DHH：智能体工程、Vibe Coding 与编程的未来

Lex Fridman

下面是和 David Heinemeier Hansson——也就是 DHH——的对话。他是 Ruby on Rails 的创造者、37signals 的 CTO、新的 Omarchy Linux 操作系统的作者、畅销书作者、赛车手，也是世界上最直言不讳、产量最高的程序员之一。过去二十多年，对他来说编程意味着一刀一刀雕出漂亮的 Ruby 代码。但最近，自 2025 年末起，DHH 拥抱了 AI 革命，再次变成最直言不讳、产量最高的「智能体工程师」——尽管他讨厌这个词。可以说，他成了那种实践者：AI 做大部分实际编程，人用高层设计、愿景和品味掌舵。这是 Lex Fridman Podcast。亲爱的朋友们，DHH。

The following is a conversation with David Heinemeier Hansson, also known as DHH, creator of Ruby on Rails, CTO of 37signals, creator of the new Omarchy Linux operating system, best-selling author, race car driver, and one of the most outspoken and prolific programmers in the world. For the past 20-plus years, for him, programming meant meticulously handcrafting beautiful Ruby code. But recently, since late 2025, DHH has embraced the AI revolution and has openly transformed himself into, once again, one of the most outspoken and prolific agentic engineers, even though he hates that term. So let’s say practitioners of whatever programming is becoming where AI is doing most of the actual programming and the human steers the ship with high-level design, vision, and taste. This is a Lex Fridman Podcast. And now, dear friends, here’s DHH.

DHH

有几十年什么都不发生，也有几周发生几十年的事。过去九个月，我们看到了几十年的进步。如果你不承认这一刻的分量，那才是妄想，那才是精神病。如果瓶子里突然跳出一个精灵说「你要什么都有，操作系统里你梦想过的每个功能我都能给你，多数五分钟，少数二十分钟，真要疯玩大概两小时」，谁不会发狂？我要最快的、能破限速的车。我要能潜到马里亚纳海沟的潜水表。我要能在不到 60 秒内装完的操作系统。

There are decades where nothing happens and weeks where decades happen. And we have seen decades of progress happen in the last nine months. If you’re not recognizing the gravity of the moment, that’s the delusion. That’s the psychosis. Who would not get delirious if suddenly a genie pops out of the bottle and says, “You can have whatever you want. Every feature you’ve ever dreamed of in an operating system, I can deliver them to you, most of them in five minutes, a few in 20, and if we really go hog wild, it’s gonna take me two hours.” I want the fastest car breaking the speed limits. I want the diver watch that can go down the Mariana Trench. I want the operating system that can install in less than 60 seconds.

Lex Fridman

13 个月前我们坐在这儿谈编程，然后一切都变了。那时你对 AI 在编程里的角色还有点怀疑，之后我们经历了智能体工程的急速演化。简单开场：你对 AI 在编程中角色的看法怎么变了？兴奋？恐惧？还是情绪过山车？

13 months ago, we sat down right here to talk about programming, and then everything changed. At that time, you were a bit skeptical about the role of AI in the process of programming, and then we went through this rapid evolution of agentic engineering. So simple question to start: How has your view on AI’s role in programming changed? Are you excited? Are you terrified? Are you on an emotional rollercoaster ride?

DHH

我超级兴奋。对我来说不存在情绪层面的存在主义威胁，那只在智识层面。情绪层面是百分之百纯粹、不加稀释的喜悦、乐观，以及「我们居然让计算机做到这个」的惊叹。我觉得有趣的是，才隔了 13 个月，就像隔了不同宇宙、不同纪元。我爱这句——我想是列宁说的：有几十年什么都不发生，也有几周发生几十年的事。过去九个月，我们看到了几十年的进步。想象你站在莱特兄弟起飞的那一刻。

I am incredibly excited. There is none of the existential threat. That doesn’t exist for me as an emotional component. It is only there as an intellectual component, and the emotional component for me is 100% pure, unadulterated joy and optimism and amazement that we’ve made computers do this. And I find it so interesting that we talked just 13 months ago because it’s like we talked in different universes, different eras. I love this quote. I think it’s Lenin. There are decades where nothing happens and weeks where decades happen. And we have seen decades of progress happen in the last nine months. I mean, imagine you’re there when the Wright brothers take flight.

昨天《纽约时报》还会写「还要一万年才能飞」，隔天我们已经在天上，再过几年就有跨大西洋航班。整个世界彻底变了。能站在那个时刻是福气。放大到整个人类史，有多少人一生都活在出生时的那个纪元里，从没见过世界和社会彻底翻转？能赶上两次这种翻转，简直是特权。我赶上了从前互联网到后互联网，现在又从前 AI 到后 AI。多幸运的一程。

Yesterday, the New York Times would write, “It’s gonna be 10,000 years before we fly,” and then the day after, we’re up in the skies, and just a few years after that, there were cross-Atlantic planes. The whole world has completely changed. What a blessing to be there in that moment. If you zoom out and look at all of human history, how many humans got to live within the same epoch that they were born in? They never saw that complete change of the world and of society. And to have been blessed with two of those feels just such a privilege. I got to see the internet from pre-internet to post-internet, and then now pre-AI, post-AI. What an amazing run. How fortunate.

Lex Fridman

而且这感觉比人类史上任何事都快。我会说大概在十二月，或许十一月末——

Yeah, but the thing is, this feels like a thing that happened faster than anything else in human history. So I would say somewhere around December, maybe late November—

DHH

11 月 24 日。就是那个精确时刻。

November 24. That’s the exact moment.

Lex Fridman

很多人谈 AI 改变一切的方式，像谈世界大战里一国入侵另一国。对许多优秀开发者来说，真的就从 AI 做点基本自动补全、写 5%、10%、15%、20% 的代码，跳到写 80%。

This is how we talk about AI changing everything. And it really just shifted for many great developers. It shifted to where AI is doing some basic autocomplete, maybe writing 5, 10, 15, 20% of code, to writing 80% of code.

DHH

或者 100%。所以即使回头看一年前那场对话，我其实没有不同意见，意见还是那些。一年前我不喜欢当时提供的 AI 模式：自动补全，或者 AI 聊天机器人。聊天机器人我其实喜欢——从一开始就是好导师，查互联网也很好，但不是要取代我凿代码的东西。然后我们有了智能体。

Or 100. To me, that’s why even when I look back upon our conversation a year ago, I don’t actually have different opinions. I have the same opinions. A year ago, I did not like the mode of AI we were offered. It was the autocomplete mode, or it was the AI chatbot mode. Now, the chatbot I actually liked, as we talked about. Great tutor right from the get-go, great way of looking things up on the internet, not what was gonna replace me chiseling code. But then we get the agents.

智能体开头大约五分钟是新奇玩意，然后变惊人，然后变成「天哪，这是 AGI 吗？」这一切都发生在去年以来，甚至就在过去九个月里。我们有这几个阶段。前智能体时代的 AI：我兴奋，但没有从根本上改写游戏规则，没有彻底改变我怎么工作。我还在凿代码，只是多了个小助手、小搭档，能碰想法、更高效地查信息。然后到了 2025 年 11 月 24 日。对我来说 Opus 4.5 是分界线。我甚至没在 24 号试，大概 26 号才试。我给了它几个任务，发现输出质量离我会写的东西近得诡异。我往后一靠，心想：「刚才发生了什么？」

And the agents start out being curiosities for about five minutes, and then they get amazing, and then they get, “Oh my God, is this AGI?” And all of that happened since last year, even just within the last nine months. We have basically these few phases here. We have AI in the pre-agentic era. I was excited about that, but it was not fundamentally rewriting the rules of the game for me. It was not completely changing how I worked. I was still chiseling code, I just had a little helper, a little sidekick who could bounce ideas off, and I could look up this information online in a more efficient way. Then we get to November 24th, 2025. Opus 4.5, to me, was the dividing line. I didn’t even try it on the 24th. I think I tried it on the 26th. I gave it a couple of tasks, and I realize that the quality of the output is uncannily close to what I would’ve written. And I remember just leaning back and thinking, “What just happened?”

「我们怎么从我夏天跟你抱怨的那种自动补全烂摊子，短短几个月就到了这儿？智能和可用性怎么一起涨的？」这是智能体 harness 的问题——我不确定 Opus 4.5 是否比夏天的 Opus 4 聪明那么多，但它驱动你的电脑、用工具、自检、把智能用到能产出真正有意义工作的能力，完全不同。几乎每个圣诞假期认真玩过的人，都经历了这种心智爆炸：这些智能体属于不同的品类。到十二月我已经在重审所有先验：「哇，如果它能做这个，那个也能吗？」「哦，能。」现在 Opus 4.5 看起来像个迟钝模型。我当时想过：「如果这是我们拿到的最后一个模型，我也够了，我会高兴。」

“How did we go from this autocomplete mess that I was talking to you about in the summer to this just a few short months later? How did we get both the increase in intelligence and then also the increase in usability?” This agent harness question where I don’t know if Opus 4.5 was that much smarter than Opus 4, which is what we had in the summer, but its ability to instrument your computer, to use tools, to check its own work, to apply its intelligence in such a way that you could get real meaningful work out of it, was completely different. And I think this is then the big change that happens for almost anyone who paid attention and started playing with it over the Christmas break. By December, I’m already revisiting all my priors — “Wow, if it can do this, can it also do that?” “Oh, yeah, it can.” And again, Opus 4.5 now looks like a retarded model. I remember thinking at the moment, “If this is the last model we get, I’ll be set. I’ll be happy.”

我可以和 Opus 4.5 过下一个二十年，你听不到我抱怨，因为它能按我想要的方式做所有这些工作。不只是能解题，我还能看它走的路径说「对，对，那儿差点，但几乎」，它会给你两记笔记。你能合并它的代码；如果是 Ruby，代码甚至可以看起来漂亮。Rust，另说。然后我们等到早春，突然有了子智能体。Harness 能把任务切开，原来 Opus 要很久的事突然变成五分之一、十分之一的时间，因为你忽然有八个子智能体在干活。但智能体时代的前两个阶段，我仍觉得必须自己开：告诉它要什么、往哪看，偏了就纠一下，会快很多，但我得坐驾驶座，当审阅者、审计者。

I could live with Opus 4.5 for the next 20 years and you would hear no complaints from me because it was just so incredibly amazing to see an agent do all this work in the way that I wanted it done. Because it was not just about it being able to solve a task, it was also that I could look at its path there and go, “Yep. Yep. Maybe not there, but almost,” and it would give you two notes. You could produce code I wanted to merge. You could produce code that actually looked beautiful if it was written in Ruby. Rust, different question. But then we wait just until early spring, and suddenly we get sub-agents. We get harnesses that can subdivide the task, and something that would take Opus quite a while suddenly took a fifth of the time, a 10th of the time, because it could get chopped up and suddenly you got eight sub-agents working for you. But both of those two first phases of the agentic age, to me, still felt like I had to drive. I could tell it what I wanted, where to look for it, and steer it a little bit when it went off, and then we’ll get there, and I would go much faster. But I had to be in the driver’s seat. I had to tell it what I wanted, and I had to be the reviewer, the auditor of what was coming out.

然后终于到了这个夏天，有了 Opus 5、Fable，还有 Sol、GPT Sol，以及程度稍低的一些开源权重模型，我们到了一个新时代：我不再告诉它我们要去哪。我告诉它我有什么问题，告诉它模糊、含糊的想法。它告诉我我们要去哪，告诉我走哪条路。我还是会看，因为我好奇、喜欢计算机、喜欢结果，但我其实不是非看不可。在产出代码、挑选路线的那部分，我已经变成可选的。记得早期 GPS 吗？比看地图厉害，但你还是得注意：会不会把你开进港口？报纸还写「啊，GPS 很糟，因为人们不注意，开进港口」。现在 GPS 上次把谁开进港口了？根本不再发生。车现在自己开。我现在到的地方是：在我眼下这个领域，我可以信任它，完全自信它会罩我。它不会干蠢事；就算干了，也能恢复。

And then finally now, this summer, with Opus 5, Fable, and Sol, GPT Sol, and to a lesser extent, some of the open weight models, we’ve arrived at a new era where I’m not telling it where we’re going. I’m telling it the problem I have. I’m telling it the fuzzy, vague idea I have. It tells me where we’re going. It tells me which path to take, and I will still look at it because I’m a curious person and I like computers and I like the outcome of it, but I really kinda don’t have to. I’ve become optional in the part that produces the code that picks the route. Remember when early GPS systems came out? They were amazing compared to looking at a map, but you still wanted to pay attention. Is it gonna drive you in the harbor? When was the last time GPS drove anyone in the harbor? Like, that just doesn’t happen anymore. In fact, the cars now just drive themselves. And this is where we’ve arrived at now, that I can trust. For the domain I’m working in right now, I can trust it and feel completely confident that it’s gonna have my back. It’s not gonna do something stupid, and if it does something stupid, it’s gonna be able to recover.

Lex Fridman

该提一下，程序员活动在各种领域。最常见的大概是 web 开发 CRUD：有数据库、有面向用户的界面，来回做事。对那个领域，你可以真的到接近 100% 代码由 AI 写；如果你是好程序员、对幕后有直觉，甚至可以合法地不看代码。然后还有你也在做的：写 Linux 发行版——因为执着速度等等，也许得多看一点代码。再往下是安全关键系统，核电站、自动驾驶，也许要更仔细地看。

We should mention that there’s all kinds of domains that programmers operate in. I think the most common domain in development is web dev CRUD. You have a database, a user-facing interface, and it does something back and forth. And I think for that, you could really get to the 100% of code written by AI and really almost — if you’re a good programmer and you have a good intuition about what’s happening behind the scenes, you can legitimately not look at the code. Then there’s domains like you’re also operating in, which is writing a Linux distribution. There, maybe you need to look at the code a little bit more. Then there’s maybe safety critical systems all the way down to operating a nuclear power plant or a self-driving car. Maybe you need to look at the code more carefully.

DHH

是，但话说回来，AI 找漏洞和修安全问题都强得离谱。这就是围绕 Fable 的那场风波：这个模型找攻击者能利用的洞太强了，根本不安全发布。讽刺的是，我们似乎到了几乎没有人类能匹敌的智能水平，因为很多安全洞是把连招串起来：这儿一个小漏洞本身未必最糟，再和另外四个一组合，突然就有 RCE——远程命令执行。能这么干的人很少，通常在国家支持的组织或其他秘密行动里。Linux 发行版是我 100% 信 AI 的原因。过去三个月我在做 Omarchy，几天前刚出的版本叫 Quattro。几乎从一开始智能体加速就接近 100%，过去两个月是 100%。我没有亲手写过 Quattro 里任何已出货的代码。我审过整体形状，审过系统模型层任何关键处的逐行，很多 UI 代码我没看，很多辅助代码没看，新功能也没有完全用手写过。

Yes, but that said, AI is insanely capable at both finding and fixing security vulnerabilities. This was the whole blowup about Fable. This model was so capable of finding holes that an attacker could exploit that it was simply not safe to release. So the irony here is that when you look at that field, it seems like we’ve reached levels of intelligence that virtually no human can match because many of these security holes are about stringing combo moves together. You find one little vulnerability here that by itself might not be the worst thing in the world, but then you combine it with four others, and suddenly you have RCE, remote command execution. Humans who are able to do that are very rare. The Linux distribution is why I’ve gotten 100% AI pilled. Because I’ve been working on Omarchy for the last three months, this version that just dropped a few days ago called Quattro. And almost right from the beginning I’m working on that version, the agent acceleration neared 100%, and in the last two months it has been 100%. I have not written any of the code that’s shipped in Quattro by hand. I’ve reviewed the shape of all of it. I’ve reviewed the individual lines of anything that’s critical in the model layer of the system, and I’ve not looked at a bunch of the UI code. I have not looked at a bunch of the auxiliary code, and I’ve not written any of the new functionality entirely by hand.

但 Web 这边——真正演进 Basecamp 和 Hey，有大量用户、相对大的代码库——用智能体全面加速出奇地难。我们不久前刚发 Basecamp 5，那是 37signals 第一个真正被智能体加速的产品。大约从二月起我们在最后冲刺，那时智能体已经不错。有一阵早期冲劲：「搞定了，让设计师编程就行，他们知道要什么功能、什么形状，让他们 vibe。」我们让他们 vibe 了。结果一堆 PR，单独看也许一时还能辩解，合起来却毁了系统架构。我们只好人手清理，用人手收拾，才回到感觉内聚、连贯的架构。那是二月。现在很不一样了。

But then the web part, actually evolving Basecamp and Hey, our professional products that have lots of users and are relatively large code bases, have proven surprisingly tricky to fully accelerate with agents. We just released Basecamp 5 not too long ago. That was the first product at 37signals that was really agent accelerated. Because we were in this final sprint phase from around February. By then, agents were already good. And we had this early surge of, “It’s solved. We can just have the designers do the programming. They know what features they want. They know what shape they want it to take. Let them vibe.” And we let them vibe. And we ended up with a lot of PRs that individually perhaps could have been justified for a hot moment, but taken all together, destroyed the architecture of the system. And we actually had to clean up manually, mop it up by hand, by human hand, to get back to an architecture that felt cohesive and coherent. So we still have a bit of that — that was February, by the way. Things are quite different now.

Lex Fridman

等一下。教训是什么？是不是现阶段要 vibe coding，你得先是程序员？

Well, hold on a second. What’s the lesson from that? Is one of the lessons from that that you have to be a programmer at this stage to be able to vibe code?

DHH

要在已有的、相当大的代码库上 vibe coding——即便是 CRUD——如果你还想保留当初把系统带到这儿的那份架构感，是的。人们骂 vibe coder 是 slop 生成器时，我会回敬：「你看过平均程序员的产出吗？」那也是另一种 slop。很多大公司代码库过了三千人之后，简直可怕。

To be able to vibe code on existing substantial code bases, even if they’re CRUD, if you wanna retain the element of architecture that got that system to where it was. Now, that’s also a point where I’ve stressed many times that when people accuse vibe coders of being slob generators, I go right back at them and say, “Have you looked at the average programmer’s output?” That is some other slob too. If you’ve looked at behind the scenes of many great companies and what their code bases look like after there’s been 3,000 humans through them, they’re awful. Absolutely awful.

Lex Fridman

你能讲讲你的直觉吗？往后退一步看我们依赖的那些应用——Adobe Photoshop 之类——那边的开发进度似乎没有加速。从 Basecamp 这种用户基数巨大的成熟产品能抽出什么教训？为什么我们没在这些老牌应用里看到更新、功能的超级快速增长？

Can you tell me the intuition you have? Just stepping back and observing the different apps that we all rely on — Adobe Photoshop, all this kind of stuff — there seems to be… the progress on development there has not accelerated. So why is it, what lessons can you draw from Basecamp, like well-established, huge user base? Why aren’t we seeing super rapid increase in new updates, features, in these well-established apps?

DHH

多重原因。最关键的第一个：一旦人类团队一起干活，瓶颈很少是实现，而是人的带宽和沟通。有产品经理、几个设计师、上面还有 VP、再上面还有 CTO，人人都想参与 shaping，因为我们都在证明自己为什么在这儿——生产力就死在那儿。过去三个月做 Omarchy 给我的启示是：要拿到那神奇的 10 倍、100 倍，少数情况 1000 倍生产力提升，你必须直接和智能体交互，不能再用人来中转带宽，因为太慢了。

Multiple reasons. I’ll start with the first one that’s the most critical. As soon as you’re having human teams work together on something, the bottleneck is rarely implementation. It’s human bandwidth and communication. When you have a product manager and a couple of designers and a VP above them and a CTO above them, and everyone wants to be part of the shaping process because we’re all justifying why we’re here, that’s where all the productivity goes to die. The revelation I’ve had working on Omarchy the last three months is that to get that magical 10X, 100X, in a few rare cases, 1000X productivity boost, you have to interact with the agents directly, and you cannot intermediate that bandwidth with another human because it’s simply too slow.

一方面有点扫兴——我喜欢人，一起干活很好。但也意味着如果由人驱动、还有三层审批和大公司那套机器，我们对智能体能做什么要压低预期。实现只是一小段。另一点：多数组织不知道自己要什么，不知道怎么让它更好。他们不卡在实现，卡在想法、愿景、品味。如果你那些要素没有超过实现能力，加实现也没用。你可以让很多烂想法成真，然后呢？要发货吗？听起来像——我不知道——微软出的什么东西。不是我们想复制的。

And on the one hand, that’s a bit of a bummer. I mean, I like humans, and it’s great to work together. But it also means we need to temper our expectations with what these agents can do if it’s humans driving it, and you have three layers of approval and all the other machinery of a large corporation. The implementation part is only a small segment of it. The other thing I’d say is that most organizations don’t know what they want. They don’t know how to make it better. They’re not bottlenecked on implementation. They’re bottlenecked on ideas. They’re bottlenecked on vision. They’re bottlenecked on taste. And if you don’t have those elements in excess of your implementational capacity, it doesn’t help. So you can make a lot of shitty ideas come true, then what? Are you gonna ship that? That sounds like, I don’t know, something coming out of Microsoft. That’s not what we’re trying to replicate here.

我们已经有过这个。想想那些拥有成千上万程序员的巨头组织。我拿微软开刀——有时我爱微软，但这回要挑——他们有几十年无尽的资源和编程产能。这恰恰说明：能写很多代码，并不产生伟大、有说服力的软件。另外，我们大概有六个月这种能力。在人类生命周期里，内化正在发生的事，这不长。有趣的是对 AI 的批评是：为什么不更快？你在说什么？我们没见过任何别的进步像 AI 这么快，你却因为过去三个月没重写整个世界、没把软件乌托邦建起来而不耐烦？

Because we already had this. If you step back for a moment and think of the towering organizations we have who have had tens of thousands of programmers at their disposal. I’m picking on Microsoft here. I love Microsoft some of the time, but I’ll pick on them in this case because they have had endless resources, endless programming capacity for decades. This is what’s showing us that just being able to write a lot of code does not produce great compelling software. Now, the other thing is we’ve had this capacity for about six months. That’s not very long in human life cycle capacity of internalizing what’s going on. And I think it’s actually funny that the critique of AI is, why isn’t it going faster? What are you talking about? We’ve not had any other form of progress that has moved as fast as AI, and you’re impatient because in the last three months we haven’t rewritten the whole world and made it a utopia of software goodness?

Lex Fridman

你基本上是说大家现在都该切到 Linux。一个论点是：我们可以把 Linux 上没有的软件都在 Linux 上重写。Linux 是我多年的挚爱，但我还粘着 Windows、现在还有 Mac，是因为视频剪辑、Premiere。谁来给 Linux 做 Premiere 和 Photoshop？感觉现在一个人就能做。

You’re basically saying everybody should be switching to Linux now. One of the arguments is, like, we can rewrite all the software that’s not available in Linux in Linux. Linux has been the love of my life for many years, but one of the reasons I’m still attached to Windows and now Mac is because of video editing, Premiere. So the question is, who’s gonna build Premiere and Photoshop for Linux? And it feels like one person can now.

DHH

百分之百，一个人能。

100% one person can.

这是经典的创新者窘境。这些公司在旧方式上太好、太稳，整个结构、管理层级、流程都调到一个已不存在的时代。你转不动。它们是超级油轮。这就是为什么科技业终于有了这场颠覆。曾有一阵我很烦 Apple 和 Google 在移动端的双寡头——移动似乎是最重要的计算平台，我看不到怎么把它们从顶上掀下去。但游戏变了。这不再是最重要的平台。手机还重要，但也有眼镜、耳机等别的形态进来。一切都在玩。桌面计算平台本身大概四十年来第一次真正在玩。Linux 从 91 年就在，桌面没起飞、没拿下；它拿下了其他一切。你桌上那些设备、冰箱、烤面包机，除了你的电脑都跑 Linux。有趣的是 Android 其实是 Linux，包得太严认不出来。现在有开口了。

This is the classic innovator’s dilemma. These companies have gotten so good, so established at the old way, and therefore their entire structure, management layers, processes are tuned for a time that no longer exists. But you can’t pivot that. These are super tankers. It just doesn’t happen. This is why we’re finally getting this upset in the technology industry. For a while, I was very upset about the duopoly between Apple and Google on mobile, because it felt like mobile was the most important computing platform that we had, and I could not see a path to unseat either Google or Apple. But the game has changed. This is no longer the most important platform. Mobile phones are important, but there’s also a lot of other form factors coming, whether it’s glasses or it’s earpieces or whatever else have you. It’s all in play now, but the computing platforms themselves are in play for the first time in probably 40 years, if you look at the desktop. Linux has been around since ’91. It’s not taken off or taken over on the desktop. It’s taken over everything else. All the devices you have on your desk, your fridge, your toaster, everything runs Linux except for your computer. Now, funnily enough, your Android is actually Linux, but wrapped so sufficiently that you can’t really recognize it. But now there’s an opening.

如果你依赖的软件把你绑在 Windows 上，你个人完全有能力开始重写。也许到不了 100% 覆盖，但这是关于 Microsoft Office 的老笑话：「我只用 5%。」对，但我们各自用不同的 5%。如果我们各自只建自己需要的那 5% 呢？那是完全不同的挑战，智能体今天就能极其胜任，我做过很多次。最新智能体参与的一个惊人发现是：我成了多语言程序员，以前绝对不是。我首先是 Ruby 程序员，必要时沾一点 Bash。过去两个月我写了 C++。我写了三个已随 Omarchy Quattro 出货的应用。我写了个写作应用。我用过 Typora——很好的应用——它本身又基于我在 Mac 上真正爱的 iA Writer，干净简单的 Markdown 写作环境，我所有文章都在那儿写。

If there’s a piece of software you depended on that bound you to Windows, it is completely within reach for you personally to start rewriting it. And maybe you won’t get to 100% coverage, but this is the old joke about Microsoft Office. “I only use 5%.” Yeah, well, we all use a different 5%. Well, what if we all just build our own 5%? What if I just took the functionality that I need and just did that? That is a completely different challenge and one agents are incredibly capable of doing right now today, and I’ve done it a lot of times over. So one of the amazing things I’ve found with the latest agent engaged is I have become a polyglot programmer, something I was absolutely not before. I was a Ruby programmer first and foremost, and then I dabbled a little bit in Bash when I had to. And in the last two months, I’ve written C++. I’ve written three applications that have shipped in Omarchy Quattro. I wrote a writing app. I was using this app called Typora, which is a very nice app, which in itself is based on another app called iA Writer, which was my real love on the Mac for a clean, simple Markdown writing environment. That’s where I write all my essays.

然后我迁到 Linux，拿不到 iA Writer，就换到 Typora。它是共享软件，有很多我不需要的功能。大约六七周前我想：「你知道吗？我只需要 Typora 的 5%，而它已经是个基础应用。」我字面告诉智能体开工，要用 C++ 和 Qt 写，因为那契合我在 Quattro 上做的美学。我想大概 20 分钟就有了第一版。我开始用，还不完美。两天内我放弃了 Typora，从那一刻起所有文章都在 Omawrite 里写。

And then I moved to Linux, and I couldn’t get iA Writer, so I moved over to this other tool called Typora. It was a piece of shareware, and it had a lot of features I just didn’t need. And about six weeks, seven weeks ago, I thought, “Do you know what? I only need 5% of Typora,” which is already a basic app. I literally told the agent to get going, that I wanted it written in C++ and Qt, because that would fit well with the aesthetic of what I was building with Quattro. And in, I think, about 20 minutes, it had the first version. I started using it, and it wasn’t quite right. Within two days, I’d given up Typora, and I wrote and have written all of my essays since that moment in Omawrite.

Lex Fridman

我能问你对一件事的建议和愿景吗？用智能体你可以写出像 Typora 替代品那样完全为你定制、只为一个用户着想的东西。然后有一种版本从为一个用户起步，扩到更大受众。我写过大量只给自己用的软件，觉得更容易、更好地解决自己的情况。比如我写了个视频编辑器，只给自己用很容易。不像你，我从没做过很多人用的东西，觉得有点吓人。智能体时代，怎么从解决自己问题——我认为这是很美的建造方式——再扩到对别人也有用？你也看到「太容易只给自己做工具」这个问题吗？

Can I ask for your advice and your vision about something? So what I find with agents, you can actually write something like a replacement for Typora that is perfectly customized to you, to your needs, with just one user in mind. And then there’s a version of that that starts with one user in mind but expands to a larger audience. I’ve actually written a large number of software just for me. I’ve never, unlike you, built something that’s used by a large number of people, and I find that a bit scary and intimidating. So with this agentic age, how do you take a step to actually build something that solves your problem, which I think is a beautiful way to build, but also expand it to actually useful for other people? Do you also see the problem of how easy it is to just build a tool for yourself?

DHH

看到，但比你想的简单。你给自己做完工具时，只要告诉智能体把它放到 GitHub。你甚至不用做别的。它会弄清怎么把仓库放到 GitHub，写漂亮的 README，开始用 GitHub releases 追踪。它实际上会是比你更好的软件维护者——因为它更有耐心，做开源管理那套苦差远比你勤勉。你可以继续享受用自己造的软件、按自己的想法演进。这进入我们开源界吵了很久的大争论：AI 贡献对维护者是好是坏？很多维护者现在很恼火，突然涌入大量 PR，来自可能不是最好的程序员、甚至根本不是程序员的人。我听着这个争论心想：「你开什么玩笑？」这儿有一条免费贡献的矿脉，你可以拿或不拿，你却在抱怨它们在那儿？听起来像「我的牛排太多汁、龙虾黄油太多」。你在抱怨什么？

Yes, but it is simpler than you think. By the time you’re done building the tool for yourself, you simply tell your agent to put it on GitHub. You don’t even have to do anything else. It will figure out how to put that repo on GitHub. It’ll write a nice README. It’ll start using GitHub releases so that you can track things. It’ll actually be a better software maintainer than you could ever be — because it is far more patient, it is far more diligent in doing the drudgery of managing open source software than you are. And you can simply get on with the joy of using the software that you built and evolving it as you see fit. Now, this gets into the big argument we’ve been having in open source for a while. Are AI contributions good or bad for the maintainer? There’s a lot of maintainers right now who are quite upset about the fact that they’re suddenly getting a huge influx of pull requests from people who may not be the best programmers or programmers at all. I look at that argument and go, “Are you kidding me?” Here is a vein of free contributions that you can take or don’t take, but you’re complaining about the fact that they’re there? It sounds straight out of — My steak is too juicy and my lobster too buttery. What are you complaining about?

这是最惊人的事。我们鼓吹了这么久的开源魔法——大家都能贡献——当然从来不是真的。只有一小撮人贡献开源，那是高技能巫师。这有点像宗教改革时刻。我们和计算机之间有中介，有一类叫程序员的教士和祭司，现在他们被中介掉了——像不像 1500 年代路德把九十五条论纲钉在门上？其中一条很清楚是关于中介：你应该直接接触更高的智能。

This is the most amazing thing. This is what we proclamated for so long was going to be the magic of open source, that we could all contribute to it, but it was, of course, never true. There was only a small subset of people who contribute to open source, and that was the very highly skilled wizards. So this is a bit of a reformation moment. We had this disintermediation between us and the computer, and it was this class of clerics and priests called programmers, and suddenly they’re being disintermediated here by — was it the 95 theses by Luther in 1500s, nailing them to the door? And one of those was very clearly about the intermediation, that you should have direct access to the higher intelligence.

Lex Fridman

但你不需要巫师来维持代码库的高卓越门槛吗？

But don’t you need the wizards to keep a high bar of excellence in the code base?

DHH

百分之百需要，这也是开源一直为真的事，也是第二个让我磨牙的论点。我跑开源项目 25 年了。我字面上看过成千上万、甚至好几万程序员的产出。我觉得我有统计基础断言：大多数程序员，他们很烂。我不是按听起来那种意思说——我停了一下。我是说他们写不出我想要的代码。他们不把 bug 报告准备齐相关信息，不在 PR 里写清为什么，懒得补需要的注释，不复查，不写单元测试，不做做出好软件该做的那些事。但你知道谁会做这些吗？智能体，如果你告诉它们。它们很勤勉地跟指令，多数时候，有时过了头。这意味着如果你拿中位程序员对平均开源项目的 PR，他们已经被智能体甩开了。

100% you do, which is also something that’s always been true about open source, which is the second argument that really grinds me. I’ve been running open source projects for 25 years. I have literally looked at the output of thousands, if not tens of thousands of programmers. I feel like I have the statistical basis to assert that most programmers, they suck. And I don’t mean that in the way it sounds. That was why I paused for a hot second here. I mean, they suck in the sense that they don’t write the code I want to have written. They don’t prepare their bug reports with all the relevant information. They don’t detail their pull requests with the why. They don’t bother to fill in needed code comments. They don’t double-check their work. They don’t write unit tests. They don’t do all of these things it takes to create good software. But do you know who’ll do all that stuff? Agents, if you tell them to. They’re very diligent at following your instructions, most of the time, sometimes to a fault. But it actually means that if you take the median programmer and their pull requests towards an average open source project, they’re already getting outclassed by agents.

我宁愿收到智能体写的 PR，也不愿收到人写的，不只因为质量更好，也因为我拒绝时没那么内疚。你都没写，我可以看你智能体代写的东西说「嗯，不要」。开源一直如此。开源维护者的问题之一是神经质和焦虑太多，把任何贡献都当成自己有义务把它弄进代码库。根本不是。你可以直接说「不」，更好是「谢谢，不要」。如果你进入那种模式——项目可以按你的愿景、你的路线图存在和演进——在智能体时代拒绝不想要的贡献就容易多了，因为你甚至不麻烦、不伤害人的感情。只是个 clanker，clanker 不会介意。事实上造 clanker 的实验室喜欢你白花 token，一样收钱。所以在我看来，这是有史以来当开源维护者最好的时刻。

I would rather get an agent-written pull request to one of my projects than I’d get one written by a human, and it’s not just because the quality’s better. It’s also because I feel a lot less bad if I just reject it. You didn’t even write it, so I can simply look at what your agent wrote on your behalf and go, “Eh, don’t want it.” That’s always been true in open source, and in fact, I think this is one of the problems with open source maintainers. They have way too much ball of neuroticism and anxiety that requires them to look upon any contribution as an obligation on their part to do everything to get that to land in the code base. That’s not true at all. You can simply say, “No,” or even better, you can say, “No thank you.” And if you get into the mode of realizing that your project is allowed to exist and evolve according to your vision, according to your roadmap, it becomes so much easier in the agentic age to decline the contributions you don’t want because you don’t even inconvenience or hurt the feelings of a human. It’s just a clanker, and the clanker won’t mind. In fact, the producers of clankers, the labs, love when you spend tokens in vain. They get paid just the same. So, in my opinion, this is the absolute best time to ever have been an open source software maintainer.

我们不但得到一堆由智能体写好、把框都勾上的光荣 PR，还能接入以前无法贡献的人的创造力。Omarchy 项目我跑了一年多一点，过去三个月做 Quattro，我合并了超过 1000 个 PR。其中相当多是由非经典程序员写的，或是其他领域的程序员，不是 Linux 操作系统或发行版开发。他们能靠智能体贡献好想法，我得以精选最好的——因为它们在那儿。否则那些想法只会活在他们脑子里。开源的目的不就是接入整个该死星球的集体智能和创造力，导向我们都能受益的公地吗？现在 Omarchy 上大约有 400 个未合并 PR，大约是一周前的两倍。加速是真的，工具也是。我不再审每一个 PR，很久没了。我让智能体替我审，然后给我摘要：是否准备好做人类决定——合还是不合。

Not only do we get this wealth of glorious pull requests made by agents with all the boxes ticked, we also get to tap into a creativity of people who did not have access to contribute to a project before. I’ve been running the Omarchy project now for a little over a year, and in the last three months working on Quattro, I have merged over 1,000 pull requests. Quite a lot of those pull requests were written by people who were not classical programmers, or were programmers in other domains, not Linux operating system or distribution development. They were able to contribute their good ideas because of agents, and I got to cherry-pick the very best of that — because it was there. Otherwise, those ideas would just have lived inside their heads. Again, is that not the purpose of open source, that we tap into the collective intelligence and creativity of the whole goddamn planet, and we channel that towards a commons where we all benefit from the combined efforts of everyone? Now, there’s currently, I think, about 400 unmerged pull requests on Omarchy, about double what it was a week ago. So the acceleration is real, but so are the tools. I’m not reviewing every pull request anymore. I haven’t been reviewing them for quite some time now. I have the agents review them for me — and then they will give me a summary about whether something is ready for the decision, the human decision. Should we merge or should we not merge?

我不需要看所有错误、重复、糟糕的 PR。智能体可以把谷壳筛掉。我只看珍珠——准备好的好想法、智能体已在 VM 里替我验证过的 bug 修复。复制和管开源项目的苦差正在以闪电速度蒸发，我们留下金色多汁的部分，软件开发的骨髓——决定这东西该做什么、该往哪走。

I don’t need to look at all the pull requests that are either wrong, duplicated, bad. An agent can sort that chaff away from me. I just get to look at the pearls — the good ideas that are ready to go, the bug fixes that the agent has validated in a VM on my behalf. All the drudgery of replicating and managing open source projects is evaporating at lightning speed, and we are left with the golden juicy parts, the bone marrow of software development — deciding what should this thing do and where should it go?

Lex Fridman

你觉得好想法的内核最终还是源于人类头脑吗？那些进来的如何改进 Omarchy 的 PR——还是可以全是一池智能体？

Do you think the good ideas, ultimately the kernel of the good ideas originated in the human mind, so the individual contributors — can it be done by agents, or is the PR ultimately requires a human idea? Like how to improve Omarchy, all those incoming… Or can it all just be a pool of agents?

DHH

那是我在第一个智能体时代——从 11 月 24 日到 2 月 28 日——想智能体的方式。我以为所有想法都源于人，他们告诉智能体建什么，然后出发。我完全不再这么想了。我见过你们不会相信的东西，模型里冒出的想法好到让我谦卑——作为一个否则以有好想法为傲的人。智能体极有能力做创造性思维；还卡在「智能体是鹦鹉、只是反刍已有想法」分析里的人，对过去六到九个月的进步是妄想。

That was the mode and the way I was thinking about agents in the first agentic age from November 24 to February 28. I thought all the ideas would originate with the humans. They would tell their agents what to build, and off they went. I don’t think that’s true anymore at all. I have seen things you people wouldn’t believe, ideas coming out of models so great that it makes me humble — as a person who otherwise prides himself on having good ideas. The agents are incredibly capable of creative thought, and folks who are still stuck in the analysis that agents are parrots just regurgitating the ideas that are already there, are delusional about the progress that’s been made in the last six to nine months.

Lex Fridman

有人听你现在说话会说：DHH 得了老一套的 AI 精神病。你能钢人论证一下——你其实处于妄想状态，像《飞越疯人院》——然后再反驳吗？

So some people listening to you right now will say DHH is suffering from the old case of AI psychosis. Can you steel man the case that you are in fact experiencing in a state of delusion, like One Flew Over the Cuckoo’s Nest, and can you argue against it?

DHH

我处于谵妄状态。那才是我的状态。因为我和计算机打了 40 年交道，过去两个月里看到的东西，这辈子从没见过。整个生涯、整个生命把对计算机的爱当作工作时间的主焦点，突然被彻底掀翻、彻底改写。我当然谵妄。看着这个现实谁不谵妄？如果你不谵妄，或至少非常兴奋……其实也不对。如果你不承认这一刻的分量，那才是妄想，那才是精神病。精神病是相信世界几乎没变，我们只是有些电子鹦鹉在背诵。在我看来，这个论点被提出，是因为人们看不到进步的果实。你前面问的：了不起的软件在哪？为什么美国 GDP 没跑到同比 12%？第一，给它一分钟。连一年都还没有，该死的。第二，就我自己而言，我其实端出了布丁。Omarchy Quattro 周五上线，已被成千上万人下载。他们喜欢。很喜欢。

I’m in a state of delirium. That’s the state I’m in. Because I have been working with computers for 40 years, and I’ve never seen the things that I’ve seen in just the last two months. A whole career, a whole life dedicating to the love of computers as my main focus for the working hours and then some, suddenly completely upended, suddenly completely rewritten. Of course I’m delirious. Who looking at this reality is not? If you are not delirious or at the very least very excited… Actually, I guess that’s not even true. If you’re not recognizing the gravity of the moment, that’s the delusion. That’s the psychosis. The psychosis is believing that the world is barely different. We just have some electronic parrots reciting things to us. Now, in my opinion, the reason this argument is actually brought up is because people are not able to see the fruits of the progress. That goes to your point earlier. Where is the amazing software? Why isn’t American GDP running at 12% year over year? First of all, give it a minute. It hasn’t even been a year, goddammit. But second of all, I feel on my own account, I have actually brought the pudding. Omarchy Quattro launched on Friday. It’s been downloaded by tens of thousands of people. And they like it. They like it a lot.

Lex Fridman

该提一下——查 Perplexity——Omarchy 是高度有主见的、基于 Arch Linux 的桌面发行版，以 Hyprland Wayland 平铺合成器为中心。由 David Heinemeier Hansson，也就是 DHH 创建，作为打磨过的开发者工作站配置，有统一视觉风格，很多日常工具已配好。

We should mention that going to Perplexity here, Omarchy is a highly opinionated Arch Linux-based desktop distribution centered on the Hyprland Wayland tiling compositor. It was created by David Heinemeier Hansson, DHH, as a polished developer workstation set up with a cohesive visual style and many everyday tools already configured.

DHH

让我这么说。Omarchy 是漂亮、现代、有主见的 Linux 系统。它是 macOS 和 Windows 的替代操作系统，而且他妈的很棒。

Let me put it. Omarchy is a beautiful, modern, and opinionated Linux system. It’s an alternative operating system to macOS and Windows, and it’s freaking amazing.

Quattro 是我们刚放出的最新版。有趣的是 Omarchy 项目才一年多一点。我去年夏天在勒芒 24 小时耐力赛场次之间开始的——看了太多 Linux Ricing 的 YouTube，被咬了一口。开始做第二次尝试：更好的 Linux 操作系统。上次我们谈过的第一版叫 Omakub，建在 Ubuntu 上，还行。但因为往栈下走了七层，我能倒进新版 Omarchy 的野心完全不同。不过那还是前智能体世界。开头那些 bash 脚本全是我手写的，然后我看到中途转折。我做了几个部分智能体加速的 Omarchy 版本，三个月前全面油门，100%，一切由智能体写、由我掌舵。那成了非常不同的体验和非常不同的操作系统，因为我突然被授予野心上的无限天花板。我可以看 Windows、Mac、别的 Linux 系统上任何功能说「我要那个」，智能体就会交付。我想要的一切突然够得着，于是我有点谵妄。如果瓶子里突然跳出精灵说「你要什么都有」，谁不会谵妄？

That’s the latest version that we just dropped out, which is also funny. The Omarchy project is only a year old and a little bit. I started it last summer in between sessions at the 24 Hours of Le Mans — where I had watched one too many YouTube videos on Linux Ricing and got bitten by that bug. Started working on this second iteration of my attempt at creating a better Linux operating system. The first iteration we talked about last time was called Omakub, was built on top of an existing system called Ubuntu, and it was fine. But the ambition I was able to pour into the new version, Omarchy, because I started seven layers deeper down the stack, was completely different. But it was still done in a pre-agentic world. I wrote all those bash scripts by hand in the beginning, and then I got to see the midway point where things changed over. I did a couple of versions of Omarchy that had partial agent acceleration, and then three months ago, it was full throttle, 100%, everything is written by agents but steered by me. And that just ended up being a very different experience and a very different operating system because I was suddenly granted a limitless ceiling on my ambition. I could look at any feature in Windows, on Mac, on other Linux systems and say, “I want that,” and the agents would deliver. So everything I wanted was suddenly within reach, which meant that I got a little delirious. Who would not get delirious if suddenly a genie pops out of the bottle and says, “You can have whatever you want”?

我没有那种压倒感、恐惧感、扭曲感，因为我有使命。我要去某个地方，因此可以把所有这些新力量导向单一目标和结果：创造完美的计算机。因此我对 AI 的所有调查都朝那个终点。我也有正职，我们也在那儿用智能体，也对准具体结果。但凭 Omarchy 体验，我直接插进后脑勺，把脑子里冒出想法和屏幕上冒出软件之间的带宽加大了。像从拨号上网到光纤。我想在你播客上 Elon 谈过 Neuralink 等项目试图加速的人类带宽。我们现在对话的带宽很低，因为受限于认知、受限于语速。你和一群智能体工作时不受那些限制。

I don’t have that sense of overwhelm. I don’t have the sense of dread. I don’t have the sense of distortion because I have a mission. I’m going somewhere, and therefore, I can channel all this new power towards a singular goal and outcome, creating the perfect computer. And therefore, all my investigations with AI are focused towards that end. Now, I also have a day job. And we also use agents there, and that’s also targeted towards sort of specific outcomes. But with the Omarchy experience, I got to tap in straight in the back of my head and increase the bandwidth between ideas arriving in my brain and software emerging on the screen. It was like going from dial-up to fiber. I think on your podcast, Elon talked about the actual human bandwidth that Neuralink and other projects are trying to accelerate. Like, the bandwidth we’re communicating over right now is quite low because we’re limited by our cognition. We’re limited by the rate of speech. You’re not limited by those things when you’re working with a swarm of agents.

让我加个免责：如果九个月前听到自己这样说话，我大概会贴上 AI 精神病的标签。会很贴切，因为九个月前这么说的人没在发货，我认为那是最终差别。有人看得早。我其实没那么早。我们谈过，我相当怀疑，我在某些方式用 AI，但我不是第一个下载 Claude Code 的人。我想 Boris 是二月底发的，我大概到九月才装 Claude Code。所以大约有六个月，先驱看到未来会是什么样的微光。还没到，因为智能不在、harness 不在，但他们能看见微光。我看不见。Tobi——Tobi Lütke，Shopify 的 CEO，现在也是 Omarchy 项目上我的同伙——他也信了 Omarchy。他很早就看到这些，试图告诉我，我没看见。

Now, let me caveat this by saying, if I’d heard myself talk like this nine months ago, I would probably have used the label AI psychosis. I think it would be fitting because nine months ago when people were talking like this, they weren’t shipping, and I think that’s the ultimate difference. There were people who saw this early. I was not actually that early. As we’ve talked about, I was rather skeptical, and I was using AI in certain ways, but I was not the first person who downloaded Claude Code. I think Boris released that in end of February. I don’t think I installed Claude Code until September. So there was about six months there where pioneers saw these glimmers of what the future was gonna look like. It wasn’t there yet because the intelligence wasn’t there. The harnesses weren’t there, but they could see the glimmers. I couldn’t. Tobi, Tobi Lutke, the CEO of Shopify and now my partner in crime on this Omarchy project in part, he’s gotten pilled with Omarchy as well. He saw these things very early, and he tried to tell me, and I wasn’t seeing it. I was not seeing what he was seeing.

Lex Fridman

你是个黑子。

You were a hater.

DHH

是。我还在地球上。而他……已经登了火箭往太空去了。我记得读他给公司的备忘录——那份 AI 备忘录——心想：「啊，似乎有点过。」你把我认为还摸不到的东西说成大事。因为产出在哪？我试着用，它写我不喜欢的代码，老打断我。你从哪看出这个？所以我完全理解。如果你没见过当前智能质量能产出什么，当然会看别人觉得「他们听起来有点疯」，因为你至此所有经验都会告诉你他们疯了。过去四五十年计算机编程史上，人们一直承诺普通人只要说话就能写代码。我们要有第四代语言。Lisp、Smalltalk 曾被说成让普通人写自己应用的环境，除了像 Microsoft Access 数据库或 Excel 电子表格那种，都没成真。那些是终端用户编程环境，但完全不像我们现在能做的。我甚至不是在谈终端用户——我在谈我，编程 25 年的人，看到产出、加速、质量，然后发货。我认为那会很快改变整个对话。还在观望这是否真有用、还困在 2025 年初机械鹦鹉迷因里的少数落后者，很快就会被即将淹没他们的证据压垮。

Yes. I was still on Earth. And he was… He had already boarded the rocket and was heading towards space. I remember actually reading his memo to the company, the AI memo, and thinking, “Ah, seems a little much.” Like, you’re making a big deal out of something that I cannot yet feel is a tangible thing. Because where’s the output of this? I try using it. It writes code I don’t like. It wants to interrupt me all the time. Where are you getting this from? So I get it. I totally get it. If you’ve not seen what the current quality of intelligence can produce, of course you’re gonna look at someone and go like, “They sound a little nutty,” because all your experience up until this point would tell you that they are nutty. Because if you look at the history of computer programming, for the last 40, 50 years, people have been promising that regular people could write code just by talking. We’re gonna have fourth generation languages. Lisp was once upon a time presented, Smalltalk was presented as these environments that would let regular people write their own applications, and none of it came true in the sense beyond, like, Microsoft Access databases or Excel spreadsheets. Those are end user programming environments, but it’s nothing like what we’re able to do now. And then, I mean, I’m not even talking about end user. I’m talking as me, someone who’s been programming for 25 years, seeing the output and the acceleration and the quality and then shipping it, and I think that’s what’s gonna change the entire conversation quite quickly. The few laggards who are still on the fence about whether this is actually useful, who are still trapped in a meme from early ’25 about mechanical parrots, they’re simply going to be overwhelmed by the evidence that’s about to flood over them.

Lex Fridman

我希望看到像 Omarchy 这样的项目，好几个真正展示 agent-first 的。在视频剪辑、Typora 这类应用那些「傻」空间里。

So I hope to see projects like Omarchy, several of them that really show agent first. In the spaces, in the silly spaces that I already mentioned, like video editing, all these kinds of apps that are like Typora — this kind of stuff.

DHH

会来的。全都在来。Omarchy 一个很 neat 的功能是更稳健的插件系统，让你能扩展操作系统，真正改写用户界面、面板和功能——通过创建自己的插件，可以从我们装箱的工具克隆。比如你想要不同的日历。Omarchy 带日历，点小钟弹出日历，它没有 iCal 支持，不消费你的约会。那种实现已经大约有 17 个，三天内我们在 Omarchy 插件市场上有了 330 个插件。我参与过的任何项目从没见过那样的增长。从没见过参与面那么广。从没见过那么多人能如此快地造出对他们有意义、也对别人可用的软件。全是因为 Omarchy 装了一套 skills，告诉你带来的任何智能体如何给操作系统做扩展，因此给你真正可塑的……我正要说 agentic。我他妈恨那个词。部分原因是它已成营销 slop 用语，甩到一切上。我希望有个不同的词，就表示「AI 在做事」。

It’s coming. All of it is coming. One of the really neat features about Omarchy is that it has a far more robust plugin system that allows you to extend your operating system and rewrite it really in terms of its user interface and its panels and its features by creating your own plugins that can be cloned off of tools we ship in the box. For example, say you want a different calendar. Omarchy ships with a calendar. You click the little clock. It pops up a calendar, and it doesn’t have iCal support, for example. It doesn’t consume your appointments. There’s about 17 implementations of that already, and in three days, we had 330 plugins on the Omarchy plugin marketplace. I have never seen growth like that with any project I’ve ever been involved in. I’ve never seen participation that broadly. I’ve never seen that many people be able to create software that’s meaningful for them — also be usable for others so quickly. All of it is driven by the fact that Omarchy ships a set of skills that tells any agent you bring to it how to create extensions to the operating system, and therefore affording you the vision of the true malleable… I was about to say agentic. I hate that fucking word. And the reason in part I hate that word is, first of all, it’s become marketing slop speak at this point. Like, it’s just slapped onto everything. I wish we had a different word that just meant AI doing stuff.

Lex Fridman

Vibe coding 感觉也不对。

Vibe coding feels wrong, too.

DHH

对，因为对我来说 vibe coding 闻起来完全像 2000 年代初的 script kiddies。人们套用刚从网上下载的 PHP 脚本，什么都不懂。某种意义上确实如此——很多 vibe coding，包括我自己 vibe 出来的项目，我刚提过有好几个。我用 C++ 写过东西，用 Rust 写过东西。Rust——我恨 Rust 恨得要死。看 Rust 代码对我就像往眼睛里倒酸。

It does, because vibe coding to me smells exactly like script kiddies did in the early 2000s. People applying PHP scripts they just downloaded offline that they don’t understand anything of. And while, I mean, that’s true in the sense that much of the vibe coding, including my own vibe coded projects, and as I just mentioned, I have several. I’ve written things in C++. I’ve written things in Rust. Rust — I hate Rust with a passion. Rust to me is like pouring acid in my eyes when I have to look at the code.

Lex Fridman

等等。你为什么恨 Rust？

Wait, wait, wait, wait. Why do you hate Rust?

DHH

在我看来，它大概是过去四十年发明的最丑的编程语言。是 Rails 和 Ruby 的反面。但我用 Rust 写过应用，因为 Rust 的产出其实惊人。内存安全、是系统语言、高效等等，难以置信。所以对我来说，Rust 可以同时是为人类消费设计的最令人厌恶的编程语言，也是智能体工程的奇妙平台。这两个真相很容易在我脑子里共存。

It is, in my opinion, the ugliest programming language that has been invented in probably the last 40 years. It’s the opposite of Rails and Ruby. But I’ve written applications in Rust because the output of Rust is actually amazing. The memory safety, the fact that it’s a system language, the fact that it’s highly efficient and so forth is incredible. So Rust, to me, can both be the most repugnant programming language ever devised for human consumption and a wonderful platform for agentic engineering. These two truths can coexist easily in my head.

Lex Fridman

你觉得会有一个点，我们可以直接把智能体工程叫作编程吗？几个月后谁还会用老派方式编程？

Do you think that there can be a point at which we can just call agentic engineering programming? Can we just change what programming means? ’Cause who’s actually going to be programming the old-school way in a few months?

DHH

我觉得不该复用这个词，因为对我来说，编程意味着理解某些原语、循环、条件、变量。编程语言的所有构造构成了编程本身，不是「程序的创造」。因为你可以说在前智能体时代，你的 CEO 也在编程——他雇一堆程序员，告诉他们做什么，出来能卖的软件。呃，多数人不会叫那个 CEO 程序员，我也不觉得该叫 vibe coder 程序员。如果我们在这儿定义 vibe coding：你告诉智能体给你建软件，你不看实现。对我来说，那就是 vibe coding 与编程——或者说智能体加速开发——的分界。

I don’t think we should reuse the term because for me, programming implies an understanding of certain primitives, loops, conditions, variables. All the constructs of programming languages are what constitutes programming itself, not the creation of programs. Because you could say in the pre-agentic era, well, your CEO is programming. He’s hiring a bunch of programmers, he’s telling them what to do, and out comes software that he can sell. Eh, I don’t think most people would call that CEO a programmer, and I don’t think we should call the vibe coder a programmer either. And vibe coding, if we define it here, is you tell an agent to build software for you. You do not look at the implementation. That, to me, is what separates vibe coding from programming or, let’s say, agent-accelerated development.

Lex Fridman

但推一下：你这个老派程序员做智能体工程，和非程序员做智能体工程，难道不是不同种类吗？如果你知道 for 循环是什么、函数式编程是什么、软件工程一些基本原则，你用自然语言做智能体工程的方式会不同、更系统，能建的规模和种类也比 vibe coder 大得多。

But don’t you think, so to push back a little bit, don’t you think a programmer, old-school programmer, you, doing agentic engineering is a different kind of agentic engineering than a non-programmer doing agentic engineering? What I mean is, if you know what a for loop is, if you know what function programming is, if you know some of the basic good principles of software engineering, the way you do natural language-based agentic engineering will be different and more systematic, and the scale and the variety of things you can actually build is much bigger than a vibe coder.

DHH

我只是为了辩论把正方再伸一点。我其实认为有一阵子，我知道那么多编程对我是劣势，因为我指示智能体按我规定的方式做，它们很擅长那样。在第一个智能体时刻——别叫纪元，大概持续了三个月——那感觉很有生产力。我能用程序员经验告诉智能体按我会做的方式做，拿到更多产出。然后我在下一个时刻晚了一步，那个时刻允许我、允许任何人描述结果、描述问题给智能体，得到比程序员规定路径更好的解。

I’m just gonna extend the pro case here a little bit just for the sake of the argument. I actually think for a while it was to my deficit to know as much as I know about programming because I was instructing the agents to do things as I prescribed them to do, and they were very good at that. And in the first agentic moment, let’s not call it an era, it was something that lasted three, three months. In the first agentic moment, that felt very productive. I could get more productivity out of telling the agents what to do in the way I would’ve done it, and I used all of my experience as a programmer to do just that. Then I was a little late on the next moment, and the next moment allowed me, allowed anyone, to describe outcomes — to describe problems to the agents and get better solutions than if you had a programmer prescribe the path.

Lex Fridman

所以你是说，有些问题程序员做智能体工程会比非程序员更差？

See, okay, so we’re having fun arguing here. Well, so are you suggesting that there’s problems for which programmers are worse than non-programmers at agentic engineering?

DHH

百分之百。原因是：很多程序员不是很好的产品经理。软件就是产品管理。它该做什么？为谁做？怎么做？长什么样？优先级？先从什么开始？第一版包含什么？那些技能不是平均、平等地分布在所有程序员身上。在让智能体、AI 做实现的智能体时代，你需要的是这些技能。我恰好坐在两个阵营，所以我爱这一刻。我还能活一点在旧世界，看完整实现，因为有些领域我仍觉得那有价值；我也活在未来。我说过，做 Omawrite 时，我一行 C++ 都没看。我特意把它当实验：这是 100% 黑箱。我会像任何对数字打字机有意见的用户那样对待它，因此这是一个公平实验——像任何对软件该怎么做、怎么工作有意见的写作者。

100%. And the reason I say that is there’s a lot of programmers who are not very good product managers. Software is product management. What should it do? Who should it do it for? How should it do it? How should it look? What’s our priorities? What do we start with first? What does version one include? All of those skills are not easily or equally distributed across all programmers. And in the agentic era where you are letting an agent, an AI, do the implementation, these are the skills you need. And I happen to sit in both camps, and this is why I love this moment. I still get to live a little bit in the old world where I look at the full implementation because there are domains and areas where I feel that still brings value, and then I also live in the future. And as I said, when I created Omawrite, I’ve not looked at a single line of that C++. I actually made it a point to myself that I was gonna treat it as an experiment. This is 100% a black box. I will treat it as though I was any other user who had opinions about how their digital typewriter should work, and therefore felt like it was a fair experiment as if a writer of any other sort who had opinions about how software should be done and how it should work.

六个月前我也会说同样的话。六个月前大概也是真的。从那以后我发现——尤其是最近几周智能体加速开发——更多时候我有谦卑去承认：智能体最懂。你可以合法地说「确保安全」。它会更懂那意味着什么。听起来疯了，但是真的。一个好平行是 AGENTS.md / CLAUDE.md 文件。有一阵微优化那玩意很火：哦，你告诉智能体这个、那个，结果是一份巨大的指令文件。Boris 在 Claude Code 上做的，最近一次访谈里关于 Opus 5 分享的一点是：他们给 Opus 5——大概还有 Fable——装的系统提示缩了 80%，因为智能体不但需要少得多的人类指令，过度规定的人类实际上在伤害它。任何有过尖头老板的程序员都确切知道那是什么感觉。老板走进房间，什么都不懂，开始告诉你怎么编程。你怎么办？闷闷不乐。如果你被强制做违背更好判断的事，你就写出更烂的代码。智能体为什么会不一样？

I would’ve said the same thing six months ago. And I think the same thing would’ve been true six months ago. What I have found since, especially even just the last few weeks of the agentically accelerated development, is that more often than not, I have the humility to recognize that the agent knows best. You can legitimately say, “Make sure it’s secure.” And it would know more about what that entails. Now, you’re laughing, but it’s true. And in fact, a good parallel to this is the AGENTS.md/CLAUDE.md files. There was a hot moment where it was all the rage about micro-optimizing that. Oh, you tell your agent this, you tell your agent that, and you ended up with this huge file of instructions for your agent. One of the things that Boris, working on Claude Code, shared in an interview recently about Opus 5 was that the system prompt that they ship for Opus 5, and presumably also Fable, shrunk by 80% because the agent not only needed far less human instruction, it was actually being damaged by overly prescriptive humans. And any programmer who’s had a pointy-haired boss knows exactly what that is like. When the boss walks into the room, doesn’t know anything, starts telling you how to program, how to code. What do you do? You sulk. You write shittier code if you’re mandated to do things that are against your better judgment. Why would an agent not be the same?

Lex Fridman

这问你很有意思，因为你长期以来出名地谈过一刀一刀雕出漂亮 Rails Ruby 代码，现在几个月内就转了。你现在的 setup 在不同抽象层是什么样？

So this is a fascinating thing to ask you because you have sort of famously for a long time talked about detailed chiseling of beautiful Rails Ruby code, and now you have switched in a matter of months to not doing that. So what does your current setup look like?

DHH

在不同抽象层。当我在 Ruby 代码里工作——我们业务全是 Ruby，有很多——让智能体改那代码时，我仍抠细节。我开始意识到：写漂亮代码的经济回报，是能用小团队快速演进、成本不高、改一处不引入一堆 bug。漂亮代码更好懂、更简单。因此写让智能体更容易处理、演进、不必重学全部上下文的系统，有很大回报。就像人一样，如果他们能在不毁架构的情况下迭代，他们就能走得更快。我见过智能体在乱成一团、什么都不连的系统上翻车；也见过我们自己的代码库：第一个 PR 质量平庸，再叠一个，越堆越差。

And at different levels of abstraction. When I’m working in Ruby code, and we have a lot of Ruby code because it’s in our entire business, and I’m asking agents to make changes to that code, I still sweat the details. What I’m coming to realize is that the economic payoff of writing beautiful code is that you can evolve quickly with a small team, and not exorbitant cost, and not introducing a bunch of bugs when you change one thing over the other. That was the driving economic argument for why you should write beautiful code, ’cause beautiful code is easier to understand, it is simpler. So therefore, there is great payoff to writing systems that agents have an easier time dealing with and evolving without having to relearn the entire context. Just like humans, if they can make iterations and changes to the code base without wrecking the architecture, they can make them faster. I’ve seen that with agents, and I’ve seen it in our own code bases where the first PR is mediocre of quality, and then if you add another PR on top of that, it gets worse.

手写漂亮代码的浪漫化已经在发生。我有一些，因为那是非常浪漫的时代。我感激有二十年经济上有价值的手写代码。那是好时光。因为手写漂亮代码今天和昨天一样存在。有人故意给 Commodore 64、Sega Mega Drive 或其他复古主机写新游戏。我最近最爱的游戏之一是 ModRetro 的。很酷。他们还出了 Nintendo 64。我玩过很多原版 Game Boy 上的俄罗斯方块，大概是我史上前三。他们重写时改了一处：按上键砖块砸下去，游戏加速大约 400%。是新的 Game Boy 游戏。有人为爱继续写新的老游戏——那很棒。而且那些漂亮代码是很好的训练数据。某种意义上，我们几十年建起的美还在延续。

It’s already happening, the romanticization of handwritten code. And I have some of it because it was a very romantic era. I’m grateful to have been alive for 20 years of economically valuable handwritten code. That was a good time. Because handwritten beautiful code exists as much today as it did yesterday. There are people who willfully write new video games for the Commodore 64, or for the Sega Mega Drive, or for other vintage consoles. In fact, one of my favorite video games of late is ModRetro’s. It’s amazing. Look it up. It’s really cool. Now they just put out the Nintendo 64 too. I played a lot of Tetris on the original Game Boy. Probably one of my top three favorite games of all time. And they rewrote it with one change. If you hit up, the brick slams. And that speeds up the game by about 400%. It is an incredible game, and it’s a new Game Boy game. The fact that there’s still people writing new old games is awesome. And also, we should say that that beautiful code is great training data. So in some sense, the beauty that we’ve built over a few decades continues.

Token 稀缺仍然让架构暂时有回报。约束还在——不是像 Commodore 64 的 1MHz CPU 和 64K 内存那种，但是另一种约束。在那些约束下写漂亮系统仍然付钱。

Token scarcity still makes architecture pay off for now. The constraints are still there — not like a Commodore 64 with one megahertz CPU and 64K of memory, but another kind of constraint. Writing beautiful systems under those constraints still pays.

Lex Fridman

能谈谈程序员在同样转型里感到的普遍焦虑吗？他们也许上了大学、主修计算机科学，梦想当程序员、建东西、拿高薪，现在一切在变，深深焦虑「我这一生怎么办？」你能共情吗？他们到底该做什么？

So can we talk about the general anxiety that programmers feel going through the same transformation? They’ve maybe gone to university, majored in computer science, dreamed of being programmers and building stuff, high salary, and now everything’s changing, and there’s deep anxiety about, “What do I do with my life?” Can you empathize with that, and what the hell are they supposed to do?

DHH

极度能。因为我虽有这种性格，我清醒知道那不是均匀分布的，不是人人都坐在这个位置、对未来有这份乐观。但我想先分开一点。如果你爱编程的唯一东西是把正确的逻辑构造拼起来、产出别人叫你产出的那种机械部分，你会很难应对新现实，因为那个机械过程受威胁。如果你像你刚说的那样兴奋于建造，我不认为你受威胁。事实上有很强的论据说我们需要远比现在更多的建造者。就业数据是模糊的。完全不清楚会发生什么。有些统计实际显示职位在增加，因为 AI 大幅压低程序的价格，人们想要更多程序。这是经典的杰文斯悖论：某样东西价格下降，需求会更多。ATM 也是例子。ATM 刚来时很多银行柜员很怕丢工作，因为突然有机器能从账户吐钱。ATM 做的是降低开一家分行的价格，银行突然能开更多分行，我们最终有了比以前更多的柜员。这些都不保证。也有真的改变的时刻。

Hugely. Because just because I have this disposition, I’m keenly aware that that’s not evenly distributed, that not everyone sits with this position and this optimism for the future. But I do think I wanna separate things a little bit. I think you’re gonna have a hard time coping with the new reality if the only thing you loved about programming was the mechanical bits of putting the right logical constructs together to produce something other people told you to produce. Because that mechanical process is under threat. If you are, as you just mentioned, excited about building things, I don’t think you’re under threat at all. In fact, I think there’s a great argument for us needing far more builders than what we have now, and if you look at the employment stats, it’s fuzzy. It’s not clear what’s gonna happen at all. Some stats actually show an increase in openings because the advent of AI is so dramatically lowering the price of programs that people want a lot more programs. This is the classic, the Jevons paradox, that says when the price of something goes down, there’s gonna be more demand for it. And this is the example of the ATMs too. When the ATM originally came, a lot of bank tellers were very afraid for their job because suddenly there was a machine that could dispense money from people’s account. Well, what the ATMs did was lower the price of a branch, and suddenly banks could afford to open a lot more branches, and we ended up with more bank tellers than we did before. Now, none of this is guaranteed. There are also moments where things do change.

直到 1800 年代末绝大多数人在田里干活，然后我们有了机械化收割工具，不再需要人拿锄头和牛。那一刻有人感到工作与生计受威胁吗？有。那合理吗？合理。英格兰砸织布机的卢德派也一样。也许那比田里干活的人更好的平行，因为那些是高技能专业人士，在相当有利的条件下做喜欢的工作，不是整天在太阳下苦干。但我们现在还想让衣服是这种高度受限的资源吗？如果这件 T 恤 400 美元？那我大概只有两件。人类总体进步依赖生产力改进。生产力意味着更少的人做同样数量或同样的工作。你想要完成的工作量可能增加，因此人更多；也可能不。可能有口袋里公司只需要固定任务集，突然能用十分之一的人做完。对当下被裁的个人是悲剧和艰难；对整体经济也惊人——突然释放这些资源去做更有生产力的事。这是整个经济演进和改进、我们获得增长的方式，增长是好的。

The vast majority of people worked in the fields up until the late 1800s, and then suddenly we got mechanized harvesting tools, and we did not need human labor with a hoe and an ox out there. In that moment, are there people who were threatened that their job and their livelihood was at stake? Yes. Was that a reasonable thing to feel threatened about? Yes. The Luddites smashing the weaving machines in England had the same thing. And I think maybe that’s the better parallel than people working the fields, because these were actually highly skilled professionals doing a job that they liked on rather favorable conditions. But where we live now, would we still like for clothing to be this heavily constrained resource? Like, what if this T-shirt was $400? I mean, okay, then I’d have two. I think the general progress of mankind depends on productivity improvements. Productivity means fewer people to do the same number or the same job. Now, the amount of job you want done may increase, and therefore you get more people, but it also may not. There may be pockets where a company just needs a certain set of fixed tasks done, and suddenly they can do them with a 10th the number of people. This, by the way, is tragic and difficult in the moment for the individual being laid off. It’s also amazing for the economy at large. Suddenly you’ve freed up these resources who can now go do more productive things. This is how the whole economy evolves and improves, and how we get growth, and growth is good.

我们不想把时钟拨回 1920 或 1950。我们该兴奋于发现了让我们能做多得多的新技术。有些被 AI 接管的任务是人类兴奋去做、只是不再经济可行的；很多任务只是苦差。这对我的编程当然为真。有些编程时刻我认为惊人——那些产生我们上次谈过的心流状态的时刻，它们是狂喜。一年里多久发生一次？如果你拿平均程序员，2000 小时工作里有多少在心流？100？200？25？我认为对某人来说那些答案都合理。把苦差交给机器，那是文明史。

We do not want to wind the clock back to 1920 or 1950. We should be excited about the fact that we’ve discovered new technology that allows us to do vastly more. And while some of the tasks that may be taken over by AI were tasks that humans were excited to do, and they’re just no longer economically viable, a lot of the tasks were just drudgery. This is certainly true of my programming. There were moments of programming I thought were amazing. These were the moments that produced that state of flow we talked about last time, and they were ecstatic. How often did they happen out of the course of a year? If you took your average programmer, how much time out of 2,000 hours at the job did that person spend in a flow state? 100? 200? 25? I think all those answers could be plausible for someone. And taking that drudgery, handing it over to machines, that is the history of civilization.

Lex Fridman

社会层面、宏观经济层面，增长令人兴奋。但个人层面会有很多焦虑，可能很多痛苦。作为建议，如果是年轻的 DHH 或现在的年轻开发者，你会建议他们做什么？

So certainly at the societal level, at the macroeconomic level, growth is exciting. But at the individual level, there’s going to be a lot of anxiety, potentially a lot of suffering. So by way of advice, if you’re like a young DHH or young developer programmer now, what would you advise they do?

DHH

别试图预判任何事。你会真的发疯。因为即便业内最聪明的脑子也无法预判从这儿跳两代模型后会什么样。绝对浪费时间，你会发展出 AI 精神病，去推演两年后什么样。专注现在，现在是爱上计算机最不可思议的时刻。如果你靠进去，可以让它们做最惊人的事。如果你强迫自己学一下前沿在哪，我双倍敢你不会对可能的事不兴奋。

Don’t try to anticipate anything. You will literally go crazy. Because even the smartest brains in the business cannot anticipate what two model hops from here is going to look like. It’s an absolute waste of time, and you will develop an AI psychosis trying to deduce what two years from now is gonna look like. Focus on right now, and right now is the most incredible time to be into computers. You can make them do the most amazing things if you lean in. If you maybe for a hot moment force yourself to learn where the state of the art is, I double dog dare you not to get excited about what’s possible.

公开建造其实是可选的。做很好，因为你能成为社区一部分，开源有巨大的同志情谊。开源是社区和同志情谊的不可思议来源，也是压住个人存在主义恐惧的好办法。如果你和其他人在一起，你会被好的心智病毒感染——兴奋的那种，告诉你一切都会没事、未来明亮，如果我们一起建一堆东西，能做没人梦想过的事。如果你只坐在孤立洞穴里担心未来，你会有点疯。我想我们从 COVID 学到的是：人独自坐在洞穴里担心会发生什么，会疯。那和抑郁无法区分，无休止反刍你控制不了的事。抱歉借一点黄仁勋的话，那就是失败者谈话。你不必当失败者。你可以选择靠进去，赢。赢的广义定义是学更多、造更多、贡献更多、参与更多。归根结底你有什么选择？你没得选，伙计。未来无论你喜不喜欢都会来，所以你不妨选择对它兴奋。

I think it’s optional actually to build publicly. I think it’s nice to do because then you get to be part of a community, and there’s an enormous amount of camaraderie. Open source is an incredible source of community and camaraderie, and it’s also a great way to stem some of that personal existential dread. If you’re amongst other humans, you can get infected with good forms of mind viruses, exciting forms of mind viruses, the ones that tell you that it’s all gonna be okay, and the future looks bright, and if we build a bunch of things together, we can do things none of us ever dreamed we could do. And if you’re just sitting in your little isolated cave worrying about the future, yeah, you’re gonna go a little nuts. I think if we learned anything during COVID, it’s that people will go nuts sitting in their little cave inside by themselves worrying about what’s gonna happen. That is indistinguishable from depression, ruminating endlessly about things you can’t control. I’m sorry to channel some Jensen Huang here, but that’s just loser talk. You don’t have to be a loser. You can choose to lean in and win. And win is very broadly defined as learning more, making more, contributing more, being part of more. And at the end of the day, what are your choices? You don’t have a choice, mate. The future’s coming whether you like it or not, so you might as well choose to be excited about it.

Lex Fridman

我刻意想找到人类存在的永恒性，真正接入：好吧，人类史上总感觉一切在变，但生命里什么重要的那些普遍性仍为真。

But then I very deliberately wanted to find the timelessness of human existence, really plug into the fact that, okay, it always feels like everything’s changing throughout human history, but the universals are still true of what matters in life.

DHH

你不会错过任何东西。这是转型早期让我磨牙的部分。人人都紧张，如果你没跟上……现在是 loops。哦不，loops 完了，是 graphs。哦不，那个也完了。没有累积，某种意义上是巨大解脱。如果你错过了过去一年，你可以在两周内追上前沿。如果你是程序员，去喜马拉雅徒步一年再回来，你可以两周追上。

You’re not gonna miss anything. This is the part that actually grinded my gears in the early phases of this transition. Everyone was so up in arms that if you didn’t… It’s loops now. Oh, no, no, we’re done with loops. It’s graphs now. Oh, no, no, we’re done with that. There’s not any accumulation, which is in some ways a great relief. If you missed the past year, you can catch up to the frontier in two weeks. If you’re a programmer who was out hiking the Himalayas for a year and you come back, you can catch up in two weeks.

Lex Fridman

但关键一步是：从背包旅行回来时，愿意成为全新、不同的人，因为变化太快。如果你去年十月出发，四月或五月回来——「什么意思，我们不编程了？」是巨大转型。

But the key step there, when you come back from the backpacking journey, is to be willing to become a totally new, different human because things are changing so fast. I mean, literally, if you left for the backpacking journey in October last year and came back in April or May — “What do you mean? We’re not programming anymore?” This is a huge transformation.

DHH

你不如在冷冻舱里待了 100 年。这令人震惊，我也承认可以悲伤一会儿。那不是我的性格，但我接受那是应对失落感的久经考验的机制。AI 不是从天上掉下来的，不是外星技术突然冲上岸。AI 和计算机科学本身一样久。如果没有 Quake、没有 Duke Nukem、没有 Unreal Tournament，我们永远得不到 AI，因为永远得不到 GPU，因此永远得不到现在这种形状的 AI。我们直到这一刻做的一切都有意义。我在那一刻就在为 AI 革命做贡献。

You might as well have been in the cryo chamber for 100 years. It’s startling, and I actually recognize that it’s okay to grief for a hot moment. That’s not my disposition. It’s not how I do it, but I accept that that’s a time-tested mechanism for coping with a sense of loss. AI did not arrive from the sky. It was not alien technology that had blasted through the universe and suddenly showed up on our shores. AI has been around for as long as computer science itself. If we had not had Quake, if we had not had Duke Nukem, if we had not had Unreal Tournament, we’d never have gotten AI because we would never have gotten the GPUs, and therefore we would never have been able to get AI in the shape it is now. So everything we did up until this moment mattered. Yeah, I was contributing to the AI revolution right there and then moment.

你可以悲伤一会儿，没关系。你甚至不必接受一切都死了。如果你仍很依恋手凿——如果一年前你问我，我会预测我会更依恋。我有点惊讶自己没有更依恋。原因是我找到了更有趣的东西。如果不是那样，如果 AI 只是取代了我爱的东西、给了我恨的东西，我大概会有点苦。我想到 Picasso 是好平行：早期他学工艺，画写实画，训练成用我们所知最好方式描绘现实的大师；然后抽象表现等形式来了，他没有哀叹自己不再做文艺复兴绘画、不再画完美的苹果。他把苹果重新想象成该死的方块，并对那兴奋。然后我碰巧爱上编程这门手艺，深潜二十年。现在我回来了。我回到有想法、迫不及待想看它存在于世界，AI 把那缩到几乎没有。

Take a moment. It’s okay. And also, you don’t even have to accept that it’s all dead. Like, if you are still very attached to the hand chiseling, which if you’d asked me a year ago, I would have predicted that I would’ve been more attached. I’m surprised a little that I’ve not been more attached. But the reason I’ve not been more attached, I think, is that I found something more fun. And if it hadn’t been like that, if AI had just replaced the thing I loved and gave me something I hated, I’d probably be a little bitter. I think Picasso is actually a good parallel because if you look at his early work where he learned his craft, it was painting realistic paintings, right? Like, he was training to be a master of the old ways of depicting reality the best way we knew how, and then along comes all these forms of abstract expression, he’s not decrying that he’s not doing Renaissance paintings anymore, that he’s not just painting the perfect depiction of an apple. He’s reimagining the apple to be a freaking square, and he’s getting excited about that. Then I happened to fall in love with programming as a craft, and then I spent two decades really diving deep on that. But now I’m back. I’m back to having an idea and being impatient beyond belief to see it exist in the world, and AI has just shrunk that down to almost nothing.

Lex Fridman

我得问你编程 setup 怎么变了？键盘、语音、IDE 是什么？

I gotta ask you about how has your programming setup changed? So, keyboard, voice, what’s the IDE?

DHH

现在想起来Crazy，但我用了 TextMate 将近 20 年。大概从 2005 年开始，我帮第一版出来，然后就没兴趣找替代。直到切到 Linux 才被迫离开栖息地。现在切到——我们叫它什么？智能体工程？哦我他妈恨那个词。我们得想出听起来和编程一样平常、但封装「和智能体一起」的事实的词。

It’s crazy to think about now, but yeah, I used TextMate for almost 20 years. I used TextMate starting in 2005, I think. I helped get the first version out, and then I just wasn’t interested. I wasn’t in the market for an alternative. And it wasn’t until the switch to Linux that I was forced out of my habitat. And now with the switch to, what are we calling it? Agentic engineering? Oh, I fucking hate that term. We gotta come up with something that sounds as plain as programming, but encapsulates the fact that it’s with agents.

Lex Fridman

我仍觉得现在该叫编程。

I still think it should be called programming at this point.

DHH

好，就叫编程。和智能体一起编程需要不同工具集。主变化是：你从脑子里的单线程编程走到并行处理。我在 TextMate 甚至不久前在 Neovim 里手凿代码时，一次专注一个问题，有条理地做完，那其实是心流的入口——深度沉浸单一问题，做到底。和智能体不是那样，部分因为智能体既太快又太慢。它们不会像键盘打字那样立刻回复。所以你得让智能体炖一会儿，于是你意识到：如果我干坐着等，首先不感觉有生产力。即便一个智能体可以很有生产力，感觉也不好，感觉你有点没用。

All right, let’s just call it programming. Programming with agents requires a different tool set. It really does. And the main change here is that you’re going from single thread programming in your head to parallel processing. When I was writing code, chiseling it by hand in TextMate or even Neovim not that long ago, I would just focus on one problem at the time, and I would methodically work my way through it, and that was actually the portal to flow. The portal to flow was deep immersion into a single problem, see it through to the end. That’s not how it works with agents, in part because the agents are at once both too fast and too slow. They don’t give you an immediate reply on something that you asked them to do that’s the same as typing on a keyboard. So you have to let the agent cook for a bit, and therefore you realize, well, if I just sit around waiting for them, first of all, that doesn’t feel productive. Even if the agent, just one of them, can be highly productive, it does not feel productive. It does not feel good. It feels actually like you’re a little bit useless.

你可以靠扔更多资源解决很多难题。这是 AI 本身的缩放律，对吧？如果你并行化，不是跑一个智能体而是一把，你可以感觉在心流，因为你不断在做编程工作——做决定，帮智能体解阻塞，或准备好新任务。要那样你需要不同 setup。我先在 tmux 里做，分开 pane、分开 split。基本上是带标签的终端。我爱智能体革命从终端踢开，因为我已经是 TUI 和终端的粉丝。那是很好的工作地方。现代终端就是好看。然后当不只是本机多个智能体，而是开始跑多台机器，单靠 tmux 不够跟踪，所以最近我切到叫 Herdr 的东西。Herdr 本质上是 tmux 加智能体通知。智能体做完、需要你时，叮一声小铃，告诉你它准备好听人的决定。它也跟踪是否在工作模式。

You can solve a lot of hard problems by simply throwing more resources at it. This is the whole scaling law of AI itself, right? That if you parallelize these things and you’re not running one agent, but you’re running a handful, you can feel like you’re in a flow state because you’re constantly doing programming work in the sense that you’re making decisions, and you’re helping either unblock an agent because it has a question about which direction to take, or you’re ready for a new task. And to do that, you need a different setup. I started first doing it in tmux and just having separate panes and having separate splits. Basically a terminal with tabs is a good way to think about it. I love the fact that this agent revolution was kicked off in the terminal because I was already a huge fan of TUIs, terminal user interfaces, and the terminal in general. That feels like a really nice place to be. And then this fact of having multiple agents, especially once it’s not just multiple agents running on your own machine, but you start running multiple machines. Now tmux alone is not enough to keep track of it, and that’s why as of late I’ve switched to this thing called Herdr. And Herdr is essentially tmux plus agent notifications. So whenever your agent is done and needs something for you, it goes ding, a little bell telling you it’s ready for its human. But it is ready for a decision, and it also keeps track of whether it is working mode or not.

大约一个月前我疯了一阵，意识到单机做这工作不够快。像发现了多核编程，但我只有两核。「如果我有 16 核？32？64？」我立刻出去买了这些惊人的 KVM，叫 GL.iNet Comet。小盒子，插 HDMI、插 USB 连到电脑。像 KVM，特别的是多容易：连上，去网页，登录一次，设一个密码。现在这东西可以上你的 tailnet。过去一年对我的另一场革命是发现这些 WireGuard 网络。Tailscale 本质上把你所有电脑变成无论你在哪都是本地网络。现在在我手机上，我直接访问 Malibu 办公室所有电脑，也访问哥本哈根办公室所有电脑。我可以把它们当坐在旁边一样，不必在防火墙打洞或设复杂 VPN。摩擦一降，我看衣橱里有一堆以前实验的迷你 PC，就说「如果全连上呢？」我连了四台，各自有小 Comet，突然能同时在更多电脑上跑智能体，全用 Herdr 控制。到了一个点我把自己的处理能力打满了。大约四五台机，我不知道，三个智能体。我大约有 16 个线程。那是我能跑的。智能体跑得越快，我能跑的线程越少，但以当前节奏，我能满加速跑大约 16 个线程。

I went on this crazy phase just about a month ago realizing that doing this work on a single machine is not fast enough. It’s like I’ve discovered multi-core programming, but I only have two cores. I’m like, “What if I had 16 cores? What if I had 32 cores? What if I had 64 cores?” So I instantly went out and I bought these amazing KVMs called GL.iNet Comets. And what they do is it’s this little box. You plug in HDMI, you plug in USB and connect it to the computer. It’s like a KVM. So a KVM is a remote way of controlling a computer, but what’s special about this is just how easy it was. You connect this thing in, you go to a webpage, log in once, set one password. Now this thing can hop on your tailnet. This has been the other revolution of the last year for me, is discovering these WireGuard networks. Tailscale is essentially turning all the computers you have into a local network wherever you are. Like right now on my phone, I have direct access to all the computers in my Malibu office. I also have access to all my computers in my Copenhagen office. And I can treat them as though I sat right next to them without having to punch holes in a firewall or set up complicated VPNs. So as soon as I discovered this, I looked at my closet, and I realized I had a bunch of mini PCs from prior experiments, and I just said, “What if I just connected all of them?” And I just connected four of the computers in a closet. They all had their little Comet, and suddenly I could run agents on more computers at the same time, and I could control them all with Herdr. And it did get to a point where I maxed out my own processing power. That I think at about four to five machines running, I don’t know, three agents. I have about 16 threads. That’s what I can run. And the faster the agents run, of course, the fewer threads I can run, but at the current pace, I can run about 16 threads at full acceleration.

这就是 setup。仍是 Neovim，但此刻我不写很多代码，所以把 Neovim 当项目浏览器，以及踢起 lazygit 看变更日志。如果 GitHub 显示 PR 再快一点，Web 也许更舒服。还有个叫 Hunk 的工具做漂亮 diff。但看 Hunk 我只看见变更集，审智能体产出时我常想看周围上下文。所以仍喜欢用 Neovim。当然全在 Omarchy 上，全在 Linux 上。这是智能体的另一大突破。智能体爱 Unix 哲学。它爱能通过命令行调用的单个工具，三大操作系统——Mac、Windows、Linux——没有哪个像 Linux 那样和这机制合拍。Linux 里一切要么是配置文件，要么是 CLI 工具。五分钟前那是主要缺点，人们不喜欢 Linux 的原因。宇宙开的巨大玩笑：五分钟前的缺点现在是主要卖点。Raycast 那么简单的东西——没有你能直接访问的配置文件。你得进 GUI，导出文件，再导入别处。你无法自动化整机 setup，也无法自动化配置 Mac 默认键绑定。那得手动，像穴居人点鼠标。

So that’s the setup. It’s still Neovim, but at this point, I’m not writing a lot of code, so I’m using Neovim as a project browser and then as a way to kick off lazygit to see the change log for what’s there. And even that, I’d say if GitHub was a little faster at showing you your pull request, the web is probably actually a nicer place to do that. There’s also this other tool I’ve been playing with a bit called Hunk, which just produces diffs in a really nice way. But I find that when I look at Hunk, I only see the change set, and when I’m reviewing output from an agent, I often want to see the surrounding context. So that’s why I still like Neovim as a way of doing it. But it’s all happening, by the way, of course, in Omarchy. So it’s all happening on Linux. And this was the other major breakthrough with agents. Agents love the Unix philosophy. It loves individual tools that it can invoke through the command line, and there is no operating system on Earth of the majors, I’m counting three here, Mac, Windows, Linux, that works as well with that mechanism as Linux. Everything in Linux is either a config file or a CLI tool. Now, that was its main drawback five minutes ago. This was the reason people didn’t like Linux. What great irony that the universe has played it upon us that now the drawbacks of Linux five minutes ago are now its major selling points. But still something as simple as Raycast — there’s no config file that you can just access. You have to go into the GUI, export a file, then take that file, I don’t know, in your freaking backpack on a USB key. And then you can import it somewhere else. You cannot automate the entire setup of your machine. You can’t automate at all the configuration of Mac’s default key bindings. That has to be a manual process where you’re clicking with a mouse like a caveman to set up your machine.

DHH

这是个好时机。我听说今天是你生日。所以我带了点小礼物。无论你喜不喜欢，我们都要把你弄上智能体操作系统。我问了 Dell 的朋友，看能不能给你弄一台生日机器。这是我用的机器——同一款，但现在是你的了。Dell XPS 14，已经装好 Omarchy 4，准备给你配置。对了，我们几乎该计时。你可以起来就跑。大约一分钟里要答大概五个问题。你愿意的话可以现场做。

So this is a good moment. I heard it was your birthday. So I brought a little gift. We are going to get you onto the agentic operating system whether you like it or not. And I asked my friends at Dell whether maybe they had a machine that I could get you for your birthday. So here’s the machine I use — which is well, the same one I use, but it’s now yours. It’s a Dell XPS 14 already set up with Omarchy 4 ready to be configured for you. And once you realize that… By the way, we should almost time this. You can be up and running. You have to answer, like, five questions in about one minute. You could literally do it live if you wanted to.

Lex Fridman

Oh, Omarchy，漂亮现代有主见的 Linux，DHH 做的。来吧。

Oh, Omarchy, beautiful, modern opinion Linux by DHH. Yeah, let’s do it.

DHH

哦，大约 40 秒就做完，然后你就进 Omarchy 了。

Oh, it’s gonna be done in about 40 seconds. And then you’re gonna be inside of Omarchy.

我要最快的车、能破限速；要能潜到马里亚纳海沟的潜水表；要能在不到 60 秒装完的操作系统。当我们突破一分钟关卡——我想大概三周前、或许两周前才真正破——我简直狂喜。能砍掉的每一秒都是兴奋。

I want the fastest car breaking the speed limits. I want the diver watch that can go down the Mariana Trench. I want the operating system that can install in less than 60 seconds. And when we broke the one-minute barrier, which was really only broken, I think, three weeks ago or something like that, maybe even two weeks ago, I just… Every second I could shave off was an excitement.

Lex Fridman

发行版。你还记得是什么让我们这么快吗？你提过智能体发现的一些事。

… distribution. Is there some stuff you remember about what it took to get us to be so fast? You mentioned a few things that were, the agents were discovering.

DHH

其中一件是把人类输入的滞后当预加载机会。装 Omarchy 机器时你答五个问题，后台同时在干活。我甚至没考虑过——游戏里最老套的把戏。我以为先问密码、用户名、时区等等，再干活；其实可以预加载。还有安装器顺序上的一堆微调，也有单纯缩小体积。上一版 Omarchy 是 7.5GB。Omarchy 用 JetBrains 字体——很棒的字体，我第一眼其实不喜欢，但它是我能在 Mitchell Hashimoto 的 Ghostty 里看起来完美的唯一字体。Ghostty 不知为何不愿按我习惯渲染心爱的 Bitstream Vera Sans，但这套 JetBrains 字体完美。于是我做了个 slim 版 JetBrains 包，一下子省了 180MB。NVIDIA 驱动也做了类似的事。

One of the things was to treat the lag of human input as an opportunity to preload. So there’s five questions you answer when you set up an Omarchy machine. So it’s doing stuff in the background. We’re doing stuff in the background. And that was one of the things I didn’t even consider, which is, I mean, the oldest trick in the book. All sorts of video games for a long time have done this. I thought, first you ask the user some questions about their password and their username and their time zone and so forth, and then you do the work, when, in fact, you could preload that. Now, the other things were just a bunch of other tweaks to how the installer itself was doing things in a certain order. Some of them was also just shrinking it. The last version of Omarchy was 7.5 gigabytes. So Omarchy uses the JetBrains font, which is an awesome font, beautiful font, and actually a font I didn’t like at first glance, but then it was the only font I could get to look perfect in Mitchell Hashimoto’s Ghostty. Ghostty, for whatever reason, did not wanna render my beloved Bitstream Vera Sans just the way I was used to it. But it rendered this JetBrains font perfectly, so I switched. So I simply just came up with a new package, the slim version of the JetBrains package. And there, right there, I saved 180 megabytes. I just went through and did that a bunch of times. We did that with the NVIDIA drivers too.

McLaren 现在做出世界上最轻的一些超级跑车，他们执着于从车上削掉每一克。所以新的 750S 也比对手轻得多。他们用碳纤维单体壳，相对 Ferrari 的铝。我那一刻削掉一兆又一兆时想：伙计，我有点像 McLaren 的汽车开发者。这其实也是我们和传统 Linux 社区在「膨胀」上的冲突。我说：我做选择，这是我认为很棒的应用集合——顺便有视频编辑器。装箱带 OBS 做录制。Kdenlive 是时间线编辑器，很好，我所有视频都用它剪。然后我做了自己的剪辑编辑器。你可以按 Control + Space 把剪辑线移到那一行，到结尾按 Alt + Space 移结束线，再 Control + S 保存。用 Omacut 我剪片段快得要命。上面有 Neovim、Herdr、Tmux、终端、一堆东西。Linux 配置部分其实大多用 Bash 建。Bash 很适合系统管理，也有局限。你不该什么都用 Bash 建。

So McLaren makes some of the lightest super sports cars in the world right now, and they are absolutely obsessed with shaving every damn last gram off their cars. That’s why even a new 750S McLaren is just way lighter than the competition. They use a carbon fiber monocoque, and that helps them over like Ferrari, for example, that uses aluminum. So I thought in that moment as I was shaving individual megabytes off the packages, man, I’m kinda like a McLaren car developer here. But this is actually this other conflict that we’ve had with Omarchy and the traditional Linux community is bloat. I’m making the choices, and this is the collection of applications I think is awesome, which by the way has a video editor. It ships with OBS to do your recording. It’s just in there. Kdenlive is a video editor, timeline editor, which is great. I edit all my videos with that. And then I made my own clip editor. And then you can hit Control + Space. It moves the clip line to that line. Then you go to the end of the clip, and you hit Alt + Space, and it moves the end line. And then you hit Control + S, and it saves. I can make clips so damn fast now with Omacut. It’s amazing. But it has all the software on it. It has Neovim on it. It has Herdr on it. It has Tmux. It has a terminal. So it’s actually mostly built in Bash for the Linux configuration part. Bash is a great language for that. It is made for that really, for system administration. It also has its limitations. You should not build everything in Bash.

Lex Fridman

Whisper Flow 之类语音转文字很好。有人把那一步做起来了。

Well, I do think there’s a few… So, like, Whisper flow is really good. A few people have stepped up that speech-to-text thing.

DHH

对。我们在用 VoxType。它是开源的，只用某个开源模型。短命令相当好用。我想按住 F9 就开始听写。你得上网，它会提议装 VoxType 包，因为包里有个 150MB 的模型，我当时想「啊，不知道能不能扛」。所以默认不带，但选项设好了。昨天我还在想我们真该把这些合起来。很多人会觉得很自然：跟电脑说「嘿，给我做个股票面板，我想跟踪 Apple 和 Dell」，然后看智能体全靠语音——输入输出都是——操作系统就变了。那是钢铁侠贾维斯的愿景。定制应用魔法般出现。

Yeah. Yeah, we’re using VoxType. It is. It’s just using one of the open models. I forget which parrot model it’s using or something like that. It works quite well for short commands, so I have used it for that. I think you just hold down F9 and it starts a dictation. You just have to get online. It’ll offer you to install the VoxType package. Because the VoxType package includes a model that’s 150 megabytes, and I was like, “Ah. I don’t know if I can carry that.” So we don’t have that in there by default, but we have the option set up. And I actually just yesterday was thinking we really need to combine all of these things. There’s a lot of people who would find it very natural just to talk to their computer and say, “Hey, can you make me a stock panel? I wanna track Apple and Dell,” and just see the agent go off through, entirely through voice, both in terms of the input and in terms of the output, and then your operating system just changes. This was the vision of Iron Man’s Jarvis. Bespoke applications that just appear magically.

Lex Fridman

发那种东西时你确实得开始想延迟。

But, you know, when you’re shipping that kind of stuff, you do have to start thinking about latency.

DHH

别让完美成为够好的敌人。如果你突然让人们以前根本不会做的事成为可能，多花五秒也行。Jobs 在这方面很强：第一代 iPhone 各方面都很糟，数据慢得可怕、性能不足，但那没关系。Linux 出问题时，智能体可以拿那条对普通人毫无意义的具体错误信息，和它在四千万行 Linux 代码上预训练过的事实关联，精确知道往哪看。从今年初起，我 Linux 机器上没有任何问题是智能体诊断不了的。以前不是这样。现在智能体知道的不只是 Linux 操作系统源码，还有我盒子上每一块软件的源码。发货前我做进了一件事：设好默认智能体后，Omarchy Quattro 有 crash watcher。机器上任何应用崩了，会弹个小东西问你要不要让智能体查。我见过子系统里有错，智能体翻日志、看 systemd log，然后检出崩掉应用的该死源码，钉到某个 Rust 文件第 472 行有个未绑定、未 unwrap 的变量溢出之类，再提出修复。

Don’t let the perfect be the enemy of good. I mean, if you’re suddenly enabling something people wouldn’t be doing before at all, it’s okay to take five seconds longer. I think this is one of the areas that Steve Jobs was so good, that he realized, like, the first iPhone was terrible in all sorts of ways, right? Absolutely horrendously slow data connection, very underpowered, all the things. When Linux has an issue, the agent can take that very specific error message that makes no sense to a normal human and correlate it with the fact that the agent was pre-trained on 40 million lines of Linux code, so it knows exactly where to look and dial it down. I have not had a single problem on my Linux machine since the beginning of this year that an agent could not diagnose. That was not true before. Now the agents know the source code of not just the Linux operating system, but every single piece of software I have on that box. The agent has access to the source code of all of it. In fact, this was one thing I built in just before we shipped. So once you set up your default agent, Omarchy Quattro has a crash watcher. If any app on your machine crashes, it’ll pop up a little thing, ask you whether to have the agent investigate. Unbelievable. I’ve seen things where there’s an error in some subsystem. The agents start digging through the logs and look up systemd log. Then it checks out the damn source code of the application that crashed, pins down that it’s in this Rust file line 472 that there’s an unbounded, unwrapped variable that overflowed or whatever it is. Then offers a fix.

对我惊人的是你可以报出这么多不同竞争者。市场那么开放，确实来回换，我们有真竞争。夏天是 Opus 5、Fable、Sol、GPT Sol，以及程度稍低的一些开源权重模型。

What’s amazing to me is that you could rattle off so many different contenders. That this market is so wide open, that it does actually change back and forth, that we have real competition. This summer, with Opus 5, Fable, and Sol, GPT Sol, and to a lesser extent, some of the open weight models.

Lex Fridman

规划方面我觉得没什么能比 Fable。我通常用 Fable 做规划和审阅，实现用别的，比如 Opus 5。这套很好，不太容易把 token 用光。

I do, you know, I just don’t think there’s anything that compares to Fable in terms of planning. So I usually do Fable for planning and for reviewing and then something else for implementation, like Opus 5. And it’s really nice. It’s a really nice setup that allows you to not run out of tokens too much.

DHH

我发现很有意思：Fable 在我看来是眼下最好的模型，但它也会犯错。要拿到最好软件，我其实宁愿有两个不同来源的——它们都不是中游，都是前沿——比如 Opus 5 和 Codex，让一个检查另一个的活。这是我现在的标准操作程序。我也开始试 Grok，也相当好，不断发现问题。工作流是本机智能体做完推到 GitHub，然后 Copilot——我不骗你——真的变好了。Copilot 不断发现真正坏掉的东西，也是不可思议的加速，因为早期 Copilot 很糟。我们不该惊讶。即便你是好程序员，做完活让同样很好的同伴审，你会得到更好代码。当然会。把那建进流程。选一个智能体来开。我主要用 Claude Code 开。在 Agent View 你可以再捡起另一个智能体。如果你想多线程——Claude Code 就是最舒服的 setup。他们也一直稍微领先一点，大概不该惊讶。Boris，做那东西的人之一，基本上是第一个想出这玩意的。用 Claude 订阅感觉像划算得疯狂的便宜货。我今早刚签了第二个——因为时差四点起来，立刻和智能体干活，离 Fable 限额重置还有三天，token 用光了。为什么那么复杂？不能叠一个订阅吗？为什么要登多次？我们其实在把多订阅支持建进 Omarchy。下一版 Omarchy 会带 multi-sub 支持。我希望实验室直接让你买一百次 max。

What I found, it’s really interesting because Fable is, in my opinion, the best model right now, but it also makes mistakes. And the best way to get the best software, I would actually rather have two differently sourced sort of… I mean, they’re not mid-tier, they’re all frontier. But have, let’s say Opus 5 and Codex — and have one check the other’s job. This is my standard operating procedure now. And I’ve also started using Grok just to test it out, and it’s also quite good. And it keeps finding stuff. And then that’s my workflow when I’m having my agents on my own machine do it, and then I push to GitHub. And then Copilot, I kid you not, has actually gotten good. Copilot keeps finding stuff that’s legitimately broken, which is also incredible acceleration because the first version of Copilot was terrible. Keeps finding things. And if you then take that, and we shouldn’t be surprised. Why are we surprised? Even if you’re a good programmer, if you finish a job and you ask your also very good peer to review it, you’re gonna end up with better code. Of course you’re gonna end up with better code. So build that into your process. Pick one of the agents to drive with. I’ve mainly been driving with Claude Code. And here in Agent View, you can pick up another agent. So if you wanna do this thing where you have multiple threads going on — Claude Code is just the nicest setup. They also just, they keep being a little further ahead, which probably shouldn’t be surprising. I mean, Boris, one of the guys that’s working on that, he was the first one and basically came up with the thing. Using Claude with a subscription does feel like a bargain, like a crazy bargain. I just signed up for my second one this morning — because I got up at 4:00 with jet lag, and I started working with the agents right away, and now I’m three days away from limits resetting on Fable, and I ran out of tokens. I mean, I don’t… Why is that so complicated? Can’t you just stack one subscription? Why do I need to log in multiple times? Oh, we’re building that into Omarchy, by the way. So the next version of Omarchy is gonna ship with multi-sub support. I wish that the labs would just make you buy a max 100 times.

Lex Fridman

我最近用 Higgsfield 生成很多视频，所以他们成了赞助。做了个赛车视频。想听听你的意见。

I’ve been generating a lot of video recently with Higgsfield, and so they became a sponsor. Created a racing video. Wanted to get your opinion on it.

DHH

哦？让我看看。如果他们像电影里那样在直道降档，我会点名。那最糟。

Oh, yeah? Oh, let me see. Let me see. If they downshift on a straight like they do in the movies, I’ll call it out. That is the worst. Let me see.

Lex Fridman

这是完全 AI 生成的。

… this is fully AI generated.

DHH

等等什么？这是 AI？该死，那是我的赛车服。那是我的车。天哪。你开玩笑吗？太疯了。你真得注意……那辆车没开大灯另一辆开了。那看起来像 60 岁版的我。但天哪……哇，太疯了。

Wait, what? This is AI? Shit, that’s my suit. That’s my car. Oh, my God. Are you kidding me? I mean, it’s crazy. You really have to notice in the… Like, that car didn’t have headlights on and the other one did. And that looked like a 60-year-old version of me. But holy crap, the… Wow, that was crazy.

Lex Fridman

和编程有平行：如果只是直线视频生成、没有人在环里，你会得到很多奇怪伪影。所以他们——我强烈建议去看 Higgsfield YouTube——有完整 90 分钟原创电影。我移不开眼。以前感觉只是预告片、广告；现在真在讲故事，人脸在说话，你被吸进去。还没到编程那个程度。问题是怎么把人集成进制片环：你得创造角色并保持一致，再喂图像。你觉得编程之外，能把编程的教训用到视频创作、艺术上吗？

So there’s parallels here to programming because if you do just straight video generation with no human in the loop — you get a lot of weird artifacts — and so on. So they do and I highly recommend people go to the Higgsfield YouTube. They have full 90-minute original movies. And, like, I can’t look away. It’s really cool. ’Cause it used to be that it’s just, like, something that just feels like trailers. Like advertisements for something. This is actually telling stories, like, people’s faces and they’re talking and you’re like, you’re drawn in. It’s not quite where programming is. And so the question is how do you integrate the human into the loop of the filmmaking process? So you have to create the people, and they have to be kept consistent. And then for the images that are kinda like feeding the thing in the creation. The question I had for you is, like, do you think outside of programming, can you apply lessons from programming to video creation, to art?

DHH

好问题，因为我认为好艺术和好软件依赖同一件事：有愿景，对你想创造什么有内聚想法。它怎么不同？怎么新颖？怎么吸引人？但我不确定创造本能转移得那么好。我不知道自己会不会有创造洞见去想出好电影。

It’s a good question because I think good art is very dependent on the same thing that good software is depending on: having a vision, having a cohesive idea of what you wanna create. How is it different? How is it novel? How is it going to appeal to people? But I’m not sure that the creative instincts transfer quite as well. I don’t know that I would have the creative insights to come up with good film.

Lex Fridman

我们谈了 AI 智能体。谈谈人的一面。当爸爸你最爱什么？

We’ve been talking about AI agents. Let’s talk about the human side. What do you love most about being a dad?

DHH

好问题。我认为从直接源于你血脉的人类身上扩展出来的那种整体的爱，很难传达，因为有自己的孩子到来之前，我其实不是特别喜欢小孩的人。现在他们多半时候是世界上最有趣的人。不总是。有时他们也只是混蛋之类。但有一种利害关系。还有就是看着人从婴儿长到学步、再到青少年的纯粹喜悦。我最大的孩子刚成青少年。我常和妻子 Jamie 谈这个：如果在最后一天错过那个机会，我会有的遗憾会是彻底的，不像任何别的。我认为即便在通常被描绘得很糟的当下，也远更满足。人们总聚焦他们糟的一面：「哦你睡不着」之类。「烦人。不知感恩。」和你爱的另一个人一起创造生命，字面就是在这个星球上的巅峰体验。

That’s a good question. I think the overall love that expands from humans that derive directly from your lineage is very difficult to communicate because I was actually not a particularly big kids person prior to the arrival of my own. And now they are the most interesting people in the world. A lot of the time. Not always. Sometimes they’re also just assholes and so forth. But there’s been… And that you have a stake in that. Now, there’s also just the sheer joys of watching a human grow from a baby to a toddler to a teenager. My oldest has just become a teenager. And I often talk to Jamie, my wife, about this, where the regret I would have of having missed that opportunity on the last day would just be total, unlike anything else. So I do think it is far more satisfying even in the moment that is normally portrayed. Again, as I say, there’s such a focus on all the ways they suck. “Oh, you don’t get to sleep,” or whatever. “They’re annoying. They’re ungrateful.” Creating life with another human that you love is literally the peak experience of being on the planet.

Lex Fridman

你怎么看他们在这个充满 AI 的世界长大？感觉——也许我只是听起来像门廊上的老头——是非常不同的世界。我在互联网之前长大，但即便互联网也不觉得有这场转型这么大。

What do you think about them coming up in this world full of AI? Is it totally different? It feels like, maybe I just sound like an old man on a porch, but it feels like a very different world. So I grew up before the internet, but even the internet doesn’t feel like as big of a transformation as this.

DHH

不。但有过更大的转型。想象出生在大约 1880 年。想象看到第一次世界大战和第二次世界大战。想象看到飞机、收音机、电视。转型——我想 Peter Thiel 有这个论点——物理世界其实很久没什么发生了。我们停滞了。所有发展都在数字和比特里。但历史上有过对他们来说当然会像我们正经历的一样重大的过渡。有那种历史感——你在担忧、焦虑上没那么特殊——是发现斯多葛著作的伟大启示之一。两千五百年前这些人处理的困境和挑战非常熟悉，认识到人性的常数。让生命值得的东西今天和那时基本一样。

No. But there’s been bigger transformations. I mean, imagine growing or being born in like 1880. Imagine seeing the First World War and the Second World War. Imagine seeing the airplane, the radio — television. I mean, the transformation, and I think Peter Thiel makes this argument, that basically nothing has happened in the physical world in quite a long time. We’ve been stagnant. All the development has been in digits and bits. But there have been transitions that for them certainly would feel as momentous as what we’re living through. And I think having that sense of history, that you’re not that special in your sense of worries, in your anxieties. This was one of the great revelations of discovering the stoic writings. You have these guys 2,500 years ago dealing with very familiar dilemmas and challenges in their life, and recognizing the constants of human nature. The thing that makes life worthwhile is basically the same today as it was then.

我读了《The Fourth Turning》。它谈历史按循环运转，有这些阶段。具体理论没「历史是圆多于直线」这个认识重要。我们一遍遍重复许多同样模式。它们感觉新，其实不是。

I read this book The Fourth Turning. You check that out. It talks about this notion of history working in cycles, and there are these phases to history, and they name them and so forth. And the specific theory is not as important as just realizing that this is a… History is a circle, more so than just a straight line. And we are repeating many of the same patterns over and over again. They feel so new, and they’re not.

Lex Fridman

你对 Linux 相当乐观。你觉得 Linux 合法能提高采用率吗？因为多年迷因是——

Do you… You’ve been pretty optimistic about Linux. Do you think legit Linux can increase its adoption? ’Cause, you know, for many years the meme is, you know—

DHH

桌面上的 Linux……每年。明年再来。

Linux on the desktop — every year. Next year again.

Lex Fridman

对。但至少按你做的论证、你投入的能量，能看到它因智能体爱 Linux 而拿下的愿景。

Yeah. But it seems like, at least the case you’re making, the energy you’re putting into it, you could see a vision where it takes over because agents love Linux.

DHH

我不但能看见，我认为这已是此刻最可能的结果。它简直太适合这一刻。再说一遍，巨大的讽刺是：Linux 所有缺陷——晦涩配置文件、各种奇怪错误信息等等——偏偏是智能体操作系统的完美东西。但就是这样，我们该庆幸命运把我们带到这儿。Linux 彻头彻尾开放。有趣的是那也不是显然的。如果你看很多 Linux 社区、开源社区对 AI 的反应，不是普遍热爱。我会说多数人即使不怀疑，也是公开敌视 AI。救命的是 BDFL 本人 Linus Torvalds——他想掌舵，想确保质量之类，但那句话大概是：如果你以为 Linux 是反 AI 项目，再想一想，你该做开源那套去 fork，因为我们要用 AI。你看到进内核的 AI 贡献数量图，是抛物线。所以 Linux 在硬靠进去。所有系统、所有服务器，一切都是 Linux。桌面和个人电脑其实没被俘获，有点像好奇，但显然只是在等这一刻。Linux 从 91 年到现在，在等智能体作为终端用户操作系统充分繁荣。

Not only can I see it, I find it to be the most probable outcome at this point. It is simply too well-suited for the moment. And again, as we talked about, it’s a great irony that all the flaws of Linux, the arcane config files, all the strange error messages and so on should just so happen to be the perfect thing for an agentic operating system. But so it is, and we should rejoice that fate has taken us here. And Linux is simply open through and through. And it’s also interesting because that was not obvious either. If you look at the way a lot of the Linux communities, open source communities, have reacted to AI, it is not universal love. I would argue that the majority of them are actually, if not skeptical, then outright hostile to AI. Now, the saving grace is that the BDFL himself, Linus Torvalds — he wants to steer it. He wants to make sure it’s good and whatever, but the line was something, if you think Linux is an anti-AI project, think again, and you should just do the open source thing and fork it, because we’re gonna use AI. And you see these graphs of the number of AI contributions going into the kernel, and it’s a parabolic curve. So Linux is leaning hard into this. All the systems, all the servers, everything is Linux. So it was kind of a curiosity that the desktop and the personal computers we were using really hadn’t been captured, but clearly it was just waiting for this moment. Linux spent the time from ’91 to now waiting for agents to fully flourish as an end user operating system.

我认为没人切换难，是因为没有令人信服的理由。这是 Omarchy 的驱动设计目标之一。它不会是 Temu Windows 或 Temu Mac。不会只是尽量熟悉然后有点更糟的廉价复制。如果不客气，我会说那是 Ubuntu 试图做的。我认为这就是 Linux 现在有机会赢的原因。作为智能体操作系统、可塑操作系统，你可以按欲望定制，无与伦比。那太有说服力，我现在能看见。就从 Quattro 发布起，拥抱可塑那一面、开始做自己的东西、分享截图的人有多少。你告诉计算机这些晦涩命令、刚性逻辑结构，它产出你想要的。现在突然你坐下用白话英语碎碎念 20 分钟。「哦如果它能那样就好。其实不对。让我们……」然后软件出来了。按你形象塑造的操作系统出来了。我认为那是如果你有过……如果你感觉到……

I think it’s hard for people to switch when there’s not a compelling reason to do so. This was one of the driving design goals for Omarchy. It was not gonna be Temu Windows or Temu Mac. It was not just gonna be a cheap copy where we try to make it as familiar as possible and then kinda worse. I mean, if I was being unkind, I would say that’s what Ubuntu has tried to do. And I think this is why Linux now has the opportunity to win. Because as an agentic operating system, as a malleable operating system, you can tailor to your desires, it is unparalleled. And that is so compelling, and I can see it now. Just since the release of Quattro, the amount of people who have embraced that aspect of it, the malleability, started making their own things, sharing screenshots. You tell the computer these arcane commands, this rigid logical structure, and it produces what you want. Now suddenly, you sit down and in plain English ramble for 20 minutes. “Oh, it’d be great if it did that. Actually, no, that’s not right. Let us…” And out comes software. Out comes an operating system shaped in your image. I think that is one of those experiences that if you have it… If you feel it…

Lex Fridman

还有一个人。你怎么看 PewDiePie 用 Linux？他是 Arch 人，对吧？

There’s another guy. What do you think about PewDiePie using Linux? He’s an Arch person, no?

DHH

他是 Arch 人。他有最华丽的 rice。一度世界最大主播，我记得孩子们看过他一些 Minecraft 视频，然后他受够了，做了健全的事。结婚，成了家庭男人，有孩子，搬到日本，然后硬核进——先 Linux，再 Arch，再 Ricing。你看到他那个 Chernobyl rice 了吗？绝对顶级不可思议。然后他也成了这一刻该死的 AI 人，建所有这些系统，你就想：那……进化故事多鼓舞人。你可以从搞笑家伙变成——

He’s an Arch person. He’s got the most gorgeous rice. So the world’s biggest streamer for a while, I remember my kids watching some of his Minecraft videos, and then he just gets fed up with that, does the wholesome thing. Marries — a family man, kid, moves to Japan, and then gets so hardcore into — first Linux, then Arch, then Ricing. Did you see his rice that was Chernobyl rice? Absolutely best tier incredible stuff. And then he also becomes a goddamn AI man of the moment building all of these systems, and you’re just going like, that… What an inspiring story of evolution. Like, you can go from being funny guy —

Lex Fridman

所以他大概是软件未来、建造未来的很好化身，对吧？因为他技术上是非程序员……正在变成程序员。

So I guess, I mean, he’s a pretty good embodiment of what the future of software, or what the future of building looks like, right? ’Cause he’s a non-programmer technically — becoming a programmer.

DHH

极度鼓舞，对那些坐在「我不太会编程，这个那个不知道」那一刻的人该极度鼓舞。好，但如果 PewDiePie 能用他定制硬件建 Council of AIs，那也许你也可以开始做事。我认为这其实是重要的一般点：我们需要榜样。我们需要鼓舞人的人。对 Omarchy 也有很多批评。很多时候像外人进来时常发生的。我用 Linux 才两年半。不长。那个社区很多人字面从 Linux 又硬又难、得光脚在雪地里双向爬坡逆风走的时候就在跑。有一种 nerd 不想把空间变成别的。我不想改变 Arch 空间。我其实不觉得 Omarchy 直接是同一社区的一部分。因为最初被 Arch 吸引的人，是被亲手建一切吸引的人。因为喜欢难所以走难路，而我建的是那的两极反面。所以我只希望那些人能说：「好，这不适合我。我想走难路。我想花 500 小时 rice 自己的 Arch 发行版」，我认为你该那样。实际上我说……我做过。Omarchy 就是我倒进去的——到这一刻。

Hugely inspiring, and should be hugely inspiring to others who sit in that moment like, “Well, I don’t know quite how to program. I don’t know this and the other thing.” Okay, but if PewDiePie can build the Council of AIs with his bespoke hardware here, then maybe you can also start on things. And I think this is actually an important general point, that we need role models. We need people to inspire. And a lot of it, I mean, is as often happens when someone comes in from the outside. I mean, I’ve been using Linux for two and a half years. That’s not very long. Plenty of people in that community who’ve literally been running Linux since it was hard and difficult, and you had to walk uphill both directions against the wind in the snow barefoot. And I think there is a certain kind of nerd who’s like, now you’re changing it into another space. I don’t wanna do that. Like, I don’t wanna change the Arch space at all. Like, I don’t really think of Omarchy as being directly part of the same community. Because the people who were attracted to Arch in the first place were the ones who were attracted to building everything by hand themselves. Doing it the hard way because they like to do it hard, and I’m building the polar opposite of that. So I just wish that those people could go like, “Okay, well, this is not for me. I wanna build it the hard way. I wanna put in 500 hours to ricing my own Arch distribution,” and I think you should. I mean, actually, I said… I mean, I did it. This is what Omarchy is, me pouring in literally, at this point.

DHH

但人们兴奋的原因是：我毫不保留地说，「我们要去这儿。我认为个人计算机的未来是可塑计算机、智能体计算机。我要全押。」然后任何对计算机类似轨迹兴奋的人，上车。车里地方很多。事实上我引以为傲的一件事……「……dot files 合集。现在我试了 Quattro，真棒。我现在喜欢了。」对我来说，我为长辩论而活。我为这种辩论而活：你两年前播下种子，两年后他们说，「该死，这王八蛋是对的。」

But the reason people are excited is because I don’t have any reservation about saying, “This is where we’re going. I think the future of the personal computer is the malleable computer, is the agentic computer. I’m gonna go all in on that.” And then anyone who’s excited about a similar trajectory for computers, come along. Plenty of, plenty of room in the car. In fact, one of the things I pride myself on… “… dot files collection. And now I tried Quattro and awesome. I like it now.” That to me, I live for the long argument. I live for this argument where you plant the seed like two years ago, and then two years later they go, “Goddammit, this son of a bitch was right.”

Lex Fridman

对。你意识到如果你继续这么成功，OpenAI 和 Anthropic 会杀进来试图做他们的操作系统，或者像 Peter 和 OpenClaw 那样开巨额支票买。

Yeah. You do realize if you continue to be as successful as you are, OpenAI and Anthropic are gonna roll in and try to do their operating system or to buy, like with Peter with OpenClaw, there’d be a gigantic check.

DHH

对。我认为我当前位置的好处是我真的不需要钱。所以我能纯粹为享受和愿景纯度建这些东西，这有种奇怪品质：有时当你停止在意别人怎么想、要什么，只追求单一愿景，它最终变得远更有吸引力。当你试图做……方向上我只是想要计算机是这样。所以如果 Omarchy 最终只是历史脚注，是智能体 OS 和可塑计算机早期运动的一部分，也没关系。我完全平静。只要我能有一台像 Omarchy 一样好玩的计算机——那就很棒。如果更好的 Rails 来了，我会用。我爱 Ruby，会继续写 Ruby，哪怕只为纯粹乐趣，像马厩里有马、车库里有 Model Y 我仍会去骑马。语言现在是英语。是陈词滥调，但也是真的。过去三个月我一直在用英语编程。

Yeah. I think the good thing about my current position is I really don’t need the money. So I get to build these things purely for my own enjoyment and purity of vision, which has this weird quality where sometimes when you stop caring about what everyone else thinks or wants or whatever, and you just pursue this singular vision, it ends up becoming far more appealing. So I mean, again, the thing too here is directionally, I just want computers to be this way. So if Omarchy should end up being just a little footnote in history that it was part of this early movement of the agentic OS and the malleable computer, that’s also okay. I’m completely at peace with that. As long as I get to have a computer that is as fun to work with as Omarchy — that’s great. If a better Rails comes along, I’m gonna use it. And I think, and I love Ruby, and I will continue to write Ruby even just for the sheer fun of it, like I would go ride a horse. I have in the stables, even though I have a Model Y in the garage. The language now is English. It’s a cliche, but it’s also true. Like I’ve been programming in English for the last three months.

Lex Fridman

现在我们得以把它带进人类文明的自然演化：我们的赛博格未来。别像机器人那样跟它说话。像写诗那样跟它说话。自然语言的力量是：通过歧义仍能携带大量意义而不过度规定。对面的智能实体通过解释可以装进去、整合，传达不直接在字里、而在字与深度语境里的高带宽信息。

And now we get to carry it into the natural evolution of human civilization, which is our cyborg future. So don’t talk to it like a robot. Talk to it like you would write a poem. Yeah. And that’s where, like, the power of natural language is, that you can, through ambiguity, still carry a lot of meaning without overspecifying. And the intelligent entity on the other side, through interpretation, can, like, load it all in, integrate it in a way that you can actually convey the high bandwidth information that’s not directly in the words — but in the words, given the deep context.

DHH

我认为完全说对了，这是很多程序员对 AI 的根本误解：他们希望它是确定性的。不不不。Temperature 是 AI setup 最美的部分。它非确定性的事实、创造力需要路上的小微调、如果人脑完全确定就不会有创造力的事实——那才是魔法。

I think that’s exactly spot on, and this is the fundamental misunderstanding that a lot of programmers have of AI, is that they wish it was deterministic. No, no, no. Temperature is the most beautiful part of the AI setup. The fact that it is not deterministic, the fact that creativity requires little tweaks in the road, that the human brain, if it was perfectly deterministic, would not be the creative thing — that’s the magic.

Lex Fridman

你提到 Omarchy 受到一些批评，那我们再往批评那条路走远一点。你让一些政治意见为人所知。大概可以放进移民或非法移民类别，或者——大规模移民、人口结构在形成文化、社会、国家中的角色，你为此挨了不少批评、制造了不少戏剧。你后悔那些戏剧吗？你大体怎么想？

You mentioned getting some criticism for Omarchy, so let’s walk further down that path of criticism. You have made some of your political opinions known. I guess you could broadly put in the category of immigration or illegal immigration, or what— Mass immigration, the role of demographics in the formation of a culture, of a society, of a nation, and you’ve gotten quite a lot of criticism for that and created a lot of drama. Do you regret any of that drama? Like, what’s your general thinking about—

DHH

我会说大规模移民。原因是 Overton 窗口不会自己打开。它一次被推开一点，靠的是人冒一点险。一点声誉、一点反击、一点批评，或者有时很多声誉、很多批评、很多反击。我认为欧洲大规模移民问题多年来是彻底禁忌，在几个欧洲国家仍是。丹麦其实相当早。有些可以追溯到一个人，Mogens Glistrup，非常古怪的角色。我记得 80 年代电视上看到他谈大规模移民的危险。那时——我在 80 年代长大，在 Brønshøj 那个街区，大概 84 年，99% 是族裔丹麦人，能把血统追溯到丹麦人。然后 90 年代初我想到了 5% 之类，到现在大概掉到 60 多或 70%。那是大变化。丹麦人大约在 90 年代开始注意到或谈论：这不是无条件的福气，大规模移民有 downside。辩论在 2000 年代真正咆哮，邻国没有。瑞典尤其根本没有那场辩论。彻底禁忌。挪威也一样。

I’d say mass immigration. And the reason I say that is that the Overton window does not open itself. It opens one nudge at a time by people risking a little. A little reputation, a little pushback, a little criticism, or maybe sometimes a lot of reputation or a lot of criticism or a lot of pushback. And I think the question of mass immigration in Europe was for many years this total taboo, and it still is in several European countries. Denmark really early. And some of it can be traced back to one man, Mogens Glistrup, who was a very quirky character. I remember him, seeing him on the TV in the ’80s talking about the dangers of mass immigration. And this was at a time where — I grew up in the ’80s, and in the neighborhood I grew up in, in Brønshøj, in, what was it? ’84, 99% ethnic Danes, people who could trace their lineage back to Danes. And then early ’90s, I think it goes to 5% or something like that, and then at this point, it’s down to around 60-something or 70% in Brønshøj. That’s a big change. And the Danes started noticing that or talking about that, that this was not an undivided blessing, that there were downsides to mass immigration in about the ’90s. And the debate really roared through the 2000s in a way that didn’t happen in the neighboring countries. So Sweden in particular just didn’t have that debate at all. Total taboo. Same thing with Norway.

因为这种变化已经发生。我觉得你能发现文化里哪里不对劲的一种方式是：你被允许谈什么？你被允许注意到什么，该这么说。我在一篇文章里注意到——大概一年半前我想起伦敦——我 90 年代末开始来伦敦。如果你再往前推 20 年，你会回到和丹麦人差不多的比例。那是族裔英国人占 85、90 百分位的国家——如果你再往回。所以一代人里是不同的国家。注意到那是可以的。也可以想：「你知道吗？我不同意那样。」「我希望不是那样。」我上次去中国是 19 年，2019。我在上海。全是中国人。如果我今年十月回去，突然发现只有 30% 中国人、70% 其他人，我会有点怪。我大概会想「发生了什么？」如果中国人在那一刻说「啊，我们不喜欢那样」，我不会冒犯。我认为有自决的道德论证，我们在世界其他地区一直援引：哦，住在那地区的人对国家有自决权。我认为丹麦人、瑞典人、挪威人完全可以公平地说：「我们在这儿当王国一千年了……」基于能力是关键——因为欧洲多数地方其实几乎没有非法移民。南欧有一些，但欧洲多数没有非法移民的大问题。问题是其他种类的大规模移民，以及你最终得到的人口结构和短短几十年前完全不同。我认为欧洲国家会受益于移民。事实上丹麦有趣的一点是他们对移民有细致统计。

Because this change has happened. And one of the ways I find that you can spot where sort of there’s something that isn’t right in the culture is, like, what are you allowed to talk about? What are you allowed to notice, actually, is how I should put it. And what I noticed in an essay, when was that? Only a year and a half ago as I remember London, was that I started coming to London in the late ’90s. And if you went 20 years before that, you get back to similar rates as what the Danes have. This was like a country of ethnic Brits in the 85s, 90s percentiles — if you get further back. So that’s a different country in a generation. It’s okay to notice that. It’s also okay to think, “Do you know what? I don’t agree with that.” “I wish it wasn’t that way.” I mean, the last time I was in China was ’19, 2019. I was in Shanghai. It was all Chinese. If I, when I go back here this October, suddenly find that there’s only 30% Chinese and then there’s 70% other people, I’d be a little weird. I’d probably be like, “What happened?” And I would not take offense if the Chinese in that instant would go like, “Ah, we don’t like that.” Like there’s just — I think there’s a moral argument for self-determination, and we invoke that argument all the time in other regions around the world that like, oh, the people who live in that region, they have a right to self-determination for their country. And I think it’s completely fair for, say, the Danes or the Swedes or the Norwegians to go like, “Well, we’ve been kingdoms here for a thousand years…” Merit-based is key here because there’s actually virtually no illegal immigration in most of Europe. Southern Europe has some illegal immigration, but most of Europe does not have a big problem with illegal immigration. And the problem is with mass immigration of other kinds, and that you end up with demographics that are just totally different from what they were even a few short decades ago. And I think the countries of Europe would benefit from immigration. In fact, one of the interesting things about the Danes is they carry meticulous statistics on immigration.

Lex Fridman

这个我觉得挺好笑：你提到妻子说的一句话粘在你身上——科技圈附近男人对极端长寿的执着，像女人对厌食，是焦虑和控制感缺失的身体表现。那听起来有点真。

This I found pretty funny, that you mentioned something your wife said that stuck with you, that all this tech-adjacent extreme longevity focus in men is like anorexia in women, a physical manifestation of anxiety and lack of control. That somehow rang true a little bit.

DHH

对很多人当然是。我想那条推特真火了，不知道你注意没有，Bryan Johnson 也进了帖，我能看出 Bryan 会把那读成有点戳。我也认为 Jamie 那次谈话里大概想到他和其他人。这也是个例子：即便我不订阅 Bryan「我想活到 180」之类的使命，我其实想死。我不想永远做这个。我认为人类大约 90 到 100 岁的寿命听起来刚好。我拥抱生命的有限。我们不必在那点上同意。事实上 Bryan 对我是远更有趣的人，因为我们不同意，Bryan 愿意字面把皮肤和其他身体元素押进那使命的游戏。

It certainly did to a lot of people. I think that tweet really popped off, and I don’t know if you noticed, but Bryan Johnson chimed in on the thread, and I could see how Bryan would read that as a bit of a jab. And I do think that Jamie was probably thinking of him amongst other people in that conversation. And it’s also an example of where even though I’m not subscribing to Bryan’s mission of, I actually do wanna die. I don’t wanna do this forever. Like, I think the human lifespan of about 90 to 100 sounds about right. I’m embracing the finitude of life. We don’t have to agree on that. In fact, Bryan is a vastly more interesting human to me because we don’t agree with that, and Bryan’s willingness to literally put his own skin and all sorts of other body elements in the game for that mission.

我昨天和 Jamie 谈这个，她在比较那种「感觉自己活得不够」的感觉。也许他们活得不够的原因之一是现代社会现在是监控国——其实不是国家，是彼此。到处是摄像手机。每一次失态甚至快乐或尴尬的爆发都被捕获。他们大概在拍你，让你在网上永远尴尬，所以也许你根本不该跳舞。因此如果我们都退缩到几乎没在活，我们就死抓着想让它永远持续。我甚至不知道这论题是否适用于 Bryan 或任何别人。但我想有一点：如果你觉得自己真的活过了，你就可以想「我活够了」，有终点不是要对抗的东西。也可能这全是存在主义事后合理化。

I was talking to Jamie yesterday about this, and she was saying or comparing it to this sense of people who didn’t feel like they had lived enough. And one of the reasons perhaps that they hadn’t lived enough was that modern society now is a surveillance state, not by the state actually, but by each other. There are camera phones everywhere. Every indiscretion or even outburst of fun or cringe is captured — they’re probably filming you to your great and eternal embarrassment on the internet, so maybe you just shouldn’t dance at all. And therefore, if we’ve all retracted to the point that we’re barely living, we’re clinging on to wanting it to last forever. Now — I mean, I’m sure that thesis does not apply in all circumstances. I don’t even know if it applies to Bryan or anyone else here. But I think there’s something to this: that if you feel like you’ve really lived, you’re okay thinking I’ve lived enough, and that it having an end is not something to be fought. Now, is it also possible that this is all just existential post-rationalization…

最近我们谈这些过度优化者。我看到 Chris Williamson，Modern Wisdom，有一段像「对，好，也许我们是有点过了」。酒精能提供的社交润滑可能被错过了，正在被错过。我们处于孤独、痛苦、抑郁之类的绝对流行病，相当一部分大概只是因为你和其他人互动不够。也许我们会回头看，或者已经在回头看，想：「呃，你知道吗？偶尔醉一次也许不是世界上最糟的事。」无论醉不醉，也许周五或周六出门时喝两杯……

But just recently, we’ve been talking about this over-optimizers. I saw Chris Williamson, Modern Wisdom had a bit where he was like, “Yeah, okay, maybe we did go a little overboard.” And the social lubricant that alcohol can provide may be missed, is missed. I mean, we are in an absolute epidemic of loneliness and misery and depression and whatever, and a fair amount of it just probably comes because you’re not interacting enough with other people. Maybe we will look back, or maybe we already are looking back and thinking like, “Eh, do you know what? Maybe getting wasted every once in a while was not the worst thing in the world.” And whether wasted or not, maybe just having a couple of drinks every Friday or Saturday when you’re out…

Lex Fridman

尼采式的，你知道，永恒轮回这个想法。如果你做土拨鼠日、永远活着，你希望是在 80 年代橙裤白点里。

So the Nietzschean, you know, this idea of eternal recurrence. If you do a Groundhog Day and you live forever, you want it to be in the orange pants with the white dots in the ’80s.

DHH

对。我想那就是检验。如果你愿意永远重复这一生——所有痛苦、所有喜悦、所有尴尬——那你大概活对了。赛车时我最爱的一件事是：我跌跌撞撞下车，彻底砸烂，几乎抬不起头，躺在车库地板上，只想：「我靠，我还活着。」那是强度。和你爱的人一起创造生命是巅峰。计算机这一刻是精灵从瓶子里出来。我要这一切再发生一遍。

Yeah. I think that’s the test. If you would willingly repeat this life forever — all the pain, all the joy, all the embarrassment — then you probably lived it right. One of the things I always loved about race cars was when I would stumble out of the car absolutely smashed and barely able to hold my head up, and I’d lay down on the garage floor and just think, “Holy fuck, I’m alive.” That’s intensity. Creating life with another human that you love is the peak. This moment with computers is the genie out of the bottle. I want all of that again.

Lex Fridman

感谢收听与 DHH 的这场对话。要支持本播客，请查看描述里的赞助商，你也能找到联系我、提问、给反馈等链接。现在让我用 Ralph Waldo Emerson 的话结束：「一旦你做了决定，宇宙就会合谋让它发生。」感谢收听，希望下次再见。

Yeah. Thanks for listening to this conversation with DHH. To support this podcast, please check out our sponsors in the description, where you can also find links to contact me, ask questions, give feedback, and so on. And now, let me leave you with some words from Ralph Waldo Emerson. “Once you make a decision, the universe conspires to make it happen.” Thank you for listening, and hope to see you next time.