or manager, it's really stunning how much of your job is just repeating yourself. And that is one of the best things about this particular AI revolution is that you don't have to repeat yourself. And so we had it like last quarter. I tend to set like one or two quarterly goals, and one of my big goals for us last quarter was 'don't repeat yourself.' So I don't want to ever say the same thing in a meeting twice if I can help it.

所以对我们来说，在每份（简报）的重要部分，我们有一份每日简报，我花很多时间对标题提供反馈，或者对如何写引言提供反馈，或者“这个想法是什么？好吗？”这类事情。我们已经开始把所有这些编码成提示词，基本上——这和模仿我不同。它不能准确说出我在会议上要说什么，但它把我的品味推到了边缘，这样那些无法和我交谈的写作者，在我看到之前，他们已经和某种“我的模拟的模拟”交谈过了。这非常强大。

So for us, at every one of the big parts of every [newsletter], we have a daily newsletter, and I'm spending a lot of time giving feedback on headlines or giving feedback on how do you write an intro or like, 'What is this idea? Any good?' That kind of stuff. And we've started to codify all that into prompts that basically—it's not the same as mimicking me. It can't exactly say exactly what I'm going to say in a meeting, but it pushes my taste out to the edge so that writers who are not able to talk to me, by the time I see it, they've already talked to like some simulation of a simulation of me. And that's incredibly powerful.

让我们顺着这条线继续。这正是我想去的地方。

Let's follow this thread. This is exactly where I wanted to go.

我觉得你正在打造的业务、团队和运营方式，正是 AI 时代公司运营方式的先锋。你们在努力做到超级 AI 优先，这与你的写作高度一致。有很多理由去研究你们在做的事情。

I feel like the business you're building, the team you're building, the way you're operating is the very bleeding edge of how companies will operate and are trying to operate in this AI era. You guys are trying to be super AI first. It's super aligned with so much of your writing. There's just so much reason to study what you guys are doing.

是的，这对我们所有人都有好处。所以，谢谢。

Yes. And this is benefiting all of us. So, thank you.

那么，首先，告诉大家 Every 到底是什么，然后分享一些你们运营方式的见解。你笑了，真有趣，但不管怎样。

So, first of all, just tell people what the heck Every is and then share a few insights into just how you operate. It's funny that you laugh, but whatever.

每个人都这么问，因为这家公司的形态非常奇特。你其实可以看到早期时代有这种形态的公司，但它们不太常见。这种形态不太合理，我认为是 AI 新近才让它成为可能，我们可以聊聊为什么。但我通常这样描述 Every：我们在 AI 前沿做创意和应用。业务核心是一份每日通讯。我们已经做了大约 5 年，有大约 10 万订阅者。顶级 AI 实验室的人都读我们的通讯。任何对前沿 AI 感兴趣或从事相关工作、想知道最新动态的人都会读。我们做了很多事，比如每当 OpenAI 或 Anthropic 发布新模型，我们都会提前拿到，然后试用并写文章，这是我最理想的工作。我爱死它了。这是最好的。我不知道在这个播客里能不能说脏话，但这是完美的绝佳用途。

Everyone asked that because it's just a very weird shape of a company. You can actually see other companies that have this shape from earlier eras, but they're a little bit less common. It doesn't make as much sense, and I think it's newly enabled by AI, and we can talk about why. But the way that I typically talk about Every is we do ideas and apps at the edge of AI. So the core of the business is we have a daily newsletter. We've been doing it for about 5 years. We have about 100,000 subscribers. All the people from the top AI labs read us. Anyone who's basically interested in or working in AI at the frontier and wants to know what's going on reads us. We do a lot of, for example, whenever OpenAI or Anthropic drop a new model, we get our hands on it early, and then we get to play with it and write about it, which is my ideal job. I love it. It's the best. I don't know if I can curse on this podcast, but it's the perfect excellent use.

你们称之为“氛围检查”。这是……

And you call those vibe checks. Is that the...

是的，我们称之为“氛围检查”。

Yeah, we call them vibe checks.

我认为这非常重要，因为这涉及到我们工作的下一部分，也就是应用部分。我认为做“氛围检查”并称之为“氛围检查”非常重要，因为它们关乎使用这个东西的感觉，以及用它来做你通常会做的工作或生活中的事情的感觉。因为我认为这捕捉到了标准基准无法捕捉、也确实无法捕捉的东西。而写“氛围检查”的最佳人选是那些真正处于前沿、用它做事的人。

Which I think is really important because this gets to the next part, the apps part of what we do. I think it's really important to do vibe checks and to call them vibe checks because they're about how does it feel to use this thing, and how does it feel to use it for work for things that you would normally use it for, like in your job or in your life. Because I think that captures something that standard benchmarks just don't capture and really can't. And the best people to write a vibe check are people that are actually at the edge using it for stuff.

所以随着时间的推移，我们发现，我们认为最好的技术写作和内容来自那些真正使用技术并用它构建的人。所以我们一直有这样的功能：除了写作，我们总是在构建小实验。这帮助我们写出很棒的内容。这已经变成了我们内部运行的一套应用。构建这些应用的人也是作者，他们也为“氛围检查”等做出贡献。所以你能从每天实际使用这些技术的人那里，真正深入了解这些东西是如何构建的。我们有一套应用：一个叫 Kora。我们在录制这一天公开发布了 Kora，这真的很棒。

And so what we found over time is we have, we love, we think the best writing and content about technology is from people that are actually using it and building with it. And so we've always had this sort of function where we're always building little experiments in addition to our writing. And that helps us write great stuff. And that has turned into a suite of apps that we run internally. And the people who are building those apps are also writers and they're contributing to things like vibe checks. So you get a really inside look into how is this stuff being built from people who are actually using it every day. And the suite of apps that we have: one's called Kora. We just launched Kora publicly on the day that we're recording this, which is really awesome.

恭喜。

Congratulations.

谢谢。你可以把它想象成一位参谋长，一个 AI 邮件参谋长。它帮你用 AI 管理邮件。非常酷。我们稍后可以深入聊聊。我们还有一个叫 Sparkle，是一个 AI 文件清理器。我们还有一个叫 Spiral，用 AI 做内容自动化。我们最初孵化了 Lex，一个 AI 文档写作工具，我们把它分拆成独立公司，我的 Every 联合创始人 Nathan 在运营。基本上我们把所有东西捆绑在一起。所以你付一个价格，就能使用我们做的所有软件，而且我们不断往捆绑包里加新东西。我可以告诉你更多关于我们喜欢孵化什么样的东西以及如何孵化，因为我觉得里面有一些非常有趣的特殊东西。但我已经说了很久了，所以先到这里。

Thank you. You can think of it like a chief of staff, an AI chief of staff for your email. It helps you manage your email with AI. It's very cool. We can go into more of it later. We have another one called Sparkle, which is an AI file cleaner. We have another one called Spiral that does content automation with AI. We originally incubated Lex, which is an AI document writer, which we spun out into its own company, and my Every co-founder Nathan runs that. And basically we bundle everything together. So you pay one price and you get access to all of the software that we make, and we're constantly putting new stuff in the bundle. And I can tell you more about what kinds of things we like to incubate and how we like to incubate it, because I think there's some really interesting special things in there. But I've been blabbing for a while, so I'll stop there.

还有咨询公司，我想谈谈，但先放一放。

There's also consulting firm which I want to talk about, but let's hold off on that.

我们有咨询业务。我们也做这个。这是业务的第三条腿。它不太符合我的“创意和应用”框架，但我们花很多时间与大公司合作，教他们如何做到 AI 优先。我们培训所有人如何使用 AI，这非常酷。真的很有趣，也是我们工作的重要组成部分。

We have consulting. We also do that. And that is another, that's like the third leg of the stool in the business. It doesn't fit quite as nicely into my ideas and app streaming, but we spend a lot of time with big companies where we teach them how to basically be AI first. We train all the people on how to use AI, and it's very cool. It's really fun and a very important part of what we do.

这感觉就像个十亿美元的业务。我想回头再谈。我觉得是这样，因为每个人都想学这个。好的。那么，分享一些你们运营的方式。你提到你的团队不写任何代码。有哪些方式让你们能如此高效地运营？我知道你的团队非常小。你们有每日通讯，有三四个产品，有咨询部门。整个团队有多大？

That feels like a billion dollar business right there. I want to come back to it. I think so because everybody wants to learn this. Okay. So, share a few ways that you guys operate. You mentioned that your team doesn't write any code. What are just some ways that allow you to operate this efficiently? I know your team's really small. You have daily newsletter. You have three or four products. You have a consulting arm. How big is the team of everything?

我们有 15 个人。

We have 15 people.

15 个人。好的。那么，给我们一些关于你们运营方式的见解，这些方式有点前沿。

15 people. Okay. So, just give us insight into some of the ways you operate that are kind of at the bleeding edge.

好的。有几件事。第一，我认为每个人都应该这样做，我们有一个 AI 运营主管。我每周和她坐在一起一次，每次我做重复性的事情时，我就会说，我们把它放到待办事项里，她就会不断构建提示词、工作流之类的，这样我和团队其他成员就能尽可能多地自动化。我认为这是一个巨大的解锁，因为如果你整天工作，忙于救火，你会想：“好吧，我是用我知道的方式做，还是用可能行不通的新方式做？比如，我要花很多时间在 Zapier 上构建无代码自动化，我不想那样做。”而有一个 AI 运营主管，你就能识别这些事情，并让它们得到解决，而不用让做实际工作的人花时间去做，我认为这大大提高了实现的可能性。这里总是有个技巧，你必须确保它被使用。所以基本上你是在内部开发小应用。但如果你擅长开发人们会用的应用，那就太好了。强烈推荐有一个 AI 运营主管。

Okay. So, a couple things. One, and I think everyone should do this, is we have an AI head of AI operations. I sit with her once a week, and every time I'm doing something repetitively, I'm like, we put it in a to-do list, and she's just constantly building prompts and building workflows and stuff like that so that I and everyone else on the team are just automating as much as possible. And I think that has been a big unlock because it's really hard if you're working in a job all day, you're fighting fires, and you're like, "Okay, am I going to do this in the way that I know how or am I going to do it in the new way that might not work? Like, I'm going to spend a bunch of time in Zapier building some no-code automation. I don't want to do that." And having an AI operations lead lets you basically identify those things and have them solved without people who are doing the work actually having to take time to do it, which I think makes it much more likely it happens. There's always a trick with that where you have to make sure it gets used. So basically you're developing little applications internally. But if you're good at making applications people use, it's great. Highly recommend having an AI operations lead.

我想你看到了 CF Kora 在推特上说要招这样的人。

I imagine you saw the CF Kora tweeted about this wanting to hire exactly this sort of person.

是的。

Yeah.

所以这显然是个趋势。那么，根据你的观点，这个人需要是公司日常工作之外的人，专门专注于帮助团队用 AI 提高效率？

So clearly this is a trend. So the idea is, per your point, that this needs to be somebody who's outside of the day-to-day work of the company and is specifically focused on helping the team be more efficient with AI?

是的。

Yeah.

那这个人主要是帮你自动化，还是也能帮其他人？

And then is this person mostly just you automating you or can they help other people?

不，她基本上能帮所有人。我们现在从编辑运营说起。编辑运营里有大量工作，我和主编 Kate 经常要做些小的文字修改，确保所有内容都符合风格，这每天要花好几个小时。现在 Opus 已经能做到，你给它一份风格指南和提示词，它就能通读你写的任何内容并做文字编辑，这太棒了。但关键在于，不只是构建这个系统，你还得让 Kate 养成习惯，别人给她东西时她会问“你用过这个提示词了吗？”。所以这里还需要一点行为上的改变，我觉得这是个很有意思的组织挑战。对我们来说会容易些，因为组织里每个人都把 AI 放在第一位，都很想用。我们没有人会说“我不知道，我不太想用这个”，那是另一大挑战，很多组织都会面临，但总是存在让人们去使用它的问题。

No, she helps she helps everyone basically. Okay, where we're starting right now is with the editorial operation. So, there's so much stuff in the editorial operation where I or our editor and chief Kate like Kate is constantly doing like little small copy edits to make sure everything is like in every style and it takes like hours hours a day. Um, and so now Opus is at a point where you can give it a style guide and a prompt and it'll go through uh go through anything you're writing and copy edit it, which is amazing. Um, the trick is it's not just building that. You also have to get Kate to be like, "Did you put this through the prompt yet?" Um, anytime someone gives her something. So, there's a little bit of like behavioral update too that has to happen, which I think is a really interesting organizational challenge. And I think for us it's a little easier because everybody inside the org is like very AI first and just like wants to go do it. Um we don't have anyone really who's like I don't know I don't really want to do this and and that's that's a whole that's a whole different challenge which I think a lot of organizations face but there's always a problem of getting people to use it.

这太酷了。这位 AI 运营人员的背景是什么？

That is super cool. What is her background this AI operations person?

她叫 Katie Parrot。她其实为我们做了很多代笔写作。当 Every 内部的创作者们通常自己写，但有时他们需要帮助，她就会帮他们写关于他们正在做的事情的内容。她就是这样开始和我们合作的。她现在还在做这个，但也花很多时间做 AI 运营的事情。在那之前，她在 Animals 工作，那是一家顶级的内容营销机构，他们非常注重流程。我觉得 Katie 之所以这么出色，是因为她非常擅长处理这类流程性的事情，或者说思考这些。但她也是个很棒的写作者，而且她对 AI 极其兴奋。她就是喜欢捣鼓、喜欢用，这让我觉得，好吧，你应该直接来做这个，而不是只做代笔。我们应该把这个加到你的职责里。结果非常棒。所以我觉得，至少你确实需要那种“我想捣鼓、我想构建”的人。也有人更有流程导向，我觉得那也很重要。而且如果他们理解他们试图构建的东西的技艺，那也大有帮助。

She her name is Katie Parrot. Um she does a lot she actually does a lot of um ghost writing for us. So she also when um when people inside of every who are builders um often they just write themselves but like sometimes they want help and she'll help um help them write about like whatever whatever they're working on. So that's that's how she started with us. She still does that but she also spends a lot of time doing the AI operations stuff. Um and then before that she was she worked at Animals which is a content marketing agency like one of the top content marketing agencies and they're very processoriented. And I think the reason Katie is so good is because she's she's incredibly good at at that kind of process stuff or like thinking about that. Um, but she's also a great writer and she's also um just incredibly uh excited about AI. She just like wants to tinker and wants to use it and like that was a thing that got me to be like, okay, you should just come and do that instead of just ghost writing. We should add this to your plate. And it's it's been really fantastic. So I think that's a at minimum you really just want someone who's just like I want to tinker. I want to build stuff. Um there's also people who have a little bit more of that process orientation. I think that is important. Um and to the extent they understand the craft of the thing that they're trying to build for that also helps a lot.

这是个很棒的建议。我觉得大家都会开始雇这样的人。

This is an amazing tip. I feel like everyone's going to start hiring these people.

我觉得是这样。还有几个人也谈过这个。我听说 Rachel Woods，她也是那种对 AI 思考很多的人。她说这正在成为一种趋势，我觉得这真的很重要，而且它会渗透到组织的每个角落。比如我们在编辑部门内部做这个，但 Kora（顺便说一句，Kora 拼写是 C O R A，所以和 q o 不同，有点容易混淆）上有很多文案，还有 spiral 或 sparkle 上的，我们都希望达到同样的质量标准。所以你会看到工程师给 Kate 发 Figma 文件，说“你能做下文字编辑吗”，这对大家都不好，而且 Kate 只有一个人，真的很难做到。所以我们做了一件事，Nateesh，Kora 的一位程序员工程师，构建了一个 Claude Code 命令，就用那个提示词检查整个代码库，找出所有需要文字编辑的地方，然后在 GitHub 上创建一个拉取请求，发给 Kate。她只需要看拉取请求，判断“这合理吗？”。所以你可以把那个提示词转换成工程师能用的格式，突然你的工程团队就能按你想要的风格写营销文案了。我觉得这太酷了。

I I think so. There's there's a couple other people who talk about this. So I heard Rachel Woods who's another um sort of she thinks a lot about AI stuff. she she's talking about I think it's becoming like it's becoming a thing and and I think it's I think it's really important and and it just like bleeds out into every other part of the org. So like we're doing this inside of the editorial or but there's a lot of copy that goes out on Kora and by the way Kora is spelled C O R A so it's different from q o um slightly confusing there's a lot of copy that goes out in Kora or spiral or sparkle that we want to have that same every quality bar for and so we have you know engineers sending Kate like here's the Figma file like can you go and like do copy edits and that sucks for everybody and Kate is one person and it's just really hard to to do that. So one thing that we did um Nateesh who's one of the programmers uh engine engineers on Kora built a Claude Code command that just uses that prompt and checks through the entire codebase um for for all the copy edits and then creates a pull request on GitHub and then sends the pull request to Kate. So she's just like looking at the pull request and being like does this make sense? And so you can translate that prompt into, for example, a format that engineers can use and suddenly your engineering team is writing marketing copy in the style you want. I think that's so cool.

这太酷了。我想稍微岔开一下话题。你一直提到 Claude，我很好奇，你和你的团队最终使用的工具栈里都有什么？看起来 Claude 是核心部分。

That is extremely cool. Uh I want to take I'm going to take us on a little tangent. You keep mentioning Claude and I'm I'm curious just what is kind of in the stack of tools that you find yourself using that your team ends up using. It seems like Claude is a core part of it.

我确实喜欢 Claude。不过一般来说，我打开的第一个东西是 03。我算是个 ChatGPT 男孩。我觉得 03 质量超高，写作、编程、各种事情都很棒。它和 Claude 相比真正有区别的是它有记忆，我特别喜欢这点。我花了很多时间跟 ChatGPT 强调“我的写作要简洁有力”，现在它已经知道了。所以当我让它写东西时，它实际上比一般用户写得更好。而且我经常用它做自我反思和个人成长之类的事情。它了解我，所以当我发会议记录给它，问“我表现得怎么样？”，它会说“你做了你常做的那件事，但另一件事你进步了很多”，我喜欢这样。我觉得这真的很棒。

I do love Claude. I would say I'm generally my first thing that I open is 03. I'm like a chatbt boy. Um and I think 03 is super high quality. I think um it's great for writing. It's great for coding. It's great for all that stuff. And what it has that really makes a difference still from from Claude is it has memory and I just love that. Like I've spent so much time yelling at Chacht about like I need my writing to be punchy and concise, you know, and it just knows that now. So I think when I ask it to write something for me, it's like actually better than yours or maybe not yours but like you your average your average CHBT user and I also find like I use it a lot for self-reflection and personal growth type stuff. So it knows me so when I send it a meeting trans I'm like how did I do? It's like well you did that thing that you normally do but you're way better on this other thing and I I like that. I think that's I think that's really great.

所以日常就是 03，那是我的首选。我觉得 Claude Opus 首先，Claude Code，Every 内部每个人基本上都在用，如果你在构建什么，你就在用 Claude Code，它好得离谱。Gemini 刚出了个东西，我很兴奋想试试，因为我觉得那是我们构建应用时用得最多的模型，在应用内部它非常强大，而且非常便宜，这很棒，所以我想试试他们出的 CLI 工具。我们也用一点 Codex，那是 OpenAI 的编程工具，用于那种一次性的、独立的，比如我想挑出这个小功能。我还用什么？回到 Claude，Claude Opus 4 能做到其他模型做不到的事，除了另一个我不能说的模型。

So day-to-day 03 that's my that's my go-to. I think claude opus is first of all Claude Code everyone inside every that's basically what we use um if you're building something you're using Claude Code it it's crazy it's so good um Gemini just came out with something so I'm very excited to try that um because I think that that's the model that we use most for the apps that we build like inside the apps u it's incredibly powerful and it's incredibly cheap which is great so I want to try the CLI tool they came out with we also use codeex a bit um which is OpenAI's coding tool. And that's for like I want a one-off self-contained like I want to pick off this little feature. What else do I use? Uh going back to Claude, Claude Opus 4 can do something that no other model except one other model that I can't talk about. Um

它能做到其他模型做不到的事。

it can do something that no other model can do.

我们就不谈那个了，不想让你惹上麻烦。好，继续说。但确实，没有其他模型能做到这个，那就是早期版本的 Claude，我觉得其他模型的版本一般也是，当你问它们“这段文字写得好吗？”，比如 Claude，它总是给 B+，然后如果你在同一对话中再来一轮，说“我更新了这个”，它总是会变成 A-。

We won't go there. We don't want to get you in trouble. Okay, go on. But yeah, no other model can do this, which is earlier versions of Claude and I think generally versions of other models when you ask them, is this piece of writing any good? Claude, for example, would always give it a B+ and then if you change if if you did another turn of the same conversation, you're like, I updated this, it would always go to A minus.

然后如果你再给它一次机会，它就会变成 A 之类的，你知道吗？所以它没有那种直觉。它有点像过度思考你想听什么。有各种方法可以通过提示工程来解决，比如给它一个模板之类的。这些方法有点用，但它就是没有那种能力——它能不能判断写作是否有趣或好不好？它有没有那种直觉？而 Opus 4 有。这真的很神奇。我认为这非常重要，因为它开启了所有这些用例，你可能想用语言模型作为评判者。比如对我们来说，我们正在开发产品 Spiral 的新版本，它做内容自动化。你以前用过这个。我们基本上是在做一个类似 Claude Code 但用于内容的产品，你说，我想写一条推文。你给它所有文档。它有很多记忆。它为自己创建待办事项列表，然后去写。其中一个有趣的点是，现在因为它能评判事物，它的待办事项列表的一部分是，我写了三条推文，我要判断它们好不好。然后它可以在返回给你之前改进。这是一个巨大的解锁，我们花了三个月试图构建这个疯狂的系统来让它评判写作，然后 Opus 4 一次就搞定了，我们觉得太好了，这个产品能用了。我们开始发布吧。所以，是的，我为此喜欢它。

And then if you give it another turn, it would go to like A, you know? So it doesn't have the same kind of gut. It's like it's sort of thinking about what you probably want to hear too much. And there are various methods that you can use to prompt engineer around this, like give it a template or whatever. And they sort of worked, but it just still doesn't have that thing where it's like can it tell if writing is interesting or any good? Does it have that gut sense? And Opus 4 has it. It's really wild. And I think that's super important because it opens up all these use cases where you might want to use a language model as a judge. So for us, for example, we're working on a new version of our product Spiral, which does content automations. You've used that in the past. And we're doing essentially a Claude Code but for content style product where you say, I want to write a tweet. You give it all the documents. It has a bunch of memories. It creates a to-do list for itself and then it goes and writes. And one of the things that is so interesting is now because it can judge things, part of its to-do list is okay, I wrote three tweets. I'm going to judge whether I think these are any good. And then it can improve before it comes back to you. And that's just a huge unlock that we were struggling for like three months to build this crazy system to try to get it to judge writing, and then Opus 4 just one shot at it and we're like great, this product works. Let's start shipping it. So yeah, I love it for that.

你还有其他经常使用的 AI 工具吗？你提到 Granola，甚至在外面也用。你觉得有哪些工具是人们可能忽视的？

Are there any other AI tools that you just use regularly? You mentioned Granola even outside of the bottles. So what are some that you think maybe people are sleeping on?

我用 Granola。我以前用 Super Whisper 和 Whisper Flow，我觉得它们很棒。我们有一个内部版本叫 Monologue，大约一个月后发布，我现在就用它，但你可以把它们看作大致等效。我认为语音转文本界面是未来，更多人应该使用它们，更多人应该把它们作为功能来构建。我一直用 Notion，我特别用它的会议录音。我觉得这基本上就是我的工具栈。

I use Granola. So I used to use Super Whisper and Whisper Flow, which I think are fantastic. We have an internal version of that called Monologue that will be shipping in like a month or so that I use now, but you can think of them as roughly equivalent. And I think generally speech-to-text interfaces are the future and more people should be using them and more people should be building them as affordances. I use Notion all the time, and I specifically use their meeting recording. I think that's mostly the stack.

好的，这真的很有帮助，也超级有趣。本期节目由 PostHog 赞助，这是一个你的工程师真正想用的产品平台。PostHog 拥有创始人、开发者和产品团队所需的所有工具，如产品分析、网络分析、会话回放、热图、实验、调查、LLM 可观测性、错误跟踪等等。PostHog 提供的所有功能都有慷慨的免费层级，每月重置。超过 90% 的客户免费使用 PostHog。你会喜欢与这样一个透明且技术精湛的团队合作。你会看到工程师为你的问题提交拉取请求，他们的支持团队在遇到棘手问题时提供代码级帮助。PostHog 让你所有数据集中在一个地方。除了分析事件，他们的数据仓库还能让你从 Postgres 数据库、Stripe、HubSpot、S3 等更多来源同步数据。最后，他们新的 AI 产品分析师 Max AI 帮助你更快取得进展。获得构建复杂查询和设置账户的帮助，专家随时待命。今天就免费注册 posthog.com/lenny，并告诉他们 Lenny 推荐的。那是 posthog.com/lenny。让我们回到你的团队运作方式。你提到有 Kate，是这个名字吗？

Okay, that was really helpful and super interesting. This episode is brought to you by PostHog, the product platform your engineers actually want to use. PostHog has all the tools that founders, developers, and product teams need like product analytics, web analytics, session replays, heat maps, experimentation, surveys, LLM observability, error tracking, and more. Everything PostHog offers comes with a generous free tier that resets every month. More than 90% of customers use PostHog for free. You are going to love working with a team this transparent and technical. You'll see engineers landing pull requests for your issues and their support team provides code-level assistance when things get tricky. PostHog lets you have all your data in one place. Beyond analytics events, their data warehouse enables you to sync data from your Postgres database, Stripe, HubSpot, S3, and many more sources. Finally, their new AI product analyst Max AI helps you get further faster. Get help building complex queries and setting up your account with an expert who's always standing by. Sign up today for free at posthog.com/lenny and make sure to tell them Lenny sent you. That's posthog.com/lenny. Let's go back to ways that your team operates. You mentioned having Kate, was that her name?

是的。

Yeah.

好的。还有什么？你们还做了什么，你认为其他公司应该做或最终会开始做的？

Okay. What else? What else do you do that you think other companies should be doing or will eventually start doing?

所以 Kora 团队，也就是 Kieran 和 Nateesh，基本上就是那个团队，两个人。嗯，是 Kora，是 Kieran、Nateesh 和 15 个 Claude Code 实例。所以它比你想象的更强大。

So the Kora team, which is Kieran and Nateesh, basically that's the team, two people. Well, it's Kora, it's Kieran, Nateesh, and 15 Claude Code instances. So it's more powerful than you think.

我喜欢这只是再次瞥见未来。我们做的一件事我觉得很酷，他们基本上发明了这个，我与此无关，他们发明了“复合工程”的概念。基本上，每完成一个工作单元，你应该让下一个工作单元更容易完成。举个例子，在 Claude Code 的世界里，你不怎么编码，你最终会花很多时间写 PRD。比如，这里有一份文档，里面正是我需要做的事情，对吧？所以你可以说，“好的，酷。那现在就是我的工作了。我就写 PRD。” 然后每个连续的 PRD，工作量都一样。或者你可以花点时间想，PRD 有一个理想形态，我要做的是写一个提示词，能把我杂乱的想法变成 PRD。所以你花一点工作，让所有接下来的 PRD 更容易写，因为你写的更少了。找到那些小加速点，每次你构建东西时，让下次做同样的事情更容易，我认为这能给你的工程团队带来更多杠杆。所以，是的，我们有 Kieran 和 Natesh，Kora 刚刚从中诞生，刚刚公开。它之前是私有测试版，有 2500 个活跃用户，有数百万封邮件通过它处理，这是我们 15 人公司做的产品之一。这有点疯狂。

This is I love that this is just again a glimpse into the future. One of the things that we do that I think is really cool and they basically invented this, like I had nothing to do with this, is they invented the idea of compounding engineering. So basically for every unit of work you should make the next unit of work easier to do. So an example is in a Claude Code world where you're not coding a lot, you end up spending a lot of time essentially typing PRDs. Like here's a document with exactly the stuff that I need to do, right? And so you could just be like, "Okay, cool. That's my job now. I'm going to just write PRDs." And so each successive PRD, it's the same amount of work. Or you could spend a little bit of time being like, there's a sort of platonic ideal of a PRD, and what I'm going to do is write a prompt that can take my rambling thoughts and then turn that into a PRD. And so you spend a little bit of work to make all of the next PRDs that you're doing easier to write because you're writing less of them. And so finding those little speedups where every time you're building something you're making it easier to do that same thing next time, I think gets you a lot more leverage in your engineering team. And so like yeah, we have Kieran and Natesh, and Kora just came out of it, it just became public. It was in private beta, has 2500 active users, and there's like millions of emails going through it, and that's one of the products that we do as a 15-person company. It's kind of crazy.

这确实疯狂。你们是怎么做这种加速的？是不断优化的提示词吗？

It is crazy. How do you do this speed-up thing? Is it prompts that they continue to refine?

很多是提示词和自动化之类的。是的。

A lot of it is prompts and automations and stuff like that. Yeah.

明白了。对于自动化，用什么工具？用什么工具来自动化自动化？

Got it. For automations, what's the tool? What's the tool used for automating automations?

他们大量使用的是 Claude Code。你可以在 Claude Code 中使用斜杠命令，这些就像你重复使用的提示词。

What they're using a lot of is Claude Code. So you can do slash commands in Claude Code, which are like repeated prompts that you're doing.

明白了。好的。所以基本上他们在构建一个提示词库，让从“这是我想构建的东西”到“一个可以输入 Claude Code 的扎实 PRD”的过程更正确、更高效。

Got it. Okay. So basically they're building a library of prompts that make the process of here's what I want to build to a good solid PRD that you can feed into Claude Code. Yeah. More correct and more efficient.

正是如此。

Exactly.

超级有趣。他们只是保留一个文件，还是把它放到项目里？他们是怎么存储的？

Super interesting. And they just keep like a file or they put this into a project. Is that how they store?

是 GitHub。就在他们的 GitHub 里，他们可以互相分享。他们做的另一件我觉得很酷的事是，他们同时使用多个 Claude 实例，但也在使用其他三个智能体。他们喜欢一个叫 Friday 的智能体。

It's a GitHub. It's like in their GitHub where they can share it with each other. Another thing that they do which I think is very cool is they use a bunch of Claude instances at once, but then they're also using like three other agents. So they love, there's an agent called Friday that they love.

那是一个叫 Friday 的 AI 智能体产品。

That's like an AI agent product called Friday.

是的。

Yeah.

听说过。好的。

Heard of that. Okay.

嗯，还有一个叫 Charlie 的，他们真的很喜欢。特别是，我觉得他们喜欢 Charlie 的地方，我们有一个完整的视频讲这个，我可以发给你。

Um, there's another one called Charlie that they really love. And in particular, I think the thing they like about Charlie, we have a whole video about this which um I can send to you.

好，我会指出来。

Yeah, I'll point to it.

他们做了一个从 S 级到 F 级的 AI 智能体排名，我觉得特别搞笑。嗯，而且，我真的很喜欢 Charlie 的一点是它住在 GitHub 里。所以当你收到一个拉取请求时，你可以直接 @Charlie 说“你能看看这个吗？”嗯，而且让不同的智能体拥有稍微不同的视角，似乎效果很好。就像不同的人，你知道，有不同的视角和不同的品味。比如 Kieran，他是那种严肃的 Rails 爱好者，就是热爱 Rails，热爱 Rails 的感觉，所以我觉得他对“这个智能体，比如 Chukg，感觉非常简洁、极简、专业，有他可能喜欢的特定风格”有真正的敏感度，而 quad 可能是稍微不同的风格。我觉得所有这些都很有趣，这些东西有个性，而那会改变你可能想用它做什么，或者为什么你可能想同时用三个。

They did like a you know S tier through F tier of AI agents which I think is so funny. Um, and um, one of the things I really like about about Charlie is that it lives in GitHub. So you can when you get a when you get a pull request, you can just be like at Charlie like can you can you check this out? Um, and that seems to seems to work really well to have like different agents that have like maybe slightly different perspectives. It's like different people, you know, that have different perspectives and have different taste. Like you can I Kieran is he's like a one of those like ra like serious Rails files who are just they just love Rails and they love the way that Rails feels and so I think he has a real sensitivity to okay this agent you know Chukg for example it's very it feels very tur and minimal and and professional and so and it has a particular kind of style that maybe he likes versus I don't know quad is a slightly different style and I think that's I think all of that is so interesting that that these things have personalities and that those that that changes what you might want to use it for or why you might want to use three of them at once.

这太迷人了。呃，这让我又想起 Peter Deng 的对话，他在其中一个关键教训里谈到了他的招聘策略，他最终雇了像 ChatGPT 的现任产品负责人、现任市场负责人、现任工程负责人，因为他雇的是那些了不起的人，他的理念是雇一支复仇者联盟团队，每个人在某些方面很强，在一起就是完美的团队，而不是每个人都擅长一切。有趣的是，你总是可以用不同的产品、不同公司的不同智能体做到这一点。

That is so fascinating. Uh it makes me think about Peter Deng's conversation again where he talks about his hiring strategy in one of his key lessons and he ended up hiring like the current head of product for JT GPT the current head of marketing at JGPT the current head of engineering like because he hires the is incredible people and his philosophy is to hire a team of Avengers where everyone is strong at certain things and together they are the perfect team versus everyone versus like the best at everything and it's interesting that you can always do that with different product different agents from different companies.

你绝对可以。

You definitely can.

而且这让我觉得，潜在的市场可能比人们想象的要大，人们会想要不同公司的智能体，而不只是所有的 Devin 或所有的 Codex。

And it makes me feel like there's a bigger market than people think potentially where people will want different companies agents, not just all Devons or not all Codex.

我觉得确实如此。肯定不是一个智能体就能统治所有。所以

I think there really is. It's definitely not like one one agent to rule them all. So

有意思。

Interesting.

是的。

Yeah.

天哪。Kora 团队的那两个人是什么背景？他们都是工程师还是什么？

Oh my god. The two people on the Kora team are what's their background? Are they both engineers or what are they?

他们都是工程师。Kieran 的背景很疯狂，

They're both engineers. Kieran's got this like crazy background where

他们俩的背景都很有意思。

They both have really interesting backgrounds.

Kieran 的背景很疯狂，他之前在一家初创公司担任副总裁和工程负责人。所以实际上相当于一家或两家初创公司的 CTO。嗯，而且他是创始人之一，但在此之前，他是一名作曲家，职业作曲家，再之前他是面包师。所以我们去年在法国做团队静修，他教我们所有人怎么做牛角包。我的牛角包做得很难看，他的却很漂亮。嗯，

Karen's got this crazy background where he was previously like VP and ange at uh at a startup. So like was effectively like the CTO of a of a startup or maybe two startups. Um and uh and was was one of the founders and then but before that he was like a composer like a professional composer and before that he was a baker. So we did like a team retreat in France last year and he like taught us all how to make croissants. My croissant was horrible. His was like beautiful. Um,

看起来

Seems that

而且总的来说，我觉得那种多维度的人才正是我喜欢在每个团队里拥有的，因为我们都是通才。我们都想用 AI 做各种奇怪、很棒、有创意的事情，而有这种背景的人会有很好的品味，不仅对智能体，还对落地页应该长什么样等等，我觉得这在你想把一个 15 人的通才团队扩展到五个产品时越来越重要。所以这就是 Kieran 的背景。Natasha 的背景让我嫉妒，因为他是在 ChatGPT 出现后才开始学编程的。嗯，他一直想学编程，而且他只在 AI 时代会编程。我一直告诉他，老兄，我是中学时从书里学编程的。我得去 Barnes & Noble 买一本书。而且我什么都搜不到，没法谷歌搜索为什么这个函数不工作之类的。

and generally I think like that kind of multi-dimensional type of talent is the kind of person that I love having at every like because we're all generalists. We all want to use AI for all these like weird awesome creative things and someone who has that background is going to have a good taste for not only agents but what should the landing page look like or whatever which I think is increasingly important where you're trying to scale a team of generalists of 15 people to like five products. So that's Kieran's background. Natasha's background is I'm jealous because he only started learning to code when Chachu came out. Um he had wanted to learn to code forever and he's only known how to code in an AI era. And I keep telling him, dude, like I I learned to program in middle school from books. Like I had to go to Barnes & Noble and like buy a book. And there was nothing I couldn't Google anything about like how this how this why this function wasn't working.

那时候甚至还没有 Stack Overflow。

Tag overflow even back then.

是的。是的。那时候没有 Stack Overflow。只有一些奇怪的 BBNet 论坛之类的东西，我当时 12 岁，可能不应该上那些网站。所以，呃，他比我在前 AI 时代见过的任何工程师都进步得快得多，我在公司其他地方也看到同样的情况。我觉得有一个大问题：当入门级工作被 AI 夺走时，孩子们会怎样？我的看法是，这值得思考，有可能在某个时候会成为一个问题，但我的看法是，每当我看到一个用 ChatGPT 的孩子，我就会想：“天哪，他们会比我共事过的任何人都快得多。”比如我们有个叫 Alex Duffy 的家伙，他和我们合作。嗯，他为 Context Window 写文章，他刚刚推出了……我们教 AI 互相玩外交游戏。嗯，这真的很酷。他做了那整件事，我觉得他真的真的真的真的很有才华。当他来找我们时，大概差不多一年前了，那是经典案例之一，我在每个地方都反复看到：你有好点子，但你写作还不够好，在你足够好之前，我很难和你一起做事。所以我得给你一些小事做，直到你变得更好，等等。我注意到他的是，他就像在一年里……他在两个月里取得了一年的进步，因为每次我坐下来告诉他：“好，故事是这样讲的。标题是这样想的。”他就把所有这些记录下来，放进一个提示词里，而且他从不犯同样的错误两次。我觉得他因为这些东西，比原本的位置加速了很多。我在其他很多地方也看到这种情况。所以 Natasha 是另一个好例子。所以我觉得总的来说，人们会发现，一个 20 岁、有 ChatGPT 订阅的人，只要你指导他们，就会非常强大。我觉得那很棒。

Yeah. Yeah. There wasn't tag overflow. there's like weird BBNet forums and stuff that like I was like 12 and I probably shouldn't have been on there or whatever. So uh it's he has gone so much faster than any other engineer I think like in a preAI era and I see the same thing in the rest of the company like I think there's this huge question about um what happens when kids uh like entry- level jobs are taken away by AI and my take is like that that's worth thinking about and it's it's possible that that might be a problem at some point but my take is whenever I see a kid with CHBT, I'm like, "Holy they're going to go so so much faster than any other person that I've worked with, like we have this guy, Alex Duffy, who works with us. Um, he writes for Context Window and he he just launched um we taught AIS how to how to play diplomacy with each other. Um, which is really cool. And he did that whole thing and he's I think he's really really really really talented. And when he came to us like I guess almost a year ago now it was one of those classic cases which I've seen like over and over at every which is you have great ideas but you're not a good writer yet and it's really hard for me to do anything with you until you're good enough at it. So I have to give you like small little things until you get better and blah blah whatever. And what I noticed with him is he was just making a year like he made like a year's worth of progress in like two months because every time I sat down with him and told him, "Okay, here's how you tell a story. Here's how you think about a headline." Like he recorded all of it, put it into a prompt, and like he never made the same mistake twice. And I think he's so much accelerated from where he would have been because of this stuff. And I see that in lots of other parts of the or so. So Natasha is another good example. And so I think generally people are going to figure out that like some 20-year-old with CHBT subscription is like super powerful if you just like mentor them. And I think that's great.

天哪，这里有很多线索我可以展开。比如所有这些对入门级人员的恐惧，好像入门级岗位正在消失，那么如果这些人不能作为入门级人员学习做事，我们怎么会有资深人员呢？而你说的是 ChatGPT 和这些工具帮助你快速加速。所以你不需要在底层待很长时间。

Man, there's so many threads I could follow here. Like there's all this fear of entrylevel people will never like the roles are disappearing for entry level people and so how will we ever have senior people if these people can't learn to do things as an entry- level person? And what you're saying is chat GPT and these tools help you accelerate really quickly. So you don't really need to be at the bottom rung for a long time.

是的。你实际上是从一开始就在学习如何比入门级高一级。而且你必须……这有点像我整个“分配经济”论点，当你审视在 AI 时代哪些技能会有价值时。嗯，一大类技能是管理者的技能。

Yeah. You're effectively like learning how to be one level above um the entry level from the beginning. And you have to and this is sort of my my whole allocation economy thesis where when you look at what skills are going to be valuable in the AI era. Um one big group of skills are the skills of managers.

今天他们是人类管理者。明天每个人都是模型管理者。现在，管理技能并没有广泛分布，因为那非常昂贵。另一个昂贵的东西——所以 8% 的劳动力是管理者。现在管理会便宜得多，所以更多人将不得不去做这件事。所以这就是我现在看到的孩子们、20 岁左右的人，除了其他东西之外，将不得不开始学习的东西——你知道，你不能只是说“好的，去做吧”，然后回来。你得能够深入正在进行的工作，帮助它变得更好。但他们同时在学习两件事：学习如何管理，以及如何做实际工作，以便他们擅长这些。

Today they're human managers. Tomorrow everyone's a model manager. Right now, management skills are not broadly distributed because it's very expensive. Another expensive thing—so 8% of the workforce is managers. It's now going to be much cheaper to manage, so more people are going to have to do it. And so that's the thing that kids, 20-year-olds, whatever I see now, are going to start to have to learn in addition to—you know, it's not like you can just say, 'Okay, go do it,' and then come back. Like, you have to be able to go into the work that's being done and help make it better. But they're learning both at the same time. They're learning how to manage and how to do the actual work so that they're good at it.

而这里的管理是管理智能体，对吧？

And the managing here is managing agents, right?

是的。你在管理 AI。对。

Yeah. You're managing AI. Yeah.

所以这正好回到你的观点，关于这个核心团队——我猜你说每个人，每个——没人写代码，零代码编写。现在只是管理为你写代码的智能体。

And so this is a good coming back to your point about how this core team—and I guess you said everyone, every—no one writes code, zero code written. Now it's just managing agents that are writing code for you.

是的。好吧，我——我从未听说过处于这个阶段的公司。所以，这非常酷。所以，工作流程是他们给出“这是我想要的”。我使用他们构建的那个很酷的提示词库来完善它，然后智能体构建代码、编写代码，然后基本上时间花在审查代码和审查输出上。它看起来怎么样？感觉怎么样？然后继续完善。

Yeah. Okay, I don't—I've never heard of a company at this stage. So, this is extremely cool. So, the workflow is they give it 'here's what I want.' I refine it using this cool prompts library that they've built, and agents build code, write the code, then basically the time is spent reviewing code and then reviewing the output. What does it look like? What does it feel like? And then continuing to refine.

哇。所以，你们已经达到了 Cursor 的 Michael 所说的我们将要达到的状态。我们——我几个月前和他聊过。他说一年后他认为事情会发展到这个地步。我们不再看代码了。你们已经在那里了。虽然你们还在看代码。好吧。你们仍然在看代码。

Wow. So, you guys are at where Michael from Cursor said we will be. So we—I chatted with him a few months ago. He said in a year this is where he thinks things will be. We're not looking at code anymore. You guys are already there. Although you're looking at code. Okay. You're still looking at code.

他们确实在看代码。所以你知道，你在做任何事情之前都要进行代码审查。我确实认为，比如 Danny，他负责 Spiral，就是我提到的我们正在构建的云代码内容工具——你知道，他花了好几天深入研究我们感兴趣的某个第三方库的内部结构，只是因为了解这些是有帮助的。理解那些东西是有帮助的，但一旦他理解了，他实际上并不写任何代码。他只是告诉 Claude Code 该做什么。我认为那非常重要。我们正在达到一个疯狂里程碑。有一种感觉，我们正在到达一个你不需要真正理解代码的地方。你不需要写任何代码。我们会到达那里，而你们已经在那里了。我认为这太容易被忽视，这有多疯狂。你有一个完全不写代码的产品团队。

They definitely are looking at code. So you know, you're doing a code review before anything. And I do think like Danny, who runs Spiral, which is the Claude Code for content tool I was talking about that we're building—you know, he spent a couple of days digging into the internals of some third-party library that we were interested in, just because it's helpful to know. It's helpful to understand those things, but then he's not actually writing any code once he understands it. He's just off telling Claude Code what to do. And I think that's really important. This is an insane milestone we're hitting here. There's this sense we're getting to a place where you don't need to really understand code. You don't have to write any code. We'll get there, and you guys are there. I think it's so easy to overlook how wild this is. You have a product team not writing code at all.

这确实很疯狂。我认为尤其疯狂的是，有一小群人，每个人都有多维度的技能，每个人都有所有这些不同的技能。每个人都是通才。每个人都以 AI 为先。所以在这样的环境中，用一个小团队能做的事情是疯狂的。你正在发明所有这些新的原则，比如我们如何合作，我们如何做工程，诸如此类。我认为这就是让写作变得那样的原因——这就是我喜欢这样做的原因——因为我认为我们从中得到的写作非常好，因为我们可以从经验的角度来谈论它。但我确实想说另一件事，那就是我们还没有到那种地步，即 Every 的员工如果不会编码也能做他们现在做的事情。

It is really wild. I think it's really wild in particular just like having a small group of people that have everyone's multi-dimensional, everyone has all these different skills. Everyone's a generalist. Everyone's AI forward. So what you can do in an environment like that with a still small team is crazy. And you're kind of inventing all these new principles for like how do we work together, how do we do engineering, all that kind of stuff. And I think that's what makes the writing like that—that's why I like doing that—because the writing that we do from that I think is really good because we can talk about it from a sort of position of experience. But I do want to say something else, which is we're not at a point yet where the people that work at Every could do what they do if they didn't know how to code.

是的，这正是我想问的。这是一个不同的标准，我认为在很长一段时间内，知道如何编码仍然是有价值的。但这是一个并非全新的进展。例如，当我上中学学习编码时，新的热门事物是脚本语言，比如 Python 和 JavaScript。如果你是一个真正的程序员，你会理解 Python 和 JavaScript 背后的语言，那是用 C 写的。而脚本语言并不完全真实。为了真正做任何有趣的事情，你必须能够学习技术栈的两个部分。C 程序员也是如此。我想在 70 年代，C 被发明时，你必须能够编写汇编语言。而英语只是脚本语言之上的一层。所以，我认为所有这些事情都是正确的，因为在过渡期间，有很多理由说明为什么能够深入技术栈的一层很重要。随着时间的推移，这变得越来越不频繁，但那仍然需要很长时间。而且有时候，即使你是 JavaScript 或 Python 程序员，了解这些东西是如何工作的、如何编写的，以及它是如何实现的，也是有用的。今天它比过去重要得多，但那花了 10 或 20 年。我认为编程也会如此。拥有这项技能非常重要，会大大加速你的发展。随着时间的推移，它会开始变得不那么重要，但我们还没有接近那个点。

Yeah, this is what I was going to ask. Which is a different bar, and I think for a long time it's going to be valuable to know how to code. But this has been a progression that is not a new progression. So for example, when I was in middle school learning to code, the new hot thing was scripting languages like Python and JavaScript. And if you were a real programmer, you would understand the language underlying Python and JavaScript, which is written in C. And scripting languages weren't totally real. And in order to really do anything interesting, you had to be able to learn both parts of the stack. Same thing for C programmers. When, I guess in the 70s, C was invented, it was like you got to be able to write assembly. And English is just like a layer on top of scripting languages. So, I think all those things were right in the sense that there's—especially during transitions—a lot of reasons why it's important to be able to go down a layer in the stack. And it gets less and less frequent over time, but that still takes a long time. And there are sometimes when even if you're a JavaScript or Python programmer, it's useful to know how all that stuff works, how it's written, and see how it's implemented. Today it's much less important than it used to be, but that took like 10 or 20 years. And I think the same thing is going to be true for programming. Having that skill is super important and will accelerate you significantly. It will start to get less important over time, but we're not close to that yet.

好的，这是一个非常重要的观点。我很高兴你谈到了这一点。那么，你是否有感觉，距离你雇佣一个不是工程师的人来构建另一个产品，比如一个真正的 SaaS 产品，还有多远？因为——

Okay, that's a really important point. I'm glad you went there. So, do you have a sense of how far we might be from you hiring someone to build another product that isn't an engineer, like a real SaaS product? Because—

是的。所以，比如，“嘿，我们有个想法。我们想找个人来真正领导它”——非常遥远。甚至不在视野之内。但有很多东西可能是产品，比那低一层，我认为你几乎现在就能做。比如一个例子，我们谈到过 Dia，来自 Browser Company 的新 AI 浏览器。Dia 有这些叫做技能的东西，实际上就像你可以在浏览器中运行的小型 AI 应用。你可以提示它们，它们在网页上运行并为你工作。一个非技术人员可以构建那个。同样，像 ChatGPT 的自定义 GPT。非技术人员肯定可以构建那个。所以我认为，虽然我肯定会坚持认为，我们离任何人都能用零编程知识构建一个传统的 SaaS 应用（除了演示之外）还很远，但会有其他形式的软件。我的一个观点是，软件正在变成内容。

Yeah. So, like, hey, we have this idea. We want to bring someone on to actually lead it—very far. Like not even within sight. But there are a lot of things that could be products that are a layer a level down from that that I think you could do almost now. So like an example, we were talking about Dia, the new AI browser from the Browser Company. Dia has these things called skills, which are effectively like little AI apps that you can run in the browser. You can prompt them and they run on the web page and do work for you. A nontechnical person could build that. Same thing for like custom GPTs from ChatGPT. A nontechnical person can definitely build that. So I think while I will definitely maintain that we're not anywhere close to anybody being able to build a conventional SaaS app with zero programming knowledge aside from just a demo, there are going to be other forms of software. One of my things is like software is becoming content.

未来会出现其他形式的软件，它们看起来不像今天的软件，但即使你不懂编程、不是技术人员，也能创办并运营一家企业，而且这很快就会实现——我的意思是，这已经在发生了。只是它看起来不像你问的那种东西。就像好莱坞电影和 YouTube 视频之间的区别。

There's going to be other forms of software that don't look like the software of today, but you can run start and run as a business as a nontechnical person even if you don't know how to code and that'll happen very soon if I mean it's already kind of happening. It's just it doesn't look like the thing that you're asking about. It's like it's sort of like the difference between a Hollywood movie and like a YouTube video.

好的。我觉得这对很多人来说真的很让人安心。基本上，你看到的是 AI 为有技能的人赋能，让他们能做更多事情。

Okay. I think that's really reassuring to a lot of people. Basically, what you're seeing is AI just supercharges people who have a skill and allows them to do a lot more.

是的。

Yeah.

好的。你们运营中还有其他有趣的方式值得分享吗？这些方式能帮助你们快速运作、用更少的资源做更多事情？

Okay. Is there any other way that you guys operate that is really interesting that might be worth sharing that helps you operate really quickly, helps you do more with less?

我想聊聊我们如何思考产品构建。比如该做什么产品、最终会做出什么，因为我觉得这里面有些特别的东西，可能有一套对人有用的方法论。

I mean I would love to talk about our like how we think about building products. Um like what products to build like what do we end up building because I think that there's something sort of special about it that probably there's a playbook that is useful for people.

所以当我思考这个问题时，这其实是最近才变得清晰的。之前很多都是凭直觉做事，没有真正思考过。但当我回顾我们最终孵化出来的东西，基本上可以追溯到我在开头说过的话：有些东西在历史上非常昂贵，只有富人或大公司才能买得起。比如你的邮件的幕僚长。我认为心理咨询师或律师是另一个有趣的例子。比如有人帮你整理衣橱或整理电脑，或者有人帮你代笔。这些正在变得便宜几个数量级，所以每个人都能用得起，即使你在一个小型初创公司。

So when I think about this is this has only sort of snapped into focus recently. So a lot of this was just like doing it intuitively without really a thought for it. But when I think about the kind of things that we have ended up incubating, it's basically it goes back to something I said at the beginning which is there are these things that were historically really expensive um that only rich people or big companies could buy. So a chief of staff for your email. Um I think a therapist or like a lawyer is another interesting example. um uh someone to like organize your closet or organize your computer is another example, someone to ghostwrite for you. Um that are uh becoming orders of magnitude cheaper so that everyone can use them even if you're at a small startup.

所以基本上，当你运营一家像我们这样的 AI 优先公司时，你会遇到很多这样的小事，你会想：我现在真希望有个代笔。但代笔真的很贵。或者我希望有个律师，但那要花我 2.5 万美元。律师真的很贵，而且对这些服务的需求远远超过供给，因为它们太贵了。而 AI 能让你想：哦，我可以用 Claude 来做，我可以用 ChatGPT 来做。所以你能利用你现有的需求——比如我们请得起律师，我们有代笔，但还有很多事情我们做不了，因为我们负担不起。所以，我们仍然有律师和代笔，但我们能做更多这类事情。

And so basically like when you're running like we are sort of this AI first company, you're running into these all these little things where you're like I wish I had a ghostwriter right now. But ghost writers are really expensive. Or I wish I had a lawyer, but it wouldn't cost me like $25,000. Lawyers are really expensive and and there's a lot more demand for those services than can be fulfilled because they're so expensive. And what AI does is it allows you to be like, "Oh, I could just use Claude for that. I can use ChatGPT for that." Um, and so you're uh you're able to use the demand that you have that like we can we can afford a lawyer. we have ghost writers, but like there's a lot more that we can't do because we can't afford it. So, we still have our lawyer and we still have our ghost writers, but we just do a lot more of that stuff.

所以我们注意到，我们开始先用 ChatGPT 和 Claude 这些通用工具来尝试，看看这有没有用？这真的有效吗？诸如此类。如果有效，我们就会把它拆分成一个独立的东西，变成一个应用。我认为这个时代特别之处在于，整个棋盘在可构建的东西方面被彻底重置了。你知道，5 年前，你可能会去构建另一个笔记应用——我们一直在构建笔记应用，或者另一个 B2B SaaS 应用——都是同样的东西，只是包装略有不同。而现在，这是一个全新的领域，没人知道会发生什么，每个人都在边发生边发明。所有这些新工作流正在被创造出来，方式非常类似于——比如当电子表格刚出现在电脑上时，我们在电子表格上摸索这些新工作流，然后它们被拆分成 B2B SaaS。ChatGPT 和 Claude 也是一样。

And um, so we notice that we start to then use like ChatGPT and Claude first, these general purpose tools to try it and see is this useful? Does this actually work? All that kind of stuff. And then if it does, we will like unbundle it into its own separate thing that um becomes an app. And I think what's really special about this time is the entire game board has been like totally reset in terms of things you can build where you know 5 years ago it was like you're going to build another notes app like we've been building notes app for forever like another B2B SaaS app like it's all the same stuff like slightly different packaging and now it's like totally new territory no one knows what's going on no like everyone's inventing it as it as it happens right all these new workflows are being created in a very similar similar way to I don't know for example when spreadsheets were first a thing on computers like we were figuring out all these new workflows on spreadsheets they got unbundled into B2B SaaS same thing for ChatGPT and Claude.

真正酷的是，你可以说：酷，我用 ChatGPT 做这个，它对我真的很有用。你可能是最早真正注意到这一点的人之一。然后，因为 Every 的每个员工都是 AI 优先的，他们来到我们这里是因为他们读 Every，他们读 Every，所以我们都有同样的氛围，我们都在做类似的事情。他们成为我们的第一批用户。所以，我们衡量产品成功的标准是：它在 Every 内部是不是一个爆款。比如 Monologue，就是我跟你说的那个应用，每个人都开始用它，我们就觉得：好了，我们有点东西了。

And what's really cool is you can be like cool I'm using I'm using ChatGPT for this it's really useful for me and you might be like one of the first people to like really notice that um and then because everybody that works at Every is AI first and came to us because they read Every they read Every so they all have the we all have the same vibe and we're all kind of doing similar stuff. They become our first our first users. So, we measure the success of the product by like is it a banger inside of Every um like Monologue the the the app that I was talking to you about like everyone just started using it and we're like okay we've got something here.

然后真正有趣的是，如果 Every 内部每个人都用它，而读 Every 的人和我们也有类似的氛围，他们就会成为下一批用户。我认为这是一个非常有趣的构建应用或构建应用的管道。这是一个全新的蓝海，所以你想到的所有东西可能都是新的，这真的很酷。随着时间的推移，我认为像我们这样的组织，那些在边缘探索的人，我们在做的事情，3 年后其他人都会做。所以现在可能有点小众，但 3 年后当其他人都有和我们一样的需求时，这会是一件大事。

Um and what's what's really interesting then is if everyone inside of Every uses it and people read Every they have a similar vibe to us too. So they become the next set of users. And that's a really I think interesting like pipeline for building applications or building apps. It's a totally new like green field so that all the stuff you're thinking about like it's probably new which is really cool. And over time what I think is organizations like ours people who are playing at the edge we're doing things that in like 3 years everybody else is going to be doing. So it may be kind of niche for now but it will be a big deal in 3 years when everyone else has the same needs that we do.

那真的很酷。我听到的是，GPT 套壳是个好主意，值得构建。

That is really cool. Uh what I'm hearing is GPT wrappers are a good idea and are worth building.

我 100% 认为 GPT 套壳很棒，它们被大肆诋毁，完全没有理由，人们不明白它们有多有价值。

I 100% think GPT wrappers are amazing and they've been much maligned for absolutely no reason and um people don't understand how absolutely valuable they are.

我觉得还有一件事，你们刚刚完成了种子轮融资。我想，现在是聊聊这个的好时机，比如这些产品不必成为某个数十亿美元的爆款。是的，你们有这样一个公司组合，你们有内容业务。所以，我认为在“这些产品需要做多大才能成功”这个问题上，有一个非常有趣的方法。也许可以聊聊这个。

I think there's also just uh you guys are you raised the seed round. Uh I want to so this is a good time to maybe talk about that just like these products don't have to become some mega billion dollar hit. Yeah, you kind of have this portfolio of companies, you have the content business. So, I think there's a really interesting approach to the how big these need to get to be successful. Maybe just talk about that.

是的，我真的希望 Every 成为一个机构，教人们如何用技术，特别是 AI，过上更好、更有人性的生活，既通过写作和内容教他们怎么做，又为他们构建工具。但我认为，建立机构的基础，至少对我来说，我希望的方式是：我希望在内部感觉像一个创意游乐场，我们有机会去冒险、去做一些奇怪的事情，这些事毫无道理，我们无法向任何人证明其合理性，但我们只是觉得会很有趣。

Yeah, I really want Every to be an institution um that teaches people um how to live a better, more human life with technology, particularly with AI, and both like teaches them how to do it with writing um and the content we make and then builds tools for them to do that. And um but I think fundamental to building an institution is at least for me the way I would like to do it is um I want internally it to feel like this creative playground where we have the opportunity to like take risk and do stuff and do weird stuff that like just doesn't make any sense. We can't justify anyone but we just feel like it would be fun.

所以我认为我一直在玩味机构严肃性之间的动态张力。我们希望这能持久且重要，但它也应该只是好玩。让我们玩玩吧。我认为拥有这种张力真的很有价值。所以我一直有点犹豫要不要融很多钱，因为我觉得那会让你陷入必须成为那个完全全力以赴的严肃事物的境地。

Um and so I think I'm always playing with that dynamic tension between institution serious. We want this to be like lasting and important and it should just be fun. Like let's play around. And I think having that tension is like really valuable. And so I've always been like sort of hesitant to raise a lot of money because I think it locks you into like having to be that serious thing that's like totally going for it.

有很多公司都在摸索这种平衡，但就我个人作为创始人而言，我想保持选择的灵活性，也想保持那种好玩的感觉。我觉得部分原因是我知道自己或多或少能掌控想做的事。可能还有一些更深层的心理因素，如果你想聊的话我很乐意谈。但我觉得那也是我想要的。所以我们起步时，融了一轮非常小的 70 万美元 pre-seed 轮，那正是创作者经济最火的时候。我们俩差不多同时开始做 newsletter，那是最狂热、最疯狂的时候，钱到处乱扔。但我们只融了 70 万，因为我想融到足够让我们能实验、有点现金缓冲，但又不会多到把我们锁死在任何东西上。我们给所有投资人发了封邮件，说“你也是我们的投资人之一，所以你可能收到过这封邮件”。

And there's lots of companies that figure out that balance, but just for me personally as a founder, I want to keep the optionality alive and I want to keep the playful feeling alive. And I think part of that comes from knowing I have the control to do what I want more or less. There's probably also some deeper psychological things going on there, which I'm happy to talk about if you want to get into it. But I think there's also just that that's kind of what I want. So when we started, we raised a very small 700k pre-seed round, and this was at the height of the creator economy. So we both started our newsletters around the same time. It was the hypiest, craziest thing. People were throwing money around. It was wild. But we raised 700k because I wanted to raise enough for us to be able to experiment and have a little cash cushion, but not so much that it locks us into anything. And we sent an email to all of our investors being like, 'And you're one of our investors, so you've probably got this email.'

最最最小的投资人，但我算一个，算一个。

Tiniest tiny investor, but I'm in there. I'm in there.

我们给所有人发了封邮件，说“这可能不是一门风险投资生意，所以你们别指望我们再融资”。我们甚至用了一种稍微修改过的 SAFE 来融资，给了每个人在三年内转换为股权的选择权，即使我们不再融更多钱。所以我们这样做，既保留了做大做强、走传统路线的选项，也保留了按我们想要的方式去做的选项。也许它不是一门大生意，但我们热爱它。

We sent an email to everyone being like, 'This is probably not a venture business, so you should not expect us to raise again.' And we even raised on this slightly modified safe that gave everyone the option to convert to equity in three years even if we didn't raise more money. So we did it in a way that allowed us the option to get really big and do the traditional thing, and also the option to do it the way we want to do it. Maybe it's not a huge business, but we love it.

那太好了。

That's great.

最近这轮融资我们也做了同样的事，从 Reid Hoffman 和 Starting Line VC 融了最多 200 万美元。我们用的是我称之为“SIP 种子轮”的方式，基本上就是他们承诺了 200 万，但我们可以随时提取。我们只是按设定的上限用 SAFE 来做。对我来说，这真的很有帮助，因为它在心理上让我能承担更多风险。如果银行账户归零，我可以拿到更多钱，太好了，我不用去想它。但同样有帮助的是，我和团队其他人都不会盯着银行账户里那个巨大的数字，想着“太好了，我们可以烧钱，那就烧吧”。而且对我们的投资人来说，我觉得 Reid 非常希望我们成功，但我不觉得他在乎这生意有多大。我觉得他在哲学上更认同我们想做的事，如果它变成一门大生意，他会很兴奋。我一直在寻找的就是这种认同，因为我觉得这件事有一个核心的创造精神，我想保持它，而且我非常在乎产生大的影响。但我觉得产生影响的方式有很多，其中一种是建立一家 100 亿美元的公司。我觉得另一种方式是真正改变人们看待世界、看待自己在世界中的位置的方式。我觉得这就是故事的作用。你未必——有时你通过建立一家巨型公司来实现，但你未必总是要那样做。比如我们最关心的很多故事，来自那些可能根本不富有的人。所以我真的很喜欢创造这样一个地方，我们既能做一门真正的好生意，我非常在乎这一点，但它的灵魂核心是改变人们看待自己在世界中的位置的方式。

And we did the same thing for this recent round where we raised up to 2 million from Reid Hoffman and Starting Line VC. And we did it as what I've been calling a SIP seed round, which is basically they've committed $2 million, but we can pull it down whenever we want. And we just do it on a safe at a set cap. For me, that's really helpful because it allows me psychologically to take a lot more risk. If we go to zero on the bank account, I can get more money, great, I don't have to think about it. But what's also really helpful is that I'm not, and the rest of the team is not, staring at a gigantic number in the bank account being like, 'Cool, we can burn this, let's burn it.' And also for our investors, I think Reid very much wants us to succeed, but I don't think he cares what size of business this is. I think he's more philosophically aligned with the thing that we're trying to do, and if it becomes a huge business, he's psyched for it. And I think that kind of alignment is what I was looking for, because I think there's this core creative spirit to the thing that I want to maintain, and I really care about having a big impact. But I think there's a lot of ways to have an impact, and one of them is building a $10 billion business. I think another way is really changing how people see the world, see themselves in the world. And I think that's what stories do. And you don't necessarily—sometimes you do that by building a gigantic company, but you don't necessarily always have to do that. Like a lot of the stories that we care about most are from people who maybe they weren't rich at all. And so I really like creating this place where we can make a really good business, and I care a lot about that, but also the core of the soul of it is changing how people see themselves in the world.

我喜欢你创新了一种新的中间地带的融资方式，既不是白手起家，也不是普通的 VC。这是“六种子”。

I love that you've kind of innovated a new middle ground way of fundraising, not bootstrap and not just regular VC. It's a six seed.

而且我喜欢这 200 万——你知道，如果我融了 5000 万，那我会说，好吧，我懂了。咱们别把 5000 万放进银行账户。但你是用 200 万做到这一点的。

And I love that this two mill—like, you know, if I raise 50 million, it'd be like, okay, I get it. Let's not put 50 million in our bank account. But you do that with 2 million.

对我们来说太多了。我们不能——我们不想看到账户里有那么多钱。

It's too much for us. We can't—we don't want to see that in our account.

这是另一回事，你知道，我们会看看这怎么发展。比如，我可能两年后回到这里哭诉，因为我们没融到足够的钱之类的。谁知道呢？但这是另一回事：我确实认为我们能用很少的钱走得更远。比如 Kora，我觉得打造 Kora 总共花了大概 30 万。也许这很疯狂，因为——

That's another thing, and you know, we'll see how this ages. Like, I might be back here in two years crying the blues because we didn't raise enough money or whatever. Who knows? But that's the other thing: I do think we can get so much further with very small amounts of money. Like, Kora, I think all in to build Kora, we've spent maybe 300k. Maybe that's crazy because—

包括工资。是的。

Includes salaries. Yeah.

哇。

Wow.

这个产品——

This product—

这个产品在三年前即使你有几十亿美元，在技术上也不可能实现。不可能，因为没有 GPT 你就无法做邮件摘要、自动回复之类的事情。所以它不仅完全不可能，而且现在我们用两个工程师就能完成过去需要 20 人团队才能完成的工作量。我觉得这意味着我们需要的钱更少。而且我不认为 VC 已经真正跟上这一点。我觉得还有其他公司在做——有个术语叫“种子自举”。所以还有其他公司也开始意识到这一点。我很好奇这会如何改变 VC 模式，当然。对我们来说，我们有一个特定的孵化模式，这和 VC 模式有点不同。我觉得我们在和创始人合作方面有一些差异化，这挺酷的。但是的，我们——我只是在试图找到一种适合我的形态，它和其他人不同，我们看看会怎么发展。

This product was not even technically possible even if you had billions of dollars three years ago. Not possible, because you can't do email summarizing and automatic responses and all that kind of stuff without GPT. So not only was it totally impossible, but now we can get, with two engineers, the amount done that would have taken a team of like 20 people. And I think that means that we need less money. And I don't think that VC has really caught up to that yet. And I think there are other companies that are doing—there's like a term called like 'seedstropping.' So there are other companies that are kind of starting to wake up to this too. And I'm curious about how it changes the VC model, for sure. For us, we have a specific incubation model, which is a bit different from a VC model. And I think there's some differentiation in the stuff that we can do with founders, which is kind of cool. But yeah, we're—I'm just trying to figure out a shape that works for me, and that's different from other people, and we'll see how this goes.

我们过几年再回访。

We'll revisit in a couple years.

是的，从外面看似乎进展顺利。

Yeah, seems like it's going great from the outside.

在结束之前，我想问几件别的事。一件是关于你们那个咨询部门。我觉得它真的很有意思，因为就像我说的，我觉得这可能是一门 10 亿美元的业务。我觉得现在每家公司都在试图弄清楚“到底别人搞明白了什么而我们没做？”我收到过很多公司首席产品官的邮件，说“你能给我介绍一些用 AI 做了很酷的事情、值得我们学习的首席产品官吗？”很多人，我就把他们互相介绍认识，这很酷，因为你们基本上就是在为很多公司解决这个问题。

I want to ask about a couple other things before we wrap up. One is around this consulting arm that you have. I think it's really interesting because, like I said, I feel like this could be a billion dollar business. I feel like every company right now is trying to figure out what the hell—what the hell's everyone else figured out that we're not doing? I've had so many emails from chief product officers at companies being like, 'Can you introduce me to some chief product officers that have done cool things with AI that we should learn from?' So many people, and I just introduce them to each other, and it's cool because you guys are basically solving that problem for a lot of companies.

那么，首先请简单分享一下这块业务是做什么的；其次，我觉得你肯定见过一些公司做得非常好，成功采用了 AI，取得了很好的生产力提升；也见过一些公司没有做到。你觉得这两者的区别是什么？

So one is just maybe share a bit about what that side of the business for folks and then two I feel like you I imagine you've seen companies that have done this really well have adopted AI things have worked really well they found really good productivity gains and then you found companies that don't what do you find is the difference between those two

我喜欢这个问题，而且我对此有非常具体的看法。首先，咨询这块业务基本上就是：我们花所有时间摆弄新模型，写关于它们的文章，用它们构建东西，而且我们有很大的受众。所以很自然地，随着时间推移，有公司来找我们说：“你能来教我们怎么做吗？”于是我们就开始做这件事了。这大概是最近 6 到 9 个月的事，但现在它已经是一项相当大的业务了。它今年可能会翻倍。去年我们做了大约 100 万美元，今年可能会更多，我们拭目以待。这取决于几个大合同，所以可能会远超这个数。我预测几年内会达到 10 亿美元。但基本上，人们就是问：“你能来帮我们学习怎么做吗？”所以我们做的是：花一些时间去研究你的组织。我们会深入了解不同团队在做什么，有哪些重复性任务，就像我们之前讨论的一些内容。然后我们会先提交一份简短的报告，告诉你我们发现的一切。不仅如此，你还可以用一个聊天机器人，和我们做的所有访谈对话，提取你自己的见解。我们有一个完整的仪表盘，显示哪些团队对此很感兴趣，哪些不感兴趣，以及根据访谈和 AI 分析，你在不同团队可能获得多大的杠杆效应。这很酷。这个应用是我大约一年前和 Devon 一起在一个周末用 vibe coding 做出来的，后来 Alex 负责咨询这块，帮助升级了它。然后我们有培训课程。我们会进去培训每个团队，并根据我们的访谈进行定制，因为 AI 的一个有趣之处在于它是一种通用技术。我认为公司内部的人中，10% 的人会说“我对这个超级好奇”，10% 的人会说“我永远不会碰这个”，80% 的人会说“如果你告诉我怎么用在我的工作上，我就会用”。所以我们定制培训，比如“这是你将使用的确切提示词，这是你将使用它们的确切场景”。我认为这确实有助于推动采用。我们每个团队花四周时间，每周一小时，诸如此类。这似乎真的很酷。之后我们还经常构建自动化，做我们之前讨论过的一些 AI 运营工作。公司真的很喜欢。我们和很多大型对冲基金、私募股权公司以及大公司合作。

I love this question um and I have a very specific opinion about this um so one yeah the consulting arm basically like we spend all of our time playing around with new models writing about them and building stuff with them and we have a big audience So naturally like we've gotten companies over time being like can you just come and teach us how to do this and so we started to do that. This is you know pretty nent. It's probably been over the last like 6 to9 months but like it's a pretty big business now. Um like it's our it's it'll probably double this year. Like last year we did about a million. Um maybe it'll be maybe it'll be more this year. We'll see. It depends on a couple we have a couple big contracts out so it might be way more than that. Um >> billion. I I predict a billion dollars in a few years. But yeah, basically people are like, can you come help us learn how to do this? So what we do is um we spend some time going and researching your organization. So we go in and try to understand like what is what are all the different teams doing? What are the repetitive tasks? Some of like some of the stuff we were talking about earlier. Um and then what we will do is uh first we present a little report tells you like here's everything that we found. Here's um not only that, but you have a chatbot where you can chat with all the interviews that we did and you can pull out your own insights. We have a whole dashboard where it shows you like here's are the teams that are really into this. Here are the teams that are not. Here's like how much um uh how much leverage you might be able to get on different teams based on the interviews and based on the AI analysis. It's pretty cool. Um and this is like that's an app that I like vibe coded like over a weekend with Devon like a year ago and then um Alex runs the part of the consulting like has helped upgrade it. Um uh then what we do is we have a training curriculum. So we go in and train each team that and we customize it based on um the interviews that we do because one of the interesting things about AI is it's such a general purpose technology and I think people who work inside companies 10% of them are like I'm super curious about this. 10% are like I will never touch this and 80% are like if you tell me how to do it for my job I'll do it. And so we customize the training to be like here are the exact prompts you're going to use um and here's the exact situations you're going to use them. And that really I think helps drive the adoption. We spend four weeks with each team, an hour a week, that kind of thing. Um it seems to be really cool and then we'll often also after this go and build automations and do some of the AI operations stuff we were talking about earlier. Companies really like it. Um I think the we work with a lot of like big hedge funds and PE firms and um big companies all that kind of stuff.

回到你的第二个问题，也就是什么区分了好公司和坏公司，或者说那些最终采用了 AI 的公司。我认为首要的预测因素是 CEO 是否使用 ChatGPT 或你选择的聊天机器人。如果 CEO 一直在用，并且说“这是最酷的东西”，那么其他人也会开始用。如果 CEO 说“我不知道，这是给别人用的”，那么没有人能领导这项变革。他们要么对此持负面态度，所以肯定没人会用；要么他们有完全不切实际的期望，因为他们对可能发生的事情没有直觉，然后就会非常失望。但那些一直在使用 AI 的 CEO 能够既激发热情，又设定合理的期望。所以这些事情最终会进展得很好。那些做得好的公司，比如我们合作的一家叫 Walleye 的对冲基金，几周前我请它的创始人上了我的播客 AI and I。他们是一家规模达 100 亿美元的大型对冲基金。他们做的一件事，我认为基本上是做这件事的典范。他做的第一件事，很多 CEO 都在做，就是发一封“我们是 AI 优先公司”的邮件。每个人都收到了备忘录，你真的必须去做。他在备忘录里说了一句话，我很喜欢：“我是用 ChatGPT 写的这封邮件，你也应该这样。”所以你必须这样带头。然后他做的，我认为很多其他很酷的公司也在做的，就是每周开会，人们分享提示词和用例。他们每周给全公司发一封邮件，说：“好的，这是我们的使用情况，这是 ChatGPT 的使用统计，这是那些提出新提示词并做出贡献的人。”这样就创造了这种意识和动力，因为回到我之前提到的观点，10% 的人是早期采用者。你需要找到并突出公司里的这些人，因为他们会花大量时间弄清楚什么有效，然后你只需要把他们的经验传授给组织的其他人。所以如果你为他们创造获得回报的论坛，你就会自动把他们的很多经验传授给其他人，并鼓励更多这样的行为。我认为这就是秘诀。

To your other to your your second question which is like what separates the good companies from the bad or the companies that end up adop adopting this. I think the the number one predictor is does the CEO use CHBT or insert your own chatbot. If the CEO is in it all the time being like this is the coolest thing everybody else is going to start doing it. If the CEO is like, I don't know, this is for someone else. Like, no one else is going to be able to lead that charge. Um, and they're either going to have uh either they're going to be negative on it and so definitely no one's going to do it or they're going to have way unrealistic expectations because they have no intuition for what's possible and they're just going to get really disappointed. But the CEOs that are using it all the time are able to like both drive the excitement and set reasonable expectations for what can be achieved. And so those things end up working really well. And the people that do this really well. So for example, we um we work with a hedge fund called Walleye, which I had the founder on my podcast AI and I um a few weeks ago. They're gigantic $10 billion hedge fund. Like one of the things that they do, which I think is I think they're basically the model for like how to do this. First thing he did, which a lot of CEOs are doing, is send the we're an AI first company email. Everyone's got the memo. You just got to really do it. And one of the things he said in his memo, which I love, is I wrote this I wrote this email with ChetT and you should too. So like you got to like >> in the memo >> you got to like lead from the front in that way. And then what he does and I think what a lot of other like really cool companies do is they're doing like weekly uh meetings where people share prompts and share use cases. They're doing um they do like a weekly email to their entire company being like, "Okay, here's our here's our usage. Here are our usage stats for CatchBT. Here are the here are the people that like um uh here are the people that came up with a new prompt and contributed to it." Like create this this sort of like awareness and momentum because what's going back to the point I made earlier about you know 10% of people are early adopters. Those are the people inside of a company that you need to find and highlight because they're going to just go spend all this time like figuring out what works and then all you have to do is like translate what they learn into the rest of the organization. And so if you create forums for them to be rewarded, you're going to automatically transfer a lot of their learnings to everybody else and encourage more of it. And I think that's kind of this the secret.

太棒了。我喜欢这个建议。那么，回顾一下你刚才分享的，你发现并鼓励公司采用的一些策略。一是发送这份备忘录，托比备忘录。我不知道这样描述是否恰当，我认为他是最早这样做的。就是“我们是 AI 优先”，它将成为你绩效评估的一部分，会问你“你能用 AI 做这件事吗？”而不是“你能和其他人谈谈吗？”所有这些，然后注明“我是用 ChatGPT 写的”。这是个好主意。还有每周会议的想法，就像现场或 Zoom 会议，人们分享“这是我在使用 AI 时学到的东西”。然后是每周统计邮件，显示“我们在整个公司使用了多少 ChatGPT”或“有些人做了很棒的工作”。

That is awesome. I love this advice. So, just to reflect back what you just shared, a few kind of uh tactics you find that you encourage within companies. One is just send send this memo, the Toby memo. I don't know if that's the right way to describe it, who I think was first along these lines. Just we're AI first. It's going to be part of your performance review. It's going to be asking can you do it in AI before you could you talk to anyone else? All these things and then just note I help I wrote this using chat JBTs. It's a great idea. Uh this idea of a weekly meeting. So, it's like a live or Zoom meeting where people share, here's the thing I've learned about using AI. Uh, and then this weekly stats email of here's how much we're using Chatbt across the or here's some people that did some awesome work.

是的，

Yeah,

太棒了。

amazing.

我特别喜欢这个简单的启发式方法：如果你的 CEO 每天使用 ChatGPT 或 Claude 之类的工具，那么一切都会顺利。

And I especially love this very simple heuristic of if your CEO uses ChatGPT or Claude or whatever daily, then it's going to work out.

是的。

Yeah.

这太酷了。我知道现在还为时过早，但你看到公司积极拥抱 AI 并广泛采用后，产生了什么样的影响？无论是轶事还是数据方面，你看到了什么？

That is super cool. I know it's early, but what kind of impact have you seen from a company leaning into this and adopting AI widely? Anything you've seen either anecdotally or numbers-wise?

现在还为时过早。除了说我认为那些做得好的人现在觉得他们可以在不增加人手的情况下完成比以前多得多的工作之外，真的很难说。所以他们只是以同样的预算走得更远、更快。我实际上没有看到很多人说“太好了，我们要解雇一堆人”。我也不想做那样的咨询，那太糟糕了。但我们从来没有说过“不”。大多数人都是说“太好了，我要用现有的人走得更远”。我还想回到我最初提到的关于安抚美国就业的观点。我见过一些公司，不是我们合作过的，而是我朋友的公司，他们说“我们在某个地方有一个呼叫中心”。但我认为我可以用美国的两名员工，使用这些客户服务平台之一，完成同样的工作量。他们仍然不是完全自动化的。就像那个 Clara CEO 的事情……但是的，你可以让美国有几个人，也许你付给他们的工资比你在其他地方付给 100 个人的工资要少一点。显然，这是每个人都要自己做的计算。但我确实看到过这种情况发生。是的，我认为这就是用同样数量的人做更多的事。

It's early. It's really hard to say other than I think generally people who do this well now feel like they can do way more work than they used to without having to hire more people. And so they're just going further faster at the same budget. I actually don't see a lot of people being like, 'Cool, we're going to fire a bunch of people.' Like, I don't really want to do consulting like that. That sucks. But we've never had to say no. Mostly people are like, 'Cool, I'm just going to go further with the people that I have.' I think also back to the first point I made about reassuring American jobs. I have seen some companies, not the ones we worked with, but I have seen some companies of people that I'm friends with where they're like, 'We have a call center somewhere.' But I think I can get the same amount done with like two employees in the US that use one of these customer service platforms. They're still not totally automatic. Like that Clara CEO thing that was... but yeah, you can have a couple people in the US that maybe you pay a little bit less to than you would for like 100 people somewhere else. And obviously, that's a calculus that everyone has to make for themselves. But I've definitely seen that happen. And yeah, I think that's the you get more done with the same amount of people.

也许在结束我们的对话之前，我想回到你提到的这个想法，但我想花更多时间讨论它，这就是“分配经济”的想法。如果我理解正确的话，我们一直处于知识经济中，人们通过做事获得报酬。而你的论点是，我们正在转向这种分配经济，技能成为管理技能，管理技能变得更加重要，我们将把更多时间花在管理上。我认为这很棒的一点是，它也告诉你未来哪些技能更重要，这是我认为很多人都在思考的事情。所以也许就回答这个问题，分享你认为重要的内容，让人们了解你的想法。

Maybe to close out our conversation, I want to come back to this idea that you referenced, but I want to spend a little more time on this, which is this idea of the allocation economy. If I understand it correctly, we've been in this knowledge economy where people get paid to do a thing. And your thesis is that we're moving to this allocation economy where skills become the manager skills become more important and we're going to be spending more of our time managing. And I think what's amazing about this is it also tells you which skills will matter more in the future, which is something I think a lot of people are thinking about. So maybe just answer that question and share whatever you think is important to share to give people a sense of what you're thinking.

是的。这基于我大约两年半前写的一篇文章。那是在智能体被认为可行之前。我当时真的在思考如何表达我每天使用这些工具的经验中，哪些技能对我有用。因为我认为这对很多其他人也会如此，而且我认为做这类预测的最好方法是你必须自己一直在做，然后这会影响你对这些事情的看法。所以，我当时使用 GPT-3 或 GPT-4 时注意到，我花了很多时间，例如，思考如何沟通问题？如何为问题收集正确的信息？如何以正确的方式呈现，以便我使用的模型能理解？如何选择将问题交给哪个模型？以及如何划分任务，比如“好吧，这个模型做这个，那个模型做那个”？基于我所知道的好与坏？如何给他们反馈？如何对我想要的东西有一个愿景，并有一套标准来判断它是否好？所有这些正是我发现自己使用这些工具的方式。我当时想，“哦，这就是管理。”一旦你明白了这一点，我想你会开始看到很多其他东西。所以，一个很好的例子是，有一个很大的抱怨是，“我怎么能让 AI 做这个？我不相信它们能做好，所以我应该自己做。”我就说，“是的，这正是每个第一次当经理的人说的话。”你总是会遇到这个问题，你会想，“好吧，如果我委派下去，它不会按照我想要的方式完成。如果我自己做，我就没有杠杆。”所以这就是经理必须学会如何当经理的方式，比如我什么时候介入，也许稍微微观管理一下，什么时候可以委派，我如何信任它，我如何划分任务，等等。所以我认为这些技能有很多重叠之处，而且这些技能目前还没有广泛分布，但未来会如此，因为当经理的成本会低得多。

Yeah. So, this is based on an article I wrote like two and a half years ago. So, this is back before agents were even thought of as viable. And I was really trying to think about how do I express what in my experience using this every day, what skills are useful for me. Because I think that'll be the case for a lot of other people and I think that's the best method to do these sorts of predictions is you have to be doing it all the time yourself and then that informs your opinion about this stuff. So, what I noticed using at the time like GPT-3 or maybe GPT-4, was that I was spending a lot of time, for example, thinking about how do I communicate the problem? How do I gather the right information for the problem? How do I put it in the right way so that the model that I'm working with gets it? How do I pick which model to give it to? And how do I maybe divide up the task to be like, 'Okay, this model does this, this model does this.' Based on what I know to be like what's good and what's bad? How do I give them feedback? How do I have a vision for what I want and a set of criteria for whether it's good? All that stuff is exactly how I found myself using these tools. And I was like, 'Oh, that's just managing.' And once that clicks for you, I think you'll start to see a lot of other things. So, a really good example is there's a big complaint that it's like, 'Well, how can I have AI do this? I can't trust that they're going to do it well, so I should just do it myself.' And I'm just like, 'Yeah, that's exactly what every first-time manager says.' You always have this problem where you're like, 'Okay, if I delegate it, it's not done in the way that I want it to be done. If I do it myself, I get no leverage.' And so that's how a manager has to learn how to be a manager, like when do I lean in and maybe micromanage a little bit and when can I delegate and how can I trust it and how do I divide up the task and all that kind of stuff. And so I think there's a lot of overlap in those skills and those skills are not broadly distributed right now, but they will be in the future because it will be so much cheaper to be a manager.

具体来说，我看了你写的文章，你强调的技能将更有价值：评估人才、愿景、品味，以及你说的，何时深入细节，何时值得深入。

And specifically I was looking at the article you wrote, the skills that you highlight will be more valuable: table is evaluating talent, vision, taste, and to your point, when to get into the details, when it makes sense to dive in.

是的。

Yeah.

太棒了。然后还有一个你提到的相关观点，那就是通才在未来会变得越来越有价值。你提到每个人在每件事上都是通才。

Awesome. And then there's also kind of a connected point you made that you referenced, which is that generalists will become more and more valuable in the future. You mentioned that everyone at every is a generalist.

是的。

Yeah.

请分享一下这方面。

Share a little bit about that.

是的。我觉得，我的意思是，也许因为我是一个通才，所以你应该对此持保留态度。

Yeah. I find, I mean, maybe it's because I'm a generalist, so you should take this with a grain of salt.

我也是。

Same.

是的。我觉得，我的意思是，也许因为我是一个通才，所以你应该对此持保留态度。但我认为这正是 AI 对我来说如此棒的原因之一，我喜欢涉猎不同的事物。所以就像在一天之内，我可以编写一个应用程序、制作一个视频、制作图像、写作等等，而 ChatGPT 就在我身边。我认为，随着文明从古希腊发展到现在，我们基本上发现的是，我们越专业化，就越能在许多不同的人之间进行协调。所以这有点像亚当·斯密，你知道，就像有一个别针工厂，有人在制造别针，或者不管他的理论是什么，就是专业化与贸易。这带来了很多非常好的影响。我认为我最喜欢的例子之一是回到古希腊，古雅典。雅典是一个通才文明，至少对公民来说是这样。就像他们有一些，你知道，对女性和奴隶的糟糕历史。但让我们暂时把这一点放在一边。如果你是一个公民通才，你可以被期望成为一名战士、一名法官、一名陪审员，也许是一名将军。

But I think that's one of the things that has made AI so awesome for me, like I love to dabble in different things. So it's like in one day I can be coding an app and making a video and making images and writing and all that kind of stuff, and ChatGPT is right there with me. And I think what we've basically what has happened as civilization has progressed from ancient Greece to now is what we've discovered is the more that we specialize, the better we can coordinate across many different people. And so it's sort of like the Adam Smith, you know, like there's a pin factory and someone's making a pin or whatever his thing is, is specialization and trade. And there have been a lot of really good impacts of that. And I think one of my favorite examples of this is back to ancient Greece, ancient Athens. Athens was a civilization of generalists, at least for citizens. It's like they have some, you know, a bad history with women and people who are slaves. But let's just put that to the side for a second. If you were a citizen generalist, you could be expected to be a fighter, a judge, a juror, maybe a general.

就像你可以预期，它在你的一生中会在你的社会中扮演许多不同的角色。但这种情况改变了，因为雅典变成了一个帝国，当它变成帝国后，如果你要派一位将军去入侵西西里岛之类的，你会希望那个人相当有技能，所以它开始打破那种通才的模式，人们开始有特定的角色，并且相互协调等等。我认为这种模式实际上对文明发展非常有益，但在很多方面，它也不那么有趣。做一个全面发展的人真的很酷。我认为人工智能的有趣之处在于，它有点像你可以把它想象成口袋里装着 1 万个博士。它几乎了解人类知识的每一个分支、每一种艺术形式，以及每一种制造或建造事物的方式，你都可以随时获取这些知识。所以，它擅长处理许多你可能需要花 10 年才能精通的专业任务，比如了解某种特定的蝉，知道它们如何繁殖。但现在你口袋里的这个东西可以在任何情境、任何时间告诉你所有相关信息。因此，你被赋予了在所有这些不同技能领域之间更频繁切换的能力。而且你可以完成更多事情，比如作为创始人，我认为我们可以保持 15 人的规模更长时间，比我们原本能做到的更长。所以，每个公司里的人都能更长时间地保持通才状态。我认为这可能会波及到经济的其他部分，不再是那种每个员工只做一个小按钮操作的巨型公司，而是会有更多由通才组成的小型组织。我认为这实际上会是一件非常好的事情。

Like there's you could expect it to have many different roles inside of your society in your lifetime. That changed though because Athens became an empire and as it became an empire if you're going to send like a general off to go and invade Sicily or whatever you want that person to be pretty skilled and so it started to break the general kind of thing into people start to have specific roles and they coordinate with each other and all that kind of stuff. And I think that that pattern has actually been really good for developing civilization, but it's also in a lot of ways like it's not as fun. It's actually really cool to be a well-rounded person. And I think the interesting thing about AI is that it's a little bit like you can think of it like having 10,000 PhDs in your pocket. It's like it knows so much about every little branch of human knowledge and every art form and every, you know, way of making things or building things and you just have access to that. So, it's doing a lot of the it's good for doing a lot of the specialized tasks that you might have had to spend like 10 years getting good at, you know, learning about this particular species of cicada, so you know exactly how they like, you know, reproduce. But now you've got this thing in your pocket that can tell you all about that in any given context at any given time. And so, you're empowered to jump a lot more between all those different domains of skill. And you can get more done as for example like a founder where I think we can stay at 15 people much longer than we would be able to. So the people inside of every can stay generalists for much longer. And I think that that may like sort of ripple out into the rest of the economy where instead of like gigantic massive corporations where like each person is doing like one little like button turning, you have many more smaller organizations with more generalists. And I think that would actually be a really good thing.

这让我想起，我和我尝试合作一段时间的私人教练聊天，她说她是一个很有远见、层次很高的人，不擅长执行，比如保持条理，而 GPT 对她来说简直是天赐之物，因为她只需要说“这是我大概想做的，帮我完成它”。

This reminds me, I was talking to my personal trainer that I'm trying out for a little bit and she said that she's a very big vision kind of high level person and not good at executing, like staying organized, and GPT is such a godsend for her because she's just like here's what I want to do roughly, just help me get it done.

那太好了。

That's great.

所以是的，这真的让我想到这些东西将释放多少价值。

And so yeah, it really made me think about just how much value all this stuff is going to unlock.

这太棒了。这正是我想要的一切。但至此，我们进入了非常激动人心的快问快答环节。丹，你准备好了吗？

This was amazing. It was everything I wanted it to be. But with that we reached our very exciting lightning round. Dan, are you ready?

我准备好了。

I'm ready.

我们开始吧。你发现自己最常推荐给别人的两三本书是什么？

Here we go. What are two or three books that you find yourself recommending most to other people?

嗯，我已经推荐过一本了，就是《战争与和平》。绝对要读。如果你想要托尔斯泰的入门读物，我会推荐《伊凡·伊里奇之死》。另一本好书是乔治·桑德斯的《雨中的池塘》，这是一本关于写作的俄罗斯短篇小说集。我特别喜欢俄罗斯作家，因为很多俄罗斯小说家都在处理技术对俄罗斯传统生活方式的影响。他们处于一种非常有趣的中立地带，介于对世界的浪漫看法和更理性主义的“我们在进步”之间。这就是你在《安娜·卡列尼娜》中会发现的东西，当列文和农民一起在田里割草时，那是托尔斯泰在思考：如果我不是一个试图让农场更高效的贵族，而是拿着镰刀，那样真的很开心。所以他们在处理很多我认为与人工智能类似的东西。《大师与使者》是另一本非常好的书，它基本上讲的是大脑的不同半球如何看待现实。它真的非常非常好，我认为它也与很多人工智能的东西有关。我想这就是我的三四本书。

Well, I already recommended one, which is War and Peace. Definitely got to read that. If you want like a Tolstoy primer, I would read The Death of Ivan Ilyich. Another good one is A Swim in a Pond in the Rain, which is by George Saunders. And that's a collection of Russian short stories that is also about writing. And I in particular I really like the Russians because a lot of the Russian novelists are dealing with the effects of technology on traditional Russian way of life. And they're very kind of in this really interesting middle ground between a sort of romantic outlook on the world and a more rationalist like we're making progress. And that's one of the things you'll find in Anna Karenina when Levin is out in the fields with the peasants doing the scythe thing, that's Tolstoy kind of thinking about oh what would it be like instead of being a nobleman who's trying to make farms way more efficient, I was just like with my scythe and that was really happy anyway. So they're dealing with a lot of similar stuff to I think AI. The Master and His Emissary is another really good one and that's about basically how the different hemispheres of the brain view reality. It's really really good and I think it relates to a lot of AI stuff too. I think those are my three or four.

很棒的书单。我觉得这些书大多数人都没提过，所以这总是一个好迹象。呃，你最近有没有特别喜欢看的电影或电视剧？

Excellent list. I think nobody's mentioned most of either any of these so this is that's always a good sign. Uh, do you have a favorite recent movie or TV show you really enjoyed?

有。我真的很喜欢《死木》。你看过吗？

Yes. I really love Deadwood. Have you seen it?

我绝对喜欢它。我记得他们因为某种原因停播了。我想他得去 HBO 做别的事情。太可惜了。它太棒了。

I absolutely love it. I remember when they stopped it for some reason. I think he had to go do something else at HBO. It was so sad. It's amazing.

是的。大卫·米尔奇太不可思议了。国宝级人物，了不起的作家。但我真正喜欢它的地方，而且我最近才看，是他谈到《死木》是关于秩序如何从混乱中形成的。所以这是一个边疆小镇。人们涌向那里，没有法律，没有规则，但到了第三季，有了市长，所有行业都进来了，变成了一个真正像样的城镇，我就是喜欢这一点。我认为从西部边疆到技术前沿有很多相似之处。所以我认为这部剧是对这种动态的一个非常有趣的研究。

Yeah. David Milch is incredible. National treasure, incredible writer. But what I really love about it and I only recently watched it is he talks about Deadwood being about how order forms out of chaos. So it's this frontier town. People are going to it and there's no law, there's no rules and by season 3 there's a mayor and all the industry has come in and it's a real proper town and I just love that. And I think there's a lot of parallels from the western frontier to technology frontiers. And so I think that show is like a really interesting study in that kind of dynamic.

我喜欢一切如何与技术运作以及人工智能如何产生联系起来。我喜欢这个。

I love how everything connects to how tech works and how AI came to be. I love this.

谢谢。

Thank you.

你最近有没有发现特别喜欢的产品？

Do you have a favorite product you've recently discovered that you really love?

我没有很好的答案，因为我花了很多时间使用我们的内部产品。但我喜欢的标准答案是 Granola。所以我真的很喜欢 Granola。我对他们的一个不满，我希望他们听这个播客，是我真的很想导出我所有的笔记。我想要 API。

I don't have a good answer for that because I just spent a lot of time using our internal products. But I like my stock answer is Granola. So I do really love Granola. My one gripe with them and I hope they listen to this podcast is I really want to export all my notes. I want API.

但除此之外，我认为这是一个很棒的产品。

But other than that I think it's a fantastic product.

这绝对是过去几个月这个环节中提到最多的产品。所以是的，Ketchup Granola。我忍不住提到，如果你成为我新闻通讯的年度订阅者，你可以免费获得一年的 Granola。真是划算。而且不仅仅是你，你的整个公司都可以免费获得一年的 Granola。真是划算。

That is definitely the most mentioned product in this segment for the past couple months. So yeah, Ketchup Granola. I can't help but mention you get a year free of Granola if you become an annual subscriber of my newsletter. What a freaking deal. And not just you, but your whole company gets free Granola for a year. What a deal.

这不是我的付费推广。我只是，你知道，这就是我的感受。所以，我很高兴它包含在套餐里。

This is not a paid promotion by me. I just, you know, that's just how I feel. So, I'm glad it's part of the bundle.

是的。太棒了。好的。你有没有最喜欢的人生格言，经常在工作中或生活中发现它有用？

Yeah. Incredible. Okay. Do you have a favorite life motto that you often come back to find useful in work or in life?

所以，基本上，就像我以前是“一切皆有记忆”。所以，我想，你知道，我要上莱尼的播客。我的人生格言会是什么？它说，你的人生格言是“深刻见证，勇敢建造”。

So, basically, like I used to be all and has memory. So, I was like, you know, I'm going on Lenny's podcast. What would my life motto be? And it said, your life motto is a witness deeply, build bravely.

你珍视缓慢而专注的观察，无论是阅读托尔斯泰、追踪冥想主题，还是剖析大卫·米尔奇的段落。所以这正好涵盖了我刚才提到的所有内容，这真的很有趣。

You prize slow, attentive seeing, whether it's reading Tolstoy, tracking meditation themes, or X-raying a David Milch paragraph. So it's hitting all the stuff I just mentioned, which is really funny.

然后勇敢地构建，你把这些洞察转化为具体的东西，比如 Every、Kora 和长文随笔等等。所以我觉得这里面有某种东西。这让我想起了真正的座右铭，这不是我发明的。我想是小普林尼说的：“做值得写的事，写值得读的东西。”这似乎是一个很好的总结。

And then build bravely, you turn those insights into concrete things like Every and Kora and long-form essays and all that kind of stuff. So I think there's something about that. This actually reminds me of the actual motto, which I didn't come up with. I think it's like Pliny the Younger said, "Do things worth writing about and write things worth reading." Seems like a pretty good summation.

做值得写的事，读值得读的东西。

Do things worth writing about and read things worth reading.

写值得读的东西。

Write things worth reading.

写值得读的东西。这应该是我们两家通讯的座右铭。

Write things worth reading. That should be the motto of both of our newsletters.

这真的很棒。好的。顺便说一句，我很喜欢你问 ChatGPT：“我的人生座右铭是什么？”

That is really good. Okay. And by the way, I love that you asked ChatGPT, "What's my life motto?"

等等，这很有意思。它没有直接给我答案，但启发了答案。是的。

And wait, this is interesting. So, it didn't give me the answer, but inspired the answer. Yeah.

我觉得这实际上正是我使用它的方式。

And I think that's actually exactly how I use it.

哇。它已经是大脑的延伸了。

Wow. It's an extension of our brains already.

是的。

Yeah.

最后一个问题。我在某处读到，你曾一度停止写作。你当时想：“我需要做其他事情，我需要建立这家公司。”然后你意识到：“我需要回到写作，因为事情开始偏离正轨。”我觉得这与你谈到的很多内容形成了有趣的推论：做让你快乐的事，靠近喜悦。请分享一下当时发生了什么，因为我之前不知道。

Last question. I was reading somewhere where you wrote that you stopped writing at one point. You were just like, "I need to do other things. I need to build this company." And then you realized, "I need to get back to writing because things started going sideways." And I feel like this is such an interesting corollary to a lot of the stuff you talked about: do things that make you happy, stay close to joy. Just share what happened there, because I didn't know that.

这绝对不是快问快答能说清的。所以我会详细说明，但我会尽量简短。

This is definitely not a lightning round thing. So I'll expound, but I'll try to do it as quickly as possible.

完美。

Perfect.

我认为一般来说，当你在建立一家公司时，即使你像我这样做事——你知道，不筹集大量资金，试图保持控制——也会有巨大的诱惑，试图以你认为应该的方式经营公司。我有个奇怪的地方，就是“我真的很喜欢写作，但我也真的很喜欢商业。”当时对我来说，既拥有成功企业又是作家的人并不多。事实证明确实有，但我有一段时间不知道。所以，在 Every 的早期，公司发展得很好，因为我写了很多，内森也写了很多。当我停止写作时，业务就不那么好了，因为媒体企业不像科技初创公司那样遵循同样的模式。如果你是媒体企业，你是创始人，然后你雇人来制作产品，这是对的，如果你之前有产品市场契合度，你就会失去它。也许你雇了优秀的写手，但这很难。这与初创公司完全相反。你构建产品的第一个版本，然后雇人构建其余部分。所以我就这么做了。我也很纠结：“好吧，这对我和我的职业生涯意味着什么？”我觉得很难承认我其实想写作，因为我没有见过任何我想成为的那种作家的例子。非常有趣的是，公司成立三年后，业务一直相当平淡。我非常痛苦，因为我没有做我真正想做的事情。我问 ChatGPT：“有没有作家建立企业的例子？”它说：“有，乔尔·斯波尔斯基，他建立了 Trello 和 Stack Overflow。还有杰森·弗里德，我认识他很久了，一直很敬佩，但在这个语境下我忘了。还有萨姆·哈里斯，他有一个很棒的播客，还有一个巨大的冥想应用。”还有比尔·西蒙斯，他是一位了不起的播客主播，还建立了 The Ringer，以几亿美元卖给了 Spotify。像这样的人有很多，他们用来建立公司的模式是众所周知的。只是它们不是典型的硅谷模式。所以我想：“酷，我只想当个作家。我觉得那会很有趣。”所以我有点翻转了。我仍然有建设者、企业家、创始人的身份部分，但我有点翻转成以写作为中心，而且我对此毫不抱歉。这实际上对业务有好处。对我有好处，对业务也有好处。我越是倾向于这样做，做那种如果你告诉任何人你要创业，比如“我们要做一份通讯，我们要孵化所有这些应用，我们要做咨询等等”，他们会说“你疯了”。每个人都想那样做。当然每个创始人都想那样做，但你必须专注。你不能写，随便。但每次我倾向于某种感觉像终极奢侈的、我隐藏的秘密欲望的东西，它实际上效果更好。我认为真正的问题是，每天做你不那么喜欢或不那么适合的事情，要付出巨大的代价。通过屈服于那些秘密欲望，你最终会为你的工作和业务找到一个适合你的形状。这总是与其他企业有些独特的形状。它总是与其他事物押韵，但我认为找到那个独特的形状，而不是像货物崇拜那样模仿你认为公司应该的样子，绝对是更好的成功方式，也是更好的生活方式。

I think generally when you're building a company, even if you do it the way that I do it or did it, which is you know you don't raise a lot of money and you try to stay in control, there's a big temptation to try to run the company in the way you think you should. And I have this weird thing where I'm like, "I really love writing, but I also really love business." And there just weren't a lot of models for me of people who had successful businesses that were also writers. Turns out there are, but I didn't know about that for a while. And so, early on at Every, it was growing really well because I was writing a lot, Nathan was writing a lot. And when I stopped writing, the business didn't work as well because media businesses don't follow the same pattern as tech startups. If you're a media business and you are a founder who then hires people to make the product, which is right, if you have product-market fit before, you lose it. And maybe you hire people that are good writers, but that's hard. It's total opposite pattern for startups. You build the first version of the product and then you hire people to build the rest of it. And so that's what I did. And I also really struggled with, "Okay, what are the implications for that and for my career?" And I think it was hard for me to admit that I actually want to write, because I just didn't have any examples of someone being the kind of writer that I wanted to be. And what's really interesting is, three years into the business, the business has been pretty flat. I was pretty miserable because I was not doing the thing that I really wanted to do. And I asked ChatGPT, "Are there any examples of writers that have built businesses?" And it was like, "Yeah, Joel Spolsky, who built Trello and Stack Overflow. There's Jason Fried, who I've known for a long time and have always looked up to, but I forgot about in this context. There is Sam Harris, who's got a great podcast and he's got a gigantic meditation app." There is Bill Simmons, who's an incredible podcaster and also built The Ringer, sold to Spotify for a couple hundred million bucks. Like there's a lot of these people, and there are patterns that they use to build companies that are pretty well understood. They're just not typical Silicon Valley patterns. And so I was like, "Cool, I just want to be a writer. I think it would be really fun." And so I sort of flipped it. I still have the builder, entrepreneur, founder part of my identity, but I sort of flipped it to be like writing is at the center, and I'm unapologetic about it. And that's actually good for the business. It's good for me and it's good for the business. And the more I've leaned into that, doing the thing that if you told anyone that you were starting a business where it's like, "Well, we're going to be a newsletter and we're going to incubate all these apps and we're going to do consulting and whatever," they would be like, "You're nuts." Like everyone wants to do that. Of course every founder wants to do that, but you have to focus. You can't write, whatever. But every time I've just leaned into something that feels like the ultimate luxury of my hidden secret desire, it's actually worked a lot better. And I think what it really is, there's a huge tax to doing something every day that you're not quite that much or you're not quite a fit for. And by sort of giving into those secret desires, you end up finding a shape for the work that you do and the business that you build that is good for you. And that's always going to be a somewhat unique shape from other businesses that have been built. It's always going to rhyme with other things, but I think finding that unique shape instead of just kind of cargo culting like what you think a company should look like is definitely a much better way to be successful, and it's also a much better way to live.

我觉得这会引起很多听众的强烈共鸣，他们可能是创始人或想成为创始人，这与本播客上分享类似经验教训的许多人产生共鸣。丹，这太棒了。最后两个问题。人们在哪里可以查看 Every，在网上找到你，以及听众如何能对你有帮助？

I think this is going to hit hard with a lot of people who are listening, who are maybe founders or want to be founders, and this resonates with a lot of people that have been on this podcast sharing similar lessons. Dan, this was incredible. Two final questions. Where can folks check out Every, find you online, and how can listeners be useful to you?

所以，你可以在 every.to 找到我们。我还在 Twitter 上，账号是 DanShipper。你可以去那里查看我们的产品、我们的通讯，如果你想了解 AI 的最新动态等等。我还有一个播客，叫 AI and I。你可以在 YouTube 和 Spotify 上找到它。至于人们如何能有用？说实话，对于像我这样想做事的人来说，最有用的就是我希望人们找到有趣、酷炫的 AI 使用方式，真正让他们的生活变得更好。所以，去做吧，然后告诉我。

So, you can find us at every.to. I'm also on Twitter at DanShipper. You can go there to check out our products, our newsletter, if you want to stay on top of AI, all that kind of stuff. I also have a podcast. It's called AI and I. You can find it on YouTube and on Spotify. And how can people be useful? Honestly, I think the most useful thing for someone like me based on what I want to do is I want people to find interesting, cool ways to use AI that actually help make their lives better. So just go do that and tell me about it.

嗯，我觉得那会很棒。嗯，最好的方式是什么？是，呃，在你 YouTube 节目下评论？还是发邮件、私信你？

Um, and I think that'll be great. Um, what's the best way to tell you? Is it, uh, comments on your YouTube show? Is it emailing you, DM you?

呃，我会说，呃，在推特上 @ 我。嗯，呃，你，如果你订阅了 Every，你也可以回复那些邮件，它们最终会转发给我。嗯。

Uh, I would say, uh, tweet me. Um, uh, you, if you subscribe to every, you can also reply to those emails and they they eventually get forwarded to me. Um,

所以，在推特上 @ 我，回复 Every 的邮件。嗯，如果你想在 YouTube 上评论，那很好。嗯，我在 YouTube 评论区出现得不够多。

So tweet me, reply to every. Um, and if you want to comment on YouTube, great. Um, I'm not in the YouTube comments as much as I should be.

别那样做。也许别那样做。嗯，好的。那么，Dan，这太棒了。非常感谢你的分享。谢谢你来做客。

Don't do that. Maybe don't do that. Um, okay. Well, Dan, this was incredible. Thank you so much for sharing. Thanks for being here.

谢谢你的邀请。

Thanks for having me.

大家再见。

Bye everyone.

非常感谢你的收听。如果你觉得这期节目有价值，你可以在 Apple Podcasts、Spotify 或你最喜欢的播客应用上订阅本节目。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发现这个播客。你可以在 lennispodcast.com 找到所有过往节目或了解更多关于本节目的信息。下期再见。

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.