Y Combinator 在 AI 时代的演变

好的。首先，如果你不知道的话，我大约 4 年前搬回了奥斯汀，我是西南偏南的常客。我是奥斯汀的超级粉丝。91 年到 93 年我在这里读商学院，并尽可能经常回到这个办公室。所以，很高兴能帮忙。能跟加里同台并和他交谈，我非常激动。Y Combinator 对创业生态系统的影响，就像她说的，我认为有点不可思议。如果你一开始听到这个想法，很难预料到他们会取得的所有成就。而且我认为也有很多模仿者试图复制 Y Combinator 的成就，但没有任何一个能接近你们所取得的成果。所以，这很了不起。真的。如果你允许的话，我想稍微深入探讨一下。

Okay. So, first of all, if you don't know, I moved back to Austin about 4 years ago and I'm a frequent participant here at South by. I'm a big Austin fan. I went to business school here in '91 to '93 and try to get back to this office as often as possible. So, thrilled to help out. I'm elated to be able to share the stage with Gary and get to talk to him. Y Combinator's impact on the startup ecosystem is, like she said, somewhat miraculous, I think. If you'd heard the idea at the beginning, it'd be hard to anticipate everything they've accomplished. And I think there've also been many look-alike attempts at recreating what Y Combinator's done, and none of them come anywhere close to the output that you guys have been able to achieve. So, it's remarkable. It really is. And if you'll allow me, I want to dive into that just a little bit.

让我们先谈谈申请流程。在漏斗顶端，你可能想要哪些人，而他们自己却不知道应该处于顶端？

Let's start by just talking about the applicant process. Who might you want at the top of the funnel that doesn't know they should be at the top of the funnel?

我想现在一切都变了。我大约 55 天前开始了自己的转型。早些时候，我坐在观众席里，实际上我买不起徽章，所以 2009 年我带着我的创业公司 Posterous 时，无法进入这个房间。我意识到的是，现在一个人可以完成的工作量，我想史蒂夫·耶格最近有篇博客文章说，平均一个 Anthropic 工程师的生产力是 2009 年或那时巅峰时期 Google 工程师的 2000 倍。我认为这就是一切都变了，对吧？所以我们在 20 年里积累了很多规则和东西。YC 实际上上周刚庆祝了 21 周年。其中一条可能就是，我们不再知道简历应该是什么样子了。

I guess everything's changed now. I'm about 55 days into my own transformation. Earlier, I was sitting in the audience, or actually I couldn't afford the badge, so I couldn't be in this room in 2009 with my startup Posterous. What I realized is that now one person can do the work of, I think Steve Yegge had a blog post recently that the average Anthropic engineer is 2,000 times more productive than a Google engineer at the height in 2009 or whenever it was. And I think that's everything just changed, right? So we have a lot of rules, a lot of things that we picked up over the course of 20 years. YC just celebrated 21 years actually last week. And one of those might just be that we don't know what the resume looks like anymore.

所以现在可能正在你脚下发生变化。是的。

So it might be changing under your feet right now. Yeah.

我的意思是，我越来越不在乎某人上过什么大学，甚至在哪里工作过。我实际上想看你的 GitHub 仓库。

I mean, I care less and less what university someone went to or even where they worked at. I actually want to see your GitHub repo.

那么，在新的 AI 赋能申请者出现之前，为了让我们有一个基础认知，在 AI 之前的申请流程中，哪三四个因素能真正让一个人脱颖而出？

So, prior to the new AI-enabled applicant, just to get it in our foundational mindset, what were the three or four things that would really make someone stand out pre-AI in the applicant process?

我的意思是，这些仍然是真的。我不认为有什么改变。只是这些不能是唯一的。我们在网上经常被批评的一个经典标准是，我们真的希望你有一个联合创始人。这有点双刃剑。比如像皮特这样的人，他创建了 OpenClaw。我的合伙人拉斐尔在他被收购前刚刚做了一次很棒的采访。

I mean, these things are still true. I don't think anything changed. It's just that those can't be the only ones. The classic one that I think we get a lot of flak for on the internet is we really want you to have a co-founder. And that one's sort of a double-edged thing. You have people like Pete who created OpenClaw. My partner Rafael just did a great interview with him literally right before he got acquired.

对。

Right.

而那个人是单干的。他做了大约一百个其他项目，纯粹靠数量，他创造了一个轰动世界的东西。所以我们正处于一个不同的时刻。也许 2 年前我们会说你绝对必须有联合创始人。我认为那会对你有帮助。在其他条件相同的情况下，你应该有一个，因为这段旅程漫长、艰难、充满挑战。有联合创始人意味着你并不孤单。我们的现实完全是社会构建的。创业之所以如此艰难，是因为你周围都是不相信你能做到的人，他们认为系统都是固定的，一切都是零和的，你一个个体人类不可能做到。所以，当然，身边有一个优秀的联合创始人会帮助你，让你更成功。

And that was a solo guy. He did like a hundred other projects and just out of pure volume, he was able to create something that set the world on fire. So we're sort of in this different moment. Maybe 2 years ago we'd say you absolutely must have a co-founder. And I think that that will help you. Everything else being equal, you should have one because this journey is long, hard, and difficult. And having a co-founder means that you're not alone. And our reality is totally socially constructed. Starting a startup is so hard because you're surrounded by people who don't believe that you could do it, that the systems are all fixed, that it's all zero-sum, that there's no way that you, an individual human being, could do it. So, of course, having a great co-founder next to you would help you, make you much more successful.

除了联合创始人之外还有什么？什么会让候选人脱颖而出？

What else in addition to a co-founder? What would cause a candidate to spike?

这实际上是一个有趣的问题。鉴于我很好奇你是如何看待投资随时间变化的。我认为风投，包括我自己，人们不断攻击 YC，说‘哦，你们只关心哈佛、斯坦福、谷歌、Meta 的工程师。’我认为这种批评是公平的，我认为我们会摆脱这种看法。我认为这越来越不重要了，因为现在又像，老兄，你一天就能写 1 万行代码。你能做 2000 人的工作。这想想真的很疯狂。

This is actually an interesting one. Given I'm curious how you've thought about how investing has sort of changed over time. I think VC, and myself included, people attack YC constantly about, 'Oh, you only care about the Harvard, Stanford, Google, Meta engineer.' And I think that critique is fair, and I think we're going to go away from that. I think that that's less and less important because now again, it's like, dude, you can just write 10,000 lines of code in a single day. You can do the work of 2,000 people. Which actually is pretty wild to think about.

Meta 有多少员工？比如在任何特定项目上有几万人。

How many employees are there at Meta? Like tens of thousands of people on any given thing.

是的。我认为我们开始进入另一个时代，我们需要煮沸海洋。我们需要做 10 倍、100 倍，实际上 2000 倍更多的工作。这就是我现在看到的。我仍然能保住我的全职工作，管理着近万亿美元的投资组合。

Yeah. And I think we're starting to enter this other age where we need to boil the ocean. We need to do 10 times or 100 times or actually 2,000 times more. That's what I'm seeing right now. I can still hold down my full-time job managing and running a portfolio of almost a trillion dollars.

是的。

Yeah.

那是我的日常工作。然后晚上，比如早上 6 点到 8 点，或者 6 点到 7 点半，我让大约 10 个工人开始工作，然后它们运行，我全天检查。然后当我回家时，在通勤路上，我在旧金山跳上一辆 Waymo，检查工人们的情况。我能完成五六次修订。所以在任何一天，我都能做我的全职工作，开八九个小时的会，同时在三个不同的项目上完成 1 万行代码。

That's my day job. And then at night, like 6:00 a.m. to 8:00 a.m. or 6:00 a.m. to 7:30, I get like 10 workers going, and then they're going, and then I check in throughout the day. And then around when I go home, on the commute home, I hop in a Waymo in San Francisco and get to check in on the workers. I can get like five or six revs in. So in any given day, I can do my full-time job doing like 8 or 9 hours of meetings and get 10,000 lines of code done on like three different projects right now.

好的。所以，我不是想拖延向 AI 的转变，但我确实有几个关于 YC 流程的快速问题。所以，现在流程是 13 周，对吧？你缩短了一段时间，但又回到了 13 周。如果有人让你总结一下它有什么特别之处，因为它看起来并不长，但它显然对经历过的人产生了巨大影响。那么，这 13 周的秘诀是什么？

All right. So, I'm not trying to delay the move to AI, but I did have just a couple quick questions about the YC process. So, the process is 13 weeks now, right? You shortened it for a while, but you're back to 13. If you were asked to summarize for someone what's so special about it, because it doesn't seem like that long a period of time, but it clearly has a massive impact on the people going through it. So, what's the secret sauce of the 13 weeks?

我想我最喜欢的一点是，人们进来时基本上带着不同的价值观。你进来，我想象当我 23 岁时，我是微软的产品经理，你带着一个想法进来，认为我能做的工作量是固定的。而且实际上我甚至不被允许和我的用户交谈，还有一个包含大约 80 万个 bug 的 bug 数据库。就像，‘哦，别去看那个。’然后在 13 周里，每个人努力的方向就是如何做出人们想要的东西？这实际上就是你第一天得到的 T 恤上写的。这听起来很微妙。听起来你应该能自己做，但再次，我在 YC 学到的，对我也是如此，现实是社会构建的。

I guess my favorite thing about it is people come in basically with a different value set. You come in, and I imagine when I was 23 years old, I was a PM at Microsoft, and you come in with an idea that here's this fixed amount of work that I can do. And actually I'm not even allowed to talk to my user, and there's a bug database with like 800,000 bugs in it. And like, 'Oh, don't go and look at that.' And then basically during the 13 weeks, what everyone is running in the direction of is just how do I make something people want? That's literally the t-shirt that you get on day one. And it sounds subtle. It sounds like you should be able to do it yourself, but again, what I've learned about YC, and this was true for me, reality is socially constructed.

然后当你身边都是同样在努力的人，而且几乎没有其他干扰，你唯一做的就是这件事，你就能跑得很快，而且你学会了跑得快。顺便说一句，我其实离开 YC 做了 5 年种子投资人，把 Initialized Capital 做到了近 10 亿美元的管理规模。我记得做种子轮和种子前投资的时候，我会坐在他们办公室里，有时候一周见好几次，但我始终无法营造出同样的氛围。就像《土拨鼠之日》一样，主角想救那个人，却永远救不了。

And then when you are surrounded by other people, all of whom are trying to do that, and there are very few other distractions, that's all you're doing, you can actually run fast. And you learn to run fast. As an aside, I actually spent 5 years away from YC as a seed investor, built Initialized Capital to be almost a billion dollars under management. I remember doing seed investing and pre-seed investing. I would sit in their office, and we would meet sometimes several times a week, and I could never create the same fabric. It was like a Groundhog Day, where the main character tries to save that guy and he never can do it.

就是在那 13 周里无法营造出同样的氛围？

Couldn't create the same fabric, just in those 13 weeks?

我会很沮丧，因为我见过它成功。我在那 13 周里见过它发生。

And I would get frustrated because I'd seen it done. I saw it happen during the 13 weeks.

其中有多少是共同学习，因为你和其他人一起学所以学得快，又有多少是同伴压力，因为你想和其他人做得一样好？

How much of it is co-learning, so you learn fast because you can learn with everyone else, versus maybe a bit of peer pressure that's put on you by wanting to do as well as everybody else?

哦，我觉得以上都有，对吧？这就像一种非常非常浓缩的大学。而且说实话，好处是，如果房间里有投资人，如果你看到一些公司有巨大的潜力，但是，嘿，如果他们只是稍微加速一点，或者如果他们改变一些关于创业的看法，这就能成。比如他们有所有的工程能力，或者他们每天写一万行代码，但他们做的是错的东西。我的建议就是送他们去 YC，我就是这么做的。我有好几家公司从快要死掉变成了价值数亿美元。

Oh, I think it's all of the above, right? It is like a very, very concentrated form of college. And the good thing is, honestly, if there are investors in the room, if you have some companies that you see incredible potential, but hey, if they just sped up a little bit, or if they changed a few things about what they believed about building a startup, this would work. Like they have all the engineering chops, or they're slinging 10,000 lines of code a day, but they're sort of doing it on the wrong stuff. My pitch would be send them to YC, and that's what I did. And I had multiple companies go from like it was going to die to worth hundreds of millions of dollars.

这种转变是通过成为那个生态系统的一部分而发生的。

That transformation happens by being a part of that ecosystem.

我觉得是这样。而且这需要一个合伙人。在 YC，有一个合伙人会选中你，并在项目期间作为你的对接人。YC 持有公司相当一部分股份。我一直想让 YC 的人完全意识到的是，我们其实把合伙人的名字改成了普通合伙人。所以我们更像你们，比尔。你们有 Benchmark，那是典型的 VC 合伙制。我希望 YC 的合伙人就像你们一样。而且我们实际上也是平等合伙制。

I think so. And it takes a partner. At YC you have one partner who picks you and is sort of your person for the program. YC owns a decent chunk of the company. What I've been trying to get people at YC to be totally aware of is that actually we changed the name of partner to general partner. So we're more like you, Bill. You got Benchmark, which is the quintessential VC partnership. I want YC partners to be just like you guys. And we're actually an equal partnership, too.

我很感激。有哪些元素更像结构化教育？在那 13 周里，你有没有特别想传达给整个群体的东西？

I appreciate that. What are elements that are more like structured education? Are there certain things that you want to get across to the group in the 13 weeks?

说实话，我现在想做的事情是，我刚刚开源了我如何使用 Claude Code，叫做 G Stack。过去几天我获得了大约 9700 个星标。

Honestly, the thing that I'm trying to do now, I just open-sourced how I use Claude Code, called G Stack. I got to about 9,700 stars in the last couple of days.

不错。

Nice.

然后我觉得那可能就是方法。就像我全身心投入其中。显然，Opus 4.5 是 Anthropic 的一个革命性启示。然后我觉得人们会想，一个人能做 2000 人的工作，这真的是非常非常新的。如果你在这个房间里，或者你在听，世界上理解这一点的人还不到 0.1%。

And then I think that that might be the way to do it. It's like I put my body on it. Obviously, Opus 4.5 was a revolution revelation from Anthropic. And then I think people are like, this one person can do the work of 2,000 people, is really, really new. And if you're in this room or you're listening along, it's like less than 0.1% of the world understands this yet.

是的。

Yeah.

所以我把它开源，实际上是一种邀请。就像，这正是对我有效的方法。而且很搞笑，因为我因此收到了很多仇恨。来自那些说“一个 Claude Code 技能”的人的仇恨多得难以置信。这里谁真的用过 Claude Code？

And so my attempt at making this open-source is actually sort of an invitation. Like, this is exactly what worked for me. And it's hilarious because I'm getting a lot of hate for it. It's an unbelievable amount of hate from people saying, you know, a Claude Code skill. Who here actually has used Claude Code?

哇！好的。我是说，你们用技能了吗？

Woo! All right. We got, I mean, you guys use skills yet?

哦，是的。是的，我的意思是，如果你还不明白，去和那些举手的人聊聊。技能只是 markdown，但现在非常不同了，因为计算机实际上能读它，而且它是图灵完备的，你可以让它运行代码。

Oh, yeah. Yeah, I mean, if you do not get it yet, go talk to the people who raised their hand. A skill is just markdown, but it's very different now because it's actually the computers can read it and it's Turing complete and you can have it run code.

是的。所以，实际上，我的意思是，这就是正在发生的转变。所以，我不再抗拒了。我们直接进入下一个话题。

Yep. So, it's actually, I mean, that's sort of the transformation that's happening. So, I'm going to quit resisting. We'll just go straight into the next that.

不，没关系。

No, it's all good.

所以，我在推广我的书时对人们说过一件事，就是这种观念：你是否对某件事如此着迷，以至于学习是免费的？你就是觉得你必须做。我想说，如果你是一个技术创始人，而你不在做 Claude Code 和你说的这些事情，你真的应该质疑自己是否有正确的素质，因为世界正在朝这个方向发展。它发展得太快了，你必须站在前沿。所以，谈谈这个想法，为什么 Garry Tan，你有很多事情要做，为什么你觉得有必要晚上用你的 AI 工具去写代码？

So, one thing I've said to people when I've been out promoting my book is this notion of, are you so fascinated with something that the learning comes for free? You just feel like you have to do it. And I would say if you're a technical founder and you're not doing Claude Code and these things you're talking about, you should really question whether you even have the right substance for this, because the world is moving in this direction. It's moving so fast and you have to be on the edge of it. So, talk through this idea of why Garry Tan, you got a lot going on, why you felt the need to go write code at night with your AI-enabled stuff.

我的意思是，有趣的是，我们一直在尝试，你知道，我们投资了很多前沿的公司，如果你看我们的 Light Cone 播客，我们这几年一直在谈论 vibe coding。

I mean, what's funny about it is we've been trying, you know, we fund a lot of companies way out on the edge, and if you watch our Light Cone podcast, we've been talking about vibe coding for years at this point.

对。

Right.

你知道，有些创业公司提供这种服务已经有一段时间了。我最喜欢的故事其实是 Replit。真的很疯狂。他们大概在 2019 年起步，有点像云托管的 IDE。所以即使你在，你知道，我打赌你的很多孩子其实都在用。它默认安装在 Chromebook 上。这种想法是任何人都能编程，而且你不想陷入设置 Unix 服务器之类的麻烦。然后他们就像这个疯狂的故事，它运行得很好，但在教育领域赚钱非常非常难，然后 GPT 3.5、4 出现了，然后 vibe coding 发生了。Andrej Karpathy 开始说，你知道吗？你应该试试点击“接受所有更改”。没人点那个，但我开始点了，然后它就开始工作了。Karpathy 是个天才，伙计。在过去一周左右，他搞出了 auto research，这是一个新东西，一夜之间他实际上提出了新的机器学习技术和突破，仅仅通过提示，仅仅通过设置一个 harness。所以这有点像现在正在发生的极端前沿。

You know, startups that have provided this service for a while. My favorite story of it is actually Replit. It's really wild. They started, I think in 2019, kind of like an IDE that's hosted in the cloud. So if you're even at, you know, I bet a lot of your kids actually use it. It's installed by default in your Chromebooks. The sort of idea that anyone can code, and you don't want to have to get bogged down in setting up Unix servers and things like that. And then they're like this wild story where it was working well, but it's very, very hard to make money in education, and along comes GPT 3.5, then 4, and then vibe coding happens. Andrej Karpathy starts talking about, you know what? You should just try clicking accept all changes. Nobody clicks on that, but I started clicking on that and it started working. Karpathy is a genius, man. In the last week or so, he's figured out auto research, which is this new thing where overnight he's actually coming up with new machine learning techniques and breakthroughs, just by prompting, just by setting up a harness. So that's sort of like the extreme bleeding edge of what's going on right now.

所以，风投界有很多人认为这是他们职业生涯中最大的浪潮，包括我这个年纪的人，以及比你更年轻的一代。我想你同意这一点？

So, there are many people in the venture community that believe this is the biggest wave in the history of their career, including people my age, your generation younger. I assume you agree with that?

哦，是的，绝对同意。

Oh, yeah, definitely.

是的，所以这就是为什么它如此具有变革性，也是为什么你说，即使你看候选人池的方式，在 21 年后也可能改变，因为这是如此不同，这些技能如此关键。一些风投甚至到了这种程度：如果不是这样的人，他们就不想谈。

Yeah, so that's part of why it's so transformational, and that's why you're saying even how you look at the candidate pool might change after 21 years, just because this is so different and these skills are so critical. Some VCs are taking it to the point where if it's not a guy, they don't want to talk to them.

你们已经到了那种硬核级别了吗？

Are you guys at that hardcore level?

我觉得正在变成这样……大概 80% 到 90% 的公司是软件，大约 10% 到 15% 越来越多地是硬科技。他们在做国防，他们在各种领域工作。

I mean, I guess it's becoming sort of... I'd say 80, 90% of the companies are software, about 10 or 15% increasingly are hard tech. They're working in defense, they're working in all kinds of places.

但即使那样，你也会希望它们具备 AI 能力。

But even in that case, you would want them to be AI-enabled.

确实很疯狂。你知道，Palmer Luckey 在 Anduril 谈到过，这个时代意味着软件不再是制约硬件的东西。坦白说，确实如此。我有很多做硬件的朋友基本上都在 vibe coding。

It's wild, like they are. You know, Palmer Luckey talks about it from Anduril, that this age means that software is no longer the thing that holds back hardware. And frankly, it does. I mean, I have friends in hardware who are basically vibe coding.

Vibe coding 现在仍然带有一种“哦，你不应该那样做”的意味。

Vibe coding right now still has this tinge of 'oh, you shouldn't do that.'

我觉得 Open Clojure 的 Pete 今天早上刚发了推文。他说：“哦，不，不，我们还是别这么叫了。”他有点厌倦打那场仗了。而我会站在另一边。我是说，我喜欢 Pete。我觉得我们就应该这么叫，并且知道我们现在处于一个不同的时刻。你知道，去年或两年前这个词刚被发明的时候，它没什么用，只是炒作。但你知道吗？一切在 11 月左右，比如感恩节前后，突然赶上了。我发现它是因为我一群朋友开始玩它，然后基本上他们在圣诞假期期间就从家人面前消失了。

I think Pete from Open Clojure just tweeted this this morning. He was like, 'Oh, no, no, let's just not call it that anymore.' He's kind of sick of fighting that war. And I would take the other side. I mean, I love Pete. I think that we should just call it that and know that we're in a different moment right now. You know, last year or two years ago when this term was sort of invented, it wasn't that useful, and it was hype. And it's like, you know what? Everything caught up in November, like Thanksgiving or so. And I found out about it because a bunch of my friends started playing with it, and then basically they disappeared from their families during the Christmas holidays, apparently.

你觉得创始人现在 100% 应该上手的三到六个工具是什么？

What are, I don't know, three to six tools that you think founders should 100% be rolling around in right now?

Claude Code 和 Codex 显然是的。有趣的是，我最近每晚只睡 4 个小时。我妻子在前排观众席，她说：“你怎么了，伙计？”我得了网络精神病，但我觉得我认识的三分之一的 CEO 也都有。一旦你试过，你会发现我能够重建我的初创公司，那家公司花了 1000 万美元的风险投资和 10 个人，我为此工作了 2 年，还吃了抗嗜睡药。我记得 2009 年在德州奥斯汀吃莫达非尼，试图找用户、参加派对、找投资人。而有了这场革命，我不需要莫达非尼了。我就是睡不着。我凌晨 4 点睡，早上 8 点醒。我妻子今早问我：“你睡了吗？”我说：“我睡够了。我想多睡，但我不能，因为……”

I mean, Claude Code and Codex are sort of obvious. What's funny is, I haven't... I've been sleeping like 4 hours a night right now. My wife's in the audience in the front, and she's like, 'What's going on with you, man?' I have cyber psychosis, but I think like a third of the CEOs that I know have it as well. And once you try it, you'll realize it's like I was able to recreate my startup that took 10 million dollars in VC capital and 10 people, and I worked on that for 2 years and I took anti-narcoleptics. Like, I remember sort of being on modafinil in Austin, Texas in 2009, trying to look for users, go to parties, and find investors. And I don't need modafinil with this revolution. I'm just up. I slept at 4:00 a.m., I woke up at 8:00 a.m. And my wife this morning was like, 'Did you sleep?' And I was like, 'I slept enough. I wanted to sleep more, but I couldn't because...'

因为你想进去看看进展如何？

Because you wanted to get in and check on how things were going?

是啊，看看那 10 个工人怎么样了。我现在同时进行着三个不同的项目。

Yeah, let's see what's going on with the 10 workers. I've got like three different projects going right now.

为什么是 10 个？为什么不是 100 个？是不是有什么管理开销？

And why 10? Why not 100? Is there some management overhead?

有的。所以，我刚发现的一件事是，与其通过 MCP 使用 Chrome（那太慢了，基本没法用），我花了半小时 vibe coding。我真的只说了这些：“嘿，Claude，我想用 Chrome 测试这个东西。我知道你很聪明，你可以帮我做手动测试。我已经准备好分支了。这是个完整的功能。它就在网页上。去以用户身份登录，做这件事，测试所有东西。你为什么不能做？”然后 20 分钟后，它说：“哦，不行，你的上下文用完了。”我说：“那我能不能直接用 Playwright？你能用 CLI 吗？”意思是，我能不能换个方式做？然后它说：“哦，对，有意思。我们为什么不那样做？”然后我有了一个分支。它做了这个东西。我说：“天哪。”然后它成功了。我做了整个功能，然后因为 Claude 和我发现了这件事，就是：“我们为什么不直接用 Unix，做一个小的命令行工具来包装那个库。”它很快，大约 200 毫秒连续响应。我说：“哦，天哪，这太疯狂了。我不仅能……”我还有其他所有部分。我有的开头部分，实际上是 CEO 模式。所以，作为初创公司创始人我必须做的所有事情，我现在都有一个 Claude Code 技能。所以，有些东西你应该试试。如果你加载 Claude Code，打开一个新东西，然后加载 G stack（在 GitHub 上），你可以直接说：“嘿，我想要一个做某某事的应用。”然后让……比如，我想在 Craigslist 上发帖，我想做一个 iPhone 应用，对吧？你只需打开 Claude Code，加载 G stack，然后在计划模式里说那句话，然后你进入“CEO 审查”计划。我的技能实际上基本上就是“盒子里的 Brian Chesky”。它会说：“哦，我们想做什么？也许我们应该这样做。”它实际上会出去规划，比如：“让我们想想你想做的初创公司的理想形态，然后从那里倒推。然后我们现在能做什么？”但你还会注意到一件事，它引用的所有时间框架，它说：“嗯，这大概需要一个月。”它会直接这么说，而且用的是人类的术语，因为一个人类需要 1 个月才能做 10,000 行代码，因为我即使在莫达非尼下、作为真正 20 小时一天的编码者，平均能写的有用代码可能也就 100 到 200 行。而现在能做 10,000 行，简直疯了。

There is. So, one thing I just figured out is, instead of using Chrome through MCP, which is so slow and basically unusable, I just vibe coded in like half an hour. I literally only had this conversation: 'Hey Claude, I want to use Chrome to test this thing. I know that you're smart and you can do manual testing for me. I have the branch ready to go. It's this whole feature. And it's just on the web. Go log in as a user and do this thing and test all the things. Why can't you do that?' And it was like 20 minutes later and it was like, 'Oh, it's not working. You're out of context.' And I said, 'Well, can't I just use Playwright directly? Can you use a CLI?' Meaning, can I just do it a different way? And it's like, 'Oh, yeah, that's interesting. Why don't we do that?' And so, I had a branch. It made this thing. I was like, 'Oh my god.' And then it worked. I made this whole feature, and then because Claude and I figured out this one thing, which is, 'Why don't we just use Unix and make a little command-line tool that wraps this other library.' It was fast. It was like 200 milliseconds back-to-back. And I was like, 'Oh, shoot. This is crazy. I can not only...' I had all the other pieces. The beginning part that I have, it's actually a CEO mode. So, all the things I had to do as a startup founder, I have a Claude Code skill for now. So, some of it is like, you should try it. If you load Claude Code, open a new thing, and then load G stack, which is on GitHub, you can literally just say, 'Hey, I want an app that does blah.' And then let... for instance, I want to post on Craigslist, and I want to make an iPhone app, right? You just open up Claude Code, you load up this G stack thing, and then you just say that in plan mode, and then you go plan CEO review. And my skill actually is basically like Brian Chesky in a box. It's like, 'Oh, well, what are we trying to do? And maybe we should do this.' And it actually goes out and plans out like, 'Let's think of the platonic ideal of the startup that you want to make, and then let's work backwards from that. And then what can we do right now?' And one thing you'll notice though is all of the time frames it quotes, it's like, 'Well, it'll take about a month to do this.' It'll say that. It'll straight-up say like, and they use human terms, because it would take a human being 1 month to do this, like 10,000 lines of code, because the average number of lines of code that I could write even on modafinil and being a really 20 hours a day kind of coder was maybe 100 or 200 lines of useful code. And to be able to do 10,000 now, it's completely nuts.

那你是否看到现代 AI 创始人正在学习用 Claude 做更多事情，而不仅仅是编码，比如销售、分销、营销以及业务的其他所有环节？

And are you seeing found modern AI founders learning to do more than just coding with Claude so sales distribution marketing all the other elements of the business?

我觉得就是这样。我还没在这方面尝试过，但很可能在座的某个人就能做到。而且这实际上就是我隐约看到的——你知道，有 CEO 模式，有边缘经理模式，然后我还摸索出了 QA 工程师模式，就是让它去使用产品，它很聪明，会截图，会说“哦，这种情况下是这样的，哦，让我写个自动化测试”。这太不可思议了，因为我花了很多年，而且我身边都是那些我整个职业生涯都在打交道的人。有一种神奇的事情是初创公司创始人能做的，比如 DoorDash 的 Tony Shue 就做到了。你知道，我们所有敬仰的那些了不起的创始人，他们做的就是这些。他们只是问非常简单的问题，进行指导，然后大量的金钱、时间和风险投资资金就流向如何找到一个高管，然后他再雇佣下面的 10 个人，然后他们再雇佣成千上万的人。然后，你知道，资本作为大棒在科技界曾经是件大事。我觉得那可能已经消失了。实际上缺失的是……

I think that's it. I think I haven't worked on that yet, but it's probably something that someone in this room could do, and that's going to be actually how I mean what I'm glimpsing at is like, you know, there's a CEO mode, there's an edge manager mode, and then I figured out this QA engineer mode where it's like go use the product and it's smart, it does screenshots, it'll say like, oh well for this case it's this, oh, let me I need to write an automated test for this. It's unbelievable because I spent years and years, and I'm surrounded by people for whom this is all I've known for my entire professional career. There is this magic thing that startup founders get to do, like the Tony Shue at DoorDash got to do that. You know, all of these incredible founders that we all look up to, this is what they did. They just asked very simple questions, they directed it, and then an incredible amount of money and time and venture capital dollars went to like, how do I access an executive who then can hire the 10 people below them, who then can hire thousands of people for that. And then, you know, capital as a bludgeon was sort of a big thing in tech. And I think that might be gone actually. The actual is missing is...

我喜欢这个说法，因为那是我亲身经历时的感受。资本作为大棒。而且在风险投资界有一些结构性的东西，后期阶段的公司越来越大。而且他们做的大多数融资轮都是主动的。所以创始人并没有决定去融资。有人敲门说：“这是 3 亿美元。”至少在我所经历的资本大棒世界里，如果你不接受，他们就会威胁把钱给你的竞争对手。所以你会遇到一种自上而下的情况，钱就这样出现在你家门口。而且，你知道，这即将与你提出的观点交汇，即公司可能不需要那么多员工。那么这怎么解决？钱去哪了？如果你的业务高度自主、高度 AI 驱动，而有人坚持给你 3 亿美元，你会怎么处理？

I love this phrase, because that's how I felt when I was living through it. Capital as a bludgeon. And there are structural things in the venture capital community where the later stage firms are getting bigger and bigger and bigger. And most of the rounds that they do are proactive. So the founder didn't decide to go raise money. Someone knocked on the door and said, "Here's $300 million." And at least in the capitalist bludgeon world that I lived in, if you don't take it, they threaten to give it to the competitor. And so you have a top-down situation where the money just kind of shows up on your doorstep. And this, you know, is about to intersect with the point you're making that the company may not need that many employees. So how is that solved? Where does the money go? If you're highly autonomous, highly AI-driven in all areas of your business, and someone insists on giving you $300 million, what do you do with it?

我的意思是，我想到了，你知道，OpenAI 的 Sam Altman，他的文章《智能时代》真的让我大开眼界，让我明白了我们接下来应该做什么，那就是我认为是时候让我们去“煮沸海洋”了，对吧？就像，现在社会上发生了太多事情。而且我认为，你知道，一些运营前沿实验室的人，也许为了筹集资金，使用了过多的末日论。

I mean, I think about, you know, Sam Altman from OpenAI, his essay of "Age of Intelligence" was what really opened my eyes to what I think we should do from here, which is I think it's time for us to boil the oceans, right? Like, there is so much going on right now in society. And I think that, you know, some of the people who run the frontier labs, maybe in order to raise the money, are using more doomerism than they should.

嗯。末日论怎么帮助他们筹集资金？

Mhm. How does doomerism help them raise money?

我的意思是，它能让人兴奋起来，伙计。因为你把它描述得很神奇，而他们想投资于神奇？

I mean, it gets people going, man. Because you describe it as magical and they want to invest in magic?

是这样吗？我觉得人们……嗯，我想说我喜欢投资于神奇，而且我相当确定你投资过一些……我的意思是，神奇就是那些令人难以置信、以至于超出我们理解范围的东西。比如市场，你知道，10 到 20 年前。你投资了所有最好的市场。随便哪一个，比如你第一次用 Uber，第一次用 DoorDash，第一次用 Instacart，这些市场对第一次使用的人来说都像是天启，对吧？

Is that... I think people... Well, I would say that I like to invest in magic, and I'm fairly certain that you have invested in things that... I mean, magic is just things that are so incredible that they strain our understanding of how they could be possible. Like marketplaces, you know, 10, 20 years ago. You invested in all the best marketplaces. And any given one, like the first time you used Uber, the first time you used DoorDash, the first time you used Instacart, any of these marketplaces were like revelations for anyone using them for the first time, right?

但我们从没说过 Uber 会在 3 年内治愈癌症，这……

But we never said that Uber would cure cancer within 3 years, which...

说得对。

That's fair.

……而这正是其中一些人在说的。

...which is what some of these people are saying.

嗯，你知道，有趣的是……那是积极的。是的。是的。

Well, you know, what's funny about that... that's positive. Yes. Yes.

我喜欢他们这么说。然后我认为可能错误的是，我觉得有杰文斯悖论。我认为我们会发现，当有更多智能时，它不是零和的。它是让我们能做更多事情的东西。所以我要冒险说一些挑衅的话。我认为这一切意味着，就像在座的各位和正在收听的人们，我们在这里是为了努力让《星际迷航：下一代》成为现实。真的。我思考过。就像，我认为我们需要一个后稀缺经济。而且我认为，你知道，治愈癌症是，我的意思是，在座的每个人都认识受其影响的人，或者很大一部分人会得癌症，对吧？这是一个人类问题。但像这样的人类问题可以通过智能来解决，通过理解癌症、理解疾病，然后真正让它消失。这就是技术。所以这就是为什么，我的意思是，能够创造那种技术的人，我们必须对我们正在努力做的事情提出一个肯定的观点和肯定的声明。比如，我们不是要消灭所有的工作。我们不是要颠覆社会。我们是在努力真正治愈、解决问题，并真正互相帮助。

I like that they said that. And then what I think is maybe the wrong thing is like, I think there's Jevons paradox. I think that we will find, when there is more intelligence, it is not zero-sum. It is something that allows us to do a lot more of it. And so I'll go out on a limb and say something provocative. I think that what all of this stuff means is, like the people here in this room and the people listening, we are here to try to make Star Trek The Next Generation happen. Literally. I think about it. It's like, I think that we need a post-scarcity economy. And I think that, you know, curing cancer is something that, I mean, everyone in this room knows someone who was touched by it, or a large percentage of the people in this room will get it, right? And it's something that's like a human problem. But a human problem like that can be solved through intelligence, through understanding cancer, understanding disease, and then actually making it go away. And that's technology. And so that's why, I mean, the people who can create that technology, we have to come out with an affirmative view and affirmative statement about what we're trying to do. Like, we're not trying to kill all of the jobs. We're not trying to upend society. We are trying to actually heal, solve problems, and actually help one another.

我和 Gary 过去讨论过一件事，你们听众可能不知道，你可以直接问你最喜欢的 AI，它会告诉你，但确实有针对 AI 焦虑的调查和民意调查，在美国，大约 70% 是负面的，而在中国，大约是 20%。所以对于这是积极还是消极，存在完全不同的心态，我确实认为这部分是由那些拥有最大话筒的人推动的，他们是最悲观的末日论者，这有点疯狂，这是我在风险投资历史上从未见过的。

One thing that Gary and I discussed in the past that you the audience may not be aware of, you can just ask your favorite AI and it'll uncover this, but there are surveys and polls done on anxiety about AI, and in the US, there's like a 70% negative, and in China, it's like 20%. And so there's a completely different mindset about whether this is positive or negative, and I do think it's partially driven by the people that had the loudest microphones being the biggest doomerist, which is kind of a crazy reality, which is something I've never seen before in my history of venture.

是的。我的意思是，我认为这里棘手的是，我有点理解。比如他们可以争辩说他们必须这样做。因为否则你怎么能得到一万亿美元？但我不明白为什么消极会创造融资，就我个人而言，但我认为还有另一个目标，但让我试着间接地解决它。

Yeah. I mean, I think the tricky thing here is like I kind of get it. Like they could argue they had to do it. Because how else are you going to get a trillion dollars? But I don't understand why being negative creates fundraising, personally, but I think there's another goal, but let me try and get at it indirectly.

不，我很好奇。是的，不。那么所有转向这些 AI 工作负载的最年轻的初创公司，他们底层使用哪些模型？

No, I'm curious. Yeah, no. So all of the youngest startups that are moving to these AI workloads, which models are they using underneath?

我的意思是，主要是 OpenAI 和 Anthropic，然后有趣的是，人们开始使用越来越多的开源模型，而且我很惊讶没有美国的开源模型。

I mean, OpenAI and Anthropic, mainly, and then interestingly, people are starting to use more and more open-source models, and I'm surprised that there isn't an American open-source model.

我们可以谈谈这个，但是什么导致他们转变到那个地步？

We can talk about that, but what causes them to get to that place where they shift?

通常实际上是用户。对。

Usually actually users. Right.

不过，你知道，现在疯狂的事情是，实际上我认为我们处于一个不错的位置。

Though, you know, the crazy thing now is like actually I think we're in a good spot.

大概一两年前，我非常担心会出现快速起飞，会有一两个实体基本上掌握 AGI。

Like a year or two ago, I was really, really worried that there'd be fast takeoff, there'd be one or two entities that basically get the AGI.

嗯。

Yeah.

我觉得我们其实处于一个非常好的位置。我的意思是，即使我把所有时间都花在智能体流程这类事情上，我可以坐在那里用 Claude，但两秒钟后我可能就在用 Codex，而且效果一样好。不仅如此，我还能问 Claude Code：“嗯，为了这个目的，我该用哪个？”比如 Anthropic 擅长超级智能的东西，但他们没有低延迟模型。所以最适合那个场景的其实是 Grok 上的 Q 和 Llama 70B，比如 3 版本实际上大约 200 毫秒。那是 Grok 上的一个开放模型，那是最好的。而且关键是，还会有另一个更好的，有时候是中国的模型，那也没关系。所有这些其实都很棒。我觉得在技术发展的这个时刻，我们真正需要的是竞争。

I think we're actually in a really, really good spot. I mean, even spending all of my time working on Agentic Flows and things like that, it's like I can sit there and use Claude, but I could be using Codex 2 seconds later, and it's as good. And not only that, I can ask Claude Code like, 'Well, for this purpose, should I be using...' Like Anthropic is great for super smart stuff, but they have no low latency model for instance. So the best one for that is actually on Grok with a Q and Llama 70B, like version 3 is actually 200 milliseconds or something. It's an open model on Grok with a Q and that's the best one. And the thing is there will be another one that's even better and sometimes it's the Chinese model and that's okay. All of that is actually great. I think what we really, really want at this moment in time in technology is we need competition.

嗯，所以这就是我要说的：如果你回顾互联网繁荣时期，比如 96、97 年，你 30% 的风险投资都投给了 Sun 和 Oracle，因为每个初创公司都想要最好的技术，他们不希望服务器宕机。五年后，100% 都是 Linux 和 MySQL。有一个创新阶段。这就是我问你转折点是什么时候的原因，那时他们有了真正的用户。然后有一个优化阶段，你开始担心性能和价格，而不仅仅是产品。所以我确实看到我们的大公司，当收入真正开始进来时，他们开始尝试不同的模型。也许不是全部，也许 60% 的 token 之类的。但这是绕了个弯子，我想说的是，我认为悲观主义的本质是他们想要监管捕获。但我已经说了好几年了。

Well, so this is where I would get at this, which is if you go back to the internet boom, right, 96-97, 30% of your venture dollars went to Sun and Oracle because every startup wanted the best technology. They didn't want the server to go down. Five years later it was 100% Linux and MySQL. There's an innovation phase. That's why I asked you when the point of change was when they had real users. And then there's an optimization phase when you start worrying about performance and price as opposed to just the product. And so I'm certainly seeing with our bigger companies when the revenue really starts to come in, they start experimenting with different models. Maybe not for the whole thing. Maybe for 60% of the tokens or something like that. But this is a roundabout way of getting to what I think the doomerism's all about, which is I think they want regulatory capture. But I've been saying that for years.

是的，我们一直在与之抗争。所以我认为很多真正想提升性能和成本的公司都在尝试开放模型，其中很多是中国的。我觉得有趣的是，看看那些大模型公司是否会试图利用华盛顿来终结这一切，从我的角度来看，这对初创公司不利。

Yeah, we've been fighting against that for a while. Yeah, so I mean I think many of the companies that are really trying to boost performance and cost are experimenting with the open models, many of which are Chinese. And I think it's going to be really interesting to see if the big models try and use Washington to put an end to that, which would be bad for startups from my perspective.

绝对。我的意思是，与此同时，我希望 Anthropic 和 OpenAI 很可能类似于 Meta、Google 和 Apple，这是大科技的经典模式。我的意思是，这是 YC 的一个老生常谈。我们想为小科技而战。我们想真正为这个房间里需要分发的人而战。过去 5 到 10 年没有消费级初创公司的部分原因是 Google 和 Meta 基本上锁死了所有。我猜如果我是他们，我也会尝试这样做。但与此同时，当你有一个第二价格拍卖模型时，你可以从每个消费业务中榨取所有毛利润。所以结果，生态系统怎么了？这将会重演。这就是为什么我认为我们需要继续支持开源。我们需要继续支持开放权重模型，因为那实际上是制衡我噩梦的东西。想象一下，你正在使用……我不会点名前沿模型实验室。老实说，可能是任何一家。可能是一个现在还不存在的。

Absolutely. I mean, simultaneous to that, I'm hoping that Anthropic and OpenAI are highly likely to be similar to Meta and Google and Apple, the classic thing for big tech. I mean, this is a whole hobby horse for us at YC. We want to fight for little tech. We want to actually fight for the people in this room who need distribution. Part of the reason why there's no consumer startups going on like five or 10 years is like Google and Meta basically locked it all up. And I guess if I were them, I would try and do that too. But at the same time, when you have a second price auction model, you can extract all the gross profit out of every consumer business. So as a result, what happened to the ecosystem? And it's going to play back again. And that's why I think we need to continue to support open source. We need to actually continue to support open weights models because that's actually the way that is a counterweight to, frankly, what I have nightmares about. Like imagine you're using... I won't name the frontier model lab. It could be any of them honestly. It could be one that doesn't even exist right now.

提高价格，你无法切换，就像 Office 365。

Raise prices and you can't switch like Office 365.

嗯，或者就像 90 年代的微软。比如，“哦，你这里生意不错。我想我们干脆……”

Well, or it's like Microsoft back in the 90s. Like, 'Oh well, what a nice business you have there. I guess we'll just...'

好吧。对吧？我的意思是，Cursor 曾是 Claude 最大的客户，现在他们正面竞争。

All right. Right? I mean Cursor was Claude's biggest customer and now they compete head-on.

没错。他们已经在向上游移动。我认为两家模型公司都意识到，如果没有终端客户连接，他们就有被替换的风险。所以他们很聪明，而且他们有很多钱。

That's right. They're already moving up the stack. I think both model companies realized that if they didn't have end customer connectivity, they were at risk of being switched out. So they're being smart and they have tons of money.

是的。所以我的意思是，在 YC，我们对此的思考有点非正统。我们相信自由市场，但一个运转良好的自由市场实际上需要竞争。有时政府确实必须介入，确保……我鼓励，这可能不是解决这个问题的地方，但我鼓励所有认同这一信念体系的人，也许创建一个联盟，帮助为开放模型提供法律支持和保护，因为我觉得事情会变得激烈。

Yeah. So I mean at YC we're kind of heterodox in our thinking around that. We believe in free markets, but a functioning free market actually requires competition. And sometimes the government does have to step in and make sure that... I would encourage, this probably ain't the place to solve it, but I would encourage everyone who shares this belief system to maybe create a consortium that can help provide legal support and protection for open models, because it's going to get spicy, I feel.

听起来没错。好吧，在 X 上找我 @Garry Tan，如果发生什么事，就把他们引到我这里，我们会找到帮助。我们必须在这里为小科技发声。

That sounds right. Well, find me on X at Garry Tan, and if something happens, just send them my way and we will find help. We have to stand for little tech here.

那么让我们……你觉得 AI 是……我们谈到了 AI 赋能的创始人将会变得不同的所有方式，现在你听起来要在申请过程中开始测试这一点了。

So let's... Do you feel AI is... We talked about all the ways that AI-enabled founders are going to look different, and now you're going to start testing for that at the application process, it sounds like.

那是我的项目之一。现在每天大约 3000 行代码。

That's one of my projects. It's like 3,000 lines of code a day right now.

好的。是的。那么你认为公司类型会有什么变化？

Okay. Yeah. And so what do you think will change about the types of companies?

哦，天哪。我的意思是，我会……

Oh god. I mean I would...

你们在收到申请时，到底有多在意公司创意？

How much do you guys even care about the company idea when they get the application?

我们不太在意创意，更在意一个人能做什么。但我的意思是，清晰的沟通很重要。我认为，无论惊讶与否，这实际上是产品经理/设计师创始人的时代。我是技术出身。我必须成为一名工程经理，雇佣软件工程师，自己写很多代码，但我现在如此有用的原因是，我可以和 10 个不同的员工交谈。也许下周希望能到 12 或 13 个，如果 G stack 继续朝着正确的方向发展。如果我能编排 10 个，或者从 10 个到 20 个员工，我认为这是可能的。但我能做到，因为我做过那份工作。这有点像经典的最大值，在我 16 或 17 岁时，我在一家名为 Adjacency 的设计公司实习。创始人实际上为史蒂夫·乔布斯创建了第一家苹果电商商店。

We don't care about the idea as much nearly as much as what someone's capable of doing. But I mean, I think clear communication matters a lot. I think, surprisingly or unsurprisingly, I think this is actually the age of the product manager / designer founder. I am technical. I had to be an engineer manager and hire software engineers and shipping a lot of code myself, but the reason why I'm so useful right now is like I can talk to 10 different workers. Maybe next week hopefully it'll be 12 or 13 if G stack still keeps going the right way. If I can be like orchestrating 10 or go from 10 to 20 workers, I think that's in the cards. But I can do that because I've done that job. That's sort of like the classic maximum for when I was 16 or 17 years old, I was interning at a design firm called Adjacency. The founder actually created the first Apple e-commerce store for Steve Jobs.

嗯。

Yep.

我说：“安德鲁，我怎样才能成为像你一样的人？我怎样才能成为 CEO？”他带我看了所有不同的工作，他说：“嗯，看到那个人，你知道，这个克里斯汀，她是销售经理。她就是那个，你知道，这就是她整天做的事情，对吧？”我必须做那份工作，然后我雇她去做那份工作。这就是我如何知道该雇谁。

And I said like, 'Andrew, how do I become like you? How do I become a CEO?' He took me around all the different jobs and he's like, 'Well, see that person, you know, this Kristen, she's a sales manager. She's the one who, you know, this is what she does all day, right?' And I had to do that job and then I hired her to go do that job. And that's how I knew who to hire.

我知道怎么管理她，也知道如果事情不对劲该怎么修。这就是 CEO 的工作。你必须了解所有不同的部分。我认为这正是当下这个时刻的完美隐喻。你需要有自主性和品味，同时你还应该是个通才。我以前会给大学生建议，说你应该一英里深，其余一英里则一英寸深。但现在社会需要的形态已经大不相同了，因为智能已经触手可及。关键在于能在这个广阔的平面上建立联系。如果你能一英尺深地横跨所有领域，智能体会填补其余的一切——那些已经在互联网和开源中存在的完整智能。

I know how to manage her and I know, you know, if things aren't going right, I know how to fix it. And that's the job of the CEO. You have to know all the different parts. And I think that's the perfect allegory for this moment right now. You need agency and taste, and also you should be a polymath. I used to give advice to college kids like you should be a mile deep and an inch deep for the rest of the mile. But now the shape of what society needs is very different because intelligence is on tap. It's about making connections across this whole wide plane. If you can be a foot deep all the way across, the agents will fill in everything else—the full intelligence that's already on the internet and open source.

所以你在预测，未来的创始人会在这方面有所不同，品味、审美以及这种设计优雅的微妙之处，将比能写 2 万行代码更具区分度。

So you're making a prediction that the founder of tomorrow will be different in these ways, and taste and aesthetic and this kind of nuance of elegance of design will be more differentiating than being able to write 20,000 lines of code.

是的。我可能是个英语专业的学生，老兄。这其实很棒。这是好消息。

Yeah. I might be an English major, man. Which is great, actually. That's the good news.

我们回到公司类型的话题。你认为在 AI 时代这会发生转变吗？你会看到公司解决与过去不同的问题吗？

We're going back to just the types of companies. Do you think that will shift in an AI age? Will you see companies attacking different problems than they did in the past?

我认为一切都一样，只是会更快、更多。这就是我思考当今公司运作方式的地方。我经常回想在微软的日子。我做过一个叫 ActiveSync 的产品，它让你的 Windows 手机与 Windows 设备通信，获取联系人、日历和邮件。与此同时，在库比蒂诺的办公室里，有一个两人团队在开发后来用于 iPhone 的整个移动版 Safari。我认为，2 万人被 2 人或 20 人击败，这一直是事实。在微软，全是政治，全是封地。我记得当时需要一项同步服务。我的产品经理导师和我去了 Windows 办公室，他带了一根棒球棒。我问：“为什么我们带棒球棒？丹，我们为什么要这样？”我们在那个部门无法修复任何 bug。你把这种无法服务用户的 dysfunction 乘以 2 万人，难怪我们的软件很糟糕。我最喜欢的例子：这里谁用 Siri？没人，对吧？这里谁对——我爱我的 iPhone，硬件很棒，很漂亮，电池也很好。但日历应用是我这辈子见过的最差的应用。这甚至不是钱的问题。这是现在整个社会都在发生的怪事。苹果有无限的资源；他们可以雇任何人，做任何事。看看他们的现金储备。我最喜欢的彼得·蒂尔名言：他著名地与埃里克·施密特坐在一起，说：“谷歌没有想法。”你为什么有 400 或 500 亿美元的现金？这些是晚期科技控股公司，不再是科技公司了。他们被护城河束缚。我们的 iPhone 里仍然没有好的大语言模型与 Siri 集成。Alexa 有数万人，却依然很糟糕。

I think it's all the same and it's just going to be faster and more. That's where I think about how companies operate today. I think about my time at Microsoft all the time. I worked on a product called ActiveSync, which was how your Windows mobile phone would talk to your Windows device to get contacts, calendar, and email. At the same time, in an office in Cupertino, there was a team of two people working on all of mobile Safari that went into the iPhone. I think that 20,000 people can be beat by two people or 20 people has always been true. At Microsoft, it was all politics, all fiefdoms. I remember needing a synchronization service. My PM mentor and I went over to the Windows office, and he brought a baseball bat. I asked, 'Why do we have a bat? Dan, why are we doing this?' We couldn't get any of our bugs fixed over in that other org. You multiply that dysfunction—not being able to serve your user—by 20,000 people, and no wonder our software is terrible. My favorite example: who here uses Siri? Nobody, right? Who here is frustrated with—I love my iPhone, the hardware is great, it's beautiful, the battery is great. But the calendar app is the worst app I've ever seen in my life. It's not even about money. That's the weird thing happening all across society. Apple has unlimited resources; they could hire anyone, do anything. Look at their cash board. My favorite Peter Thiel-ism: he famously sat with Eric Schmidt and said, 'Google has no ideas.' Why do you have 40 or 50 billion dollars in cash? These are late-stage technology holding companies, not tech companies anymore. They're held back by the moats. We still don't have good large language models in our iPhone with Siri. Alexa has tens of thousands of people and it's still terrible.

当人们告诉我新的 Claude 工具在 Excel 上远超 Copilot 时，我就想，是的，同样的概念，大公司已经被警告两年这是巨大的浪潮，却仍未找到实现的方法。这有点疯狂。是的。

When people tell me the new Claude tool for Excel runs circles around Copilot. So, yeah, same concept that the big companies now have had two years of being told this is a huge wave and they still haven't found a way to get there. It's kind of wild. Yeah.

就像它有了 8 年的先发优势，然后就在等待。所以，我认为这也许是对在座各位的提示。是的，曾经有护城河。这一直是事实。十年前、五年前、二十年前，车库里的小初创公司能勉强匹敌的唯一方式就是原始的攻击性。我们如何做到 10 倍或 100 倍？我如何与我的前一千名用户建立深厚的个人关系？这实际上比以往任何时候都更容易实现。而这正是我们所需要的。

Like just what it had an 8-year head start and then it's just waiting. So, I think maybe that's the prompt for the people in this room. Yes, there were moats. This is always true. Ten years ago, five years ago, twenty years ago, the only way a tiny startup in a garage could even hold a candle is raw aggression. How do we do 10 times or 100 times more? How do I have a deep personal relationship with my first thousand users? That's actually more accessible now than ever. And that's exactly what we need.

我非常好奇 OpenAI 或类似的公司是否会重新创建日历、联系人数据库或电子邮件，只是为了辅助所有这些功能。

I'm very curious if OpenAI or someone like that ends up recreating a calendar or contact database or email just to sidecar all that stuff.

是的。我想如果你们喜欢电影《瞬息全宇宙》，我的一个——

Yeah. I guess if you guys like the movie Everything Everywhere All at Once, one of my—

它在这里的西南偏南电影节首映。

Which debuted here at South by Southwest.

是的。我觉得那部电影让我感受到，一个寻找真正伟大想法的创始人可以在脑海中装载什么。跨越所有时间、所有不同的人、所有不同的行业，总有事情在发生——沮丧、无聊、愤怒。如果你能设身处地地理解别人的处境，那么需要完成的工作是什么？我认为大科技公司不会问这个问题，因为他们有护城河，他们自我偏好，他们不必为此努力。这就是为什么一些公开的自我偏好行为应该被定为非法。我们需要允许新进入者——在座的各位和听众——创造新技术，并与另一端的人建立直接关系。我们不想与大型匿名企业合作。我们正处于这样一个时刻：也许甚至不是软件，也许是智能体、智能体编排，帮助社会中的每个人拥有自己的个人智能体。现在，我们依赖像 Open Cloud 的皮特这样的人来创建这个新包，但这可以为世界上每个行业构建。

Yes. I feel like that movie makes me feel like what a founder looking for really great ideas can sort of load in their head. Across all this time, across all these different people, across all these different industries, there's stuff happening—frustration, boredom, anger. If you can step into the shoes of other people and understand what's going on for them, what are the jobs to be done? I think big tech doesn't ask that question because they have moats, they self-preference, they don't have to work hard for this. That's why some of the self-preferencing openly should just be illegal. We need to allow for new entrants—people in this room and people listening—to create new technology and have a direct relationship with the people on the other end. We don't want to work with big faceless corporations. We're at this moment where maybe it's not even software. Maybe it's agents, agent orchestration, helping everyone else in society have their own personal agent. Right now we rely on someone like Pete from Open Cloud to create this new package, but that could be built for every industry in the world.

比如看看，你知道，一个思维练习。

Like look at the, you know, one thought exercise.

而且这一定会发生，是板上钉钉的事。你看 GDP 这块大饼，里面真正有软件成分的占比有多少？你知道，即使在 AI 之前，我估计也就 10%、20% 左右吧？世界上大部分商业活动仍然通过文本、电子邮件、电话、传真进行。我是说，真的就是电子邮件，我觉得全球航运业主要还是靠来回发邮件和 Excel 表格报价。所以，你知道，现在确实如此，但未来不可能还是这样。

And it will be. It's like guaranteed to happen. It's like there, look at the GDP pie and like what percentage of that actually has any software in it at all? Like you know, even before AI, it was like I don't know, like 10, 20% maybe? Like the bulk of business in the world still happens over text, email, phone, fax. Like I mean, literally email, like literally I think the world's shipping industry is still mainly using emailing Excel spreadsheet quotes back and forth. And so, you know, that's true right now, but it couldn't possibly be true in the future.

在你们开始写代码之前，我有个技术问题想问你。你觉得我们会达到一个数据库对用户来说变得模糊不清的地步吗？就像今天的 SaaS，他们怎么说来着，在 CRUD 数据库之上？你可以吐槽它。但所有这些系统底层都有一个确定性的数据库。你觉得我们会达到不再关心底层具体是什么数据库、而是信任智能体去操作数据的程度吗？

And before you arrive at coding, I have a technical question for you. Do you think we reach a point where the database becomes obfuscated to the user? Like in SaaS even today, they call it what's the saying on top of a CRUD database? Like you can take shots at it. But underlying all these systems have an underlying database that's deterministic. Do you think we'll get to the point where we won't have awareness of the exact database that's underneath there and trust the agents to manipulate the data?

我个人不这么认为。我觉得 SQL 数据库会永远存在。而且我也有过这种经历——也许错的是我。比如今天早上我和团队里一个工程师争论。他说：‘嘿，我想让 AI 智能体帮我做一堆管理任务。’

I mean, personally, I don't think so. Like I think you're going to have SQL databases forever. And I mean, I had this—I mean, maybe I'm the one who's wrong. Like I was debating with an engineer on my team this morning. It's like he said, 'Hey, I want to make the AI agent be able to do a bunch of admin tasks for me.'

我是说，你知道，连 Karpathy 我都记得他说过，他的自动研究智能体以完全权限运行，像危险地跳过权限，把整个项目删了。是的。然后他不得不重新创建，对吧？

I mean, you know, even Karpathy I think said that his auto research agent that was running with like full permissions, like dangerously skip permissions, deleted the whole project. Yes. And he had to like recreate it, right?

但没花多长时间。

But it didn't take very long.

是啊。没花多长时间，这倒是挺好的。

Yeah. It didn't take very long, which is nice.

所以，你知道，我也不确定。我觉得数据库很重要，访问控制很重要，然后你知道，LLM 很了不起，但其中一些东西，我其实很好奇，也许这不是最终形态。就像宝可梦一样，外面还有一两个等级。

So, you know, I guess I don't know. Like I think that databases are important, access control is important, and then you know, LLMs are amazing, but like some of it is like I actually am curious like maybe it's not the final form. Like kind of like Pokémon, there's like another level or two out there.

嗯。

Yep.

而且我，你知道，坦白说我希望如此。说实话，我们在某种程度上不得不建造非常非常低效的数据中心来获得这一切，这真的很疯狂。我的意思是，智能如此惊人，以至于我们愿意付出几乎任何代价。我是说，Anthropic 的模式，谁会想到人们愿意付那么多钱，但老实说，既然我已经拥有了，就像给我 10 倍快速模式，我愿意付 10 倍的钱。

And like I, you know, I frankly hope so. Like it's actually really wild to what degree we have to build very, very inefficient data centers in some sense to get this. I mean, the intelligence is so astonishing that we're willing to pay almost any price. I mean, Anthropic's mode, it's like who would have thought that people would be willing to pay that much, but honestly now that I have it, it's like give me 10x fast mode, I'd pay 10 times as much.

是啊，是啊。

Yeah, yeah.

在我们进入观众提问之前，让我们谈谈业界担心或关心的几个大话题。永远保持私有。你怎么看这件事，尤其是 Y Combinator 内部有一个投资部门？

Before we go to audience questions, let's hit on a couple big topics that the industry is worried about or caring about. Stay private forever. How what's your opinion of this, especially since there is an investment arm inside of Y Combinator?

是的，我认为这是真的。我的意思是，如果你和很多 YC 创始人谈过，比如 Stripe 或 Rippling 的创始人，他们是了不起的人，非常有雄心壮志，而公开市场有时并不帮助、并不真正支持这样的创始人。所以保持私有，老实说对他们来说是有道理的，但我希望他们，你知道，上市。

Yeah, I think it's true. I mean, if you talk to a lot of the founders who are YC founders, I mean, you talk to the founders of Stripe or Rippling and people like they're incredible people who are very, very ambitious, and then the public markets sometimes do not help, do not really support founders who are like that. And so staying private, honestly kind of makes sense for them, but I would love for them to, you know, go public.

Y Combinator 有没有在私有公司中实现流动性？

Has Y Combinator gotten liquid in a private company?

呃，是的，是的。我的意思是，这是很奇怪的事情，对吧？比如 Scale AI 的退出，你知道，但我是说，你知道，只是参与二级交易。呃，我和 Initialized 过去做过。

Uh, yes, yes. I mean, it's very strange stuff, right? Like the Scale AI exit is, you know, that, but I meant, you know, just participating in a secondary transaction. Uh, I and Initialized have done it in the past.

嗯。

Yeah.

但我不——我的意思是，在 YC，我们只是尽量，你知道，我们只是尽量全程陪伴他们。

But I don't—I mean, at YC we just try to, you know, we just try to be there with them the whole time.

是的，没错。

Yeah, that's right.

是啊。你对加州提出的财富税直言不讳。你想谈谈为什么你态度这么坚决吗？

Yeah. You've been outspoken on the suggested California wealth tax. You want to talk about why you're so strongly minded?

是的，我的意思是，他们提出的方案难点在于，如果人们没注意的话，11 月的选票上会有一项，虽然还没获得资格，但他们正在争取，5% 的一次性税，我们都知道这种税从来不是一次性的，但实际上比这更深刻。这实际上是一种未实现收益税。所以，如果你在这个房间里，比如说你的公司，你很幸运地获得了 50 亿美元的纸面估值，而你在纸面上值 10 亿美元。那么，你当时就必须为你的持股支付 5%。如果你的公司归零，实际上，你就白白损失了那笔钱。

Yeah, I mean, the hard part about what they've come up with, so if people haven't been paying attention, there's going to be, on the ballot in November, it has not qualified yet, but they're going for it, a 5% one-time tax, and we all know these taxes are never one-time, but it's actually more profound than that. It's actually an unrealized gains tax. So, if you're in this room and say your company, like, you're so lucky to get a paper valuation of $5 billion and you're worth a billion dollars on paper. Well, you'd have to pay 5% of your holdings at that time. And if your company went to zero, actually, like, you'd just be out of that.

你就只能干坐着——

You'd just be sitting on a—

没错。

That's right.

基本上就是净营业亏损。

NOL, basically.

而且，问题是，这也是对你已经缴过税的东西征税。所以，你知道，假设你已经远远超过了缴清所有税款的程度，他们就能双重征税。我的意思是，老实说，就像百万富翁税——

And so, the thing is, it's also a tax on things that you already paid taxes on. So, you know, let's say you're well beyond having paid all of your taxes, like, they get to double dip. I mean, honestly, like, the millionaire tax that—

而且当前版本是不是也有这种基于投票权来衡量的事情？

And does the current version also have this thing where it measures based on voting control?

是的，是的，是的。所以，我的意思是，我有朋友不得不离开这个州，比如带着家人搬家，就在去年 1 月 1 日之前。以防它通过。

Yeah, yeah, yeah. So, I mean, I've had friends of mine have to leave the state, like, take their families and move, literally before January 1st of last year. On the off chance that it passes.

是啊，我的意思是，它有追溯效力，对吧？

Yeah, I mean, it backdates, right?

所以，我认为这是，你知道，就我个人而言，我不在乎税收，但我在乎什么是公平的，什么才能真正帮助创新。而且，你知道，加州正处于一个疯狂的时刻，它实际上是一个资源陷阱。比如，有这么多资源，但如果你环顾四周，我花了很多时间研究无家可归者政策，我们花了越来越多的钱，却得到越来越糟的结果，主要是为了作秀。我们不应该那样做。我们在旧金山或洛杉矶做的事情并不富有同情心。然后他们攻击人们说：‘哦，你们这些肥猫、商业人士、科技人士，你们只关心那个。’但事实是：‘不，那完全不对。’你知道，我非常乐意交税。我和其他人一样是社会的一部分。

So, I think this is, you know, personally, I don't care about taxes, but I care about what's fair and what will actually help innovation. And, you know, California is in sort of this insane moment where it's actually a resource trap. Like, there's so many resources, but if you look around, like, I spent a lot of time looking at homelessness policy, like, we spend more and more money to get worse and worse outcomes, mainly for virtue signal. And, like, we shouldn't do that. It's the things that we do in San Francisco or in LA are not compassionate. Like, and then they come at people saying, 'Oh, well, you know, fat cats, business, tech people, like, all they care about is that.' And it's like, 'No, that's not correct at all.' Like, you know, I'm more than happy to pay the tax. Like, I'm a part of a society just like everyone else.

对。

Right.

而且，你知道，我想交税，但我希望它用在有意义的事情上，真正为彼此服务。但现在不是这样，对吧？比如我们花钱买百万美元的公寓，被用作毒品窝点，然后有人死在里面。是的。那并不富有同情心。比如在加州，现在州政府资助无毒品永久性支持性住房是违法的。是的，我知道。这完全疯了。而且你知道，即使是收容所，你也不能进行药物检测。

And, you know, I want to pay the tax, but I want it to go to things that make sense and that actually serve one another. And that's not what's happening, right? Like we are paying for million-dollar condos that are used as drug dens and then people die in them. Yeah. And that's not compassionate. Like it is illegal in California right now for the state to fund permanent supportive housing that is dry. Yes, I know that. Which is totally nuts. And it's you know, even a shelter you can't drug test.

是啊。这是违法的。

Yeah. It's illegal.

所以你知道，我的意思是，我们谈论这个，我觉得 10 年前，在这样一个房间里，我不认为有人会谈论它。

And so you know, I mean us talking about that like I think 10 years ago, like you know, in a room like this, I don't think anyone would talk about it.

但我觉得我们必须谈谈这个，因为好像没别人会谈，对吧？我觉得科技界和商界都愿意交税，但我们需要另一端有一个高效的政府，而唯一能做到的办法就是我们得深入细节。

But I think we need to talk about it because like no one else is going to talk about it, right? Like we're you know, I think tech and business like we're fine to pay the tax, but we need an effective government on the other end and only way we can get that is like we have to go all the way into the weeds.

嗯，你在播客里提到过，但这种情况是因为加州的提案程序造成的。所以立法机构或州长并没有提出这个。它实际上是由一个工会——SEIU 工会——支持并把它放到选票上的。但如果投票通过，他们就必须接受。

Well, and you noted this on a podcast, but that situation is set up because of the proposition process in California. So the legislature or the governor didn't bring this. It's being proposed actually by a supported by a union, the SEIU union that has put it on the ballot. But if it's voted in, they have to accept it.

是的。不管你喜欢与否，这就是我追踪到的：住房不足，教育质量下滑。你知道吗，在旧金山，公立中学的学生几乎还是不可能选修代数。

Yeah. I mean like it or not like that's what I've traced you know, not enough housing, uh a watering down of our education. Did you know that it's still almost impossible to take algebra as a public middle school student in San Francisco.

因为他们取消了……是的。我的意思是，这简直难以理解。

Cuz they killed off the uh Yeah. I mean it's like unfathomable.

我本来不可能……我是公立学校的产物，我绝不可能在高三学微积分，进入斯坦福学计算机工程。

I wouldn't be able to... I'm the public... I'm the product of public schools and there's no way I would have been able to study calculus by senior year, get into Stanford computer for computer engineering.

对。

Right.

并且做我所做的一切。我并不是在富裕环境中长大的。小时候我们有时会面临食物短缺，我需要公立学校，需要数学老师教我如何做今天我能做的事情。我希望每个人……我们希望社会运转，但它对我们不利，我们必须把这些家伙选下去。科技界和商界的人必须获得政治权力，我们不能害怕。我们必须发声，因为我们确实能解决我们州的问题。

And do anything that I did. Like I didn't grow up well-off. Like I, you know, we were food insecure sometimes when I was a kid, you know, I needed my public school and I needed my math teachers to teach me how to do the things I get to do today. And I want every... you know, we want our society to work, and it's not working for us, and we've got to vote these people out. Like, we actually have tech people and people in business have to get political power, and we cannot be afraid. We have to speak up because we actually can solve the problems of our state.

在很多方面这真的很可悲。我感到非常幸运，能在硅谷度过 25 年，成为这个生态系统的一部分。其中有些部分简直奇妙无比。你知道，那种分享的程度，人们愿意指导、伸出援手的程度，你可以失败然后立刻投入其他事情，这太棒了。有人故意要摧毁它，这有点疯狂。从我的角度来看，从宏观层面看。

It's really sad in a lot of ways. I feel so fortunate to have been able to spend 25 years in Silicon Valley and be a part of the ecosystem. And there are pieces of it that are just so wonderfully magical. You know, the amount of sharing that goes on, the amount of willingness people have to mentor and reach out, the way that you can like fail and then just immediately be into something else and like it's just so wonderful. The thought that someone would intentionally target to blow it up is a little nuts. Like, from a high level, from my perspective.

但说实话，这对奥斯汀会非常好。

But it'd be really good for Austin, honestly.

不，这对奥斯汀可能很棒。我觉得在某种程度上已经如此了，但我宁愿认为它们都能成功，而不是一个必须以牺牲其他为代价。

No, it could be great for Austin. I think it already is in a certain bit, but I'd like to think they could all succeed and not that one has to succeed at the expense of the others.

是的。对我们来说，我们希望初创公司成功，而且我们之所以还在旧金山，是因为如果你留在旧金山湾区，独角兽的比率实际上高出 2.5 倍。

Yeah. Yeah, I mean, for us, like, we want startups to succeed and, you know, the reason why we're still in San Francisco is that the rate of unicorns is actually 2.5x higher if you stay in the San Francisco Bay Area.

如果他们杀了下金蛋的鹅，我的意思是，蛋很好。让我们吃蛋吧，和大家分享蛋。只是别射杀那只鹅。是的，我敢肯定鹅肉真的很好吃，但你只能吃一次。

And like, if they kill the golden goose, I mean, the eggs are great. Let's eat the eggs, you know, like, let's share the eggs with everyone. Just don't shoot the goose. Yeah, I mean, that you know, I'm sure the goose tastes really really good, but you only eat it once.

是的。让我们看看观众的问题，如果他们能快速弹出来。看看有没有什么能引起你的兴趣，Gary。好的。当 99% 的人都能进行 vibe coding 时，世界会是什么样子？

Yeah. Let's see what audience questions look like if they can pop those up real quick. See if any of these pop for you, Gary. All right. What would the world look like when 99% of people can vibe code?

我觉得这是个好问题。其中一部分是，我们将为彼此做什么？我认为它可能实际上是在操作系统层面。所以，如果你看《她》这部电影，我觉得会有那么一刻，我们不会打开手机看这个界面。我认为我们可能会有某种即时软件。那将非常不同，我们将塑造它。

I think that's a good question. Like, some of it is like, what are we going to be doing for one another? And I think it might actually be at the operating system level. So, you know, if you watch the movie Her, like I think there's going to be this moment where we're not going to open up our phones and look at this UI. Like, I think that we're probably going to have something that's just-in-time software. It's going to be very different, and we're going to be shaping that.

是的。因为 Y Combinator，早期投资的门槛是如何提高的？

Yeah. How's the bar raised for early stage investing because of Y Combinator?

我的意思是，我们自己也开始在 YC 批次中看到这一点。大约一半的公司都走在最前沿。他们每天编写 1 万、2 万行代码，就像我一样。而且他们实际上比我做得好得多。

I mean, I guess we're starting to see it in the YC batches ourselves. It's like, you know about half of the companies are all the way out on the edge. Like, they are coding 10, 20,000 lines of code a day, just like me. And that's you know, they're way better at it than I am, actually.

对。因为他们得到……

Right. Cuz they get...

这就像一套全新的技能，如果你不参与其中，你就会被抛在后面。

It's like there's a whole new skill set that if you're not a part of, you're left behind.

而且这实际上非常难。我的意思是，我有软件工程师朋友，他们处于否认状态。他们说，这不是真的。我的提词里也有。他们说，Taha，这只是一堆 markdown，这只是标准的 VC 陈词滥调，像拉高出货的骗局。我说，伙计们，最有趣的是，YC 已经很好了。我不知道。这不是拉高出货。我真心希望人们成功，这实际上可能是好消息。

And it's actually really hard. I mean I have software engineer friends who are like in denial. They say like, this isn't real. Like, it's in my mentions, too. They're like, you know, Taha, this is just a bunch of markdown, and this is just you know, the standard VC tripe, like you know, pump and dump scam. It's like, guys, like I mean, the funniest thing is like, you know YC is already good. Like, I don't know. I mean, it's like I This is not a pump and dump. It's like, I actually genuinely want people to be successful, and this is actually probably This is the good news, actually.

关于这件事非常有趣的一点是，我觉得对所有创意人士来说都是如此，你只需要打开并使用工具。因为你知道，你会走在最前沿。你会成为顶尖，如果你是一个优秀的工程师，今天的 10 倍工程师可以成为 2000 倍工程师，遇到拒绝成为 2000 倍工程师的 10 倍工程师是很奇怪的。这似乎不应该发生。但我认为有大量这样的人，我认为这种情况一直在发生。改变是困难的，尤其是当它曾经是你的超能力时。有一本书叫《谁动了我的奶酪？》我觉得它很短很简单，但我认为它是定义性的书。我注意到的另一件事是，那些深入技术的人会开始理解它还能做什么。所以一旦你进入其中，几乎就像有一种指数级或加速的启动，因为两周、四周后，你的思维已经扩展，对可能性和用例的理解也扩展了。所以你实际上在远离所有人。

Like, the very interesting thing about this, and this is I feel like it's true across all of um for creative people all you have to do is like open up and use the tool. And because you know what you know you're going to be all the way out on the edge. Like, you will be a top point, you know, if you're a good engineer, like the 10x engineer today can be that 2000x and like you know it's very strange to run across like the 10x engineer who refuses to become a 2000x engineer. Like it seems like that shouldn't happen. But I think like there's a mass like I think that's happening like pretty consistently. And change is hard. Especially if it was your superpower. There's that book Who Moved My Cheese? I feel like that you know it's a very short very simple book but I think it's the defining book. The other things I've noticed is people that roll around in the technology come to understand what else it might be able to do. And so it's almost like there's this exponential or accelerated launch once you get into it because two weeks in four weeks in your mind has expanded in terms of what you think it's possible of and the use cases that you're using it. And so you're literally running away from everyone.

有一个问题：YC 会在其他城市启动吗？

There's a question would YC launch in other cities?

是的，我们公开地，尤其是因为加州正在发生的事情，我们不喜欢把所有的鸡蛋放在一个篮子里，尤其是当外面有一些杀鹅者时。所以我们有剑桥，我认为马萨诸塞州的剑桥可能是我们的……我的意思是，也许这就是例子，你看我们并不反对税收，马萨诸塞州实际上有百万富翁税，我认为是 4% 或 5%，对每年超过一百万美元的工资部分。

Yeah, I mean we're openly especially because of what's happening in California like we don't like to have all of our eggs in one basket and especially when you have some goose killers out there. So we've got a you know Cambridge I think Cambridge Massachusetts is probably our you know I mean maybe this is the example of like look like we're not against taxes like you know Massachusetts actually has like a millionaire tax like I think it's 5% or 4% on every dollar above a million dollars a year in salary.

其实那是一种正常的税收，没那么夸张。你知道，如果你有超级投票权股票，突然要通过税收损失净资产的 170%，那才叫疯狂。有很多方法可以做得很好。而且剑桥有很好的学校系统，非常运转良好。你知道，我觉得加州在功能失调方面走得太远了。不过我们希望扭转这个局面。

It's like actually that's a normal tax, like it's not this crazy thing. You know, if you have super voting shares, suddenly you're going to lose like 170% of your net worth through taxes. There are ways to do this that work really well. And I mean, Cambridge has a great school system, like it's very functional. Like, you know, I think California is way out on the edge of what is not functioning. And we hope to turn it back though.

在奥斯汀办一期训练营，你需要什么条件？

What would it take for you to run a batch here in Austin?

天哪，我很乐意。所以，部分原因是，我需要在这里招一位合伙人。或者，你知道，我们雇了第一位常驻剑桥的合伙人，安基特·古普塔，他非常出色。

Oh my god. I mean, I'd love to. So, I mean, some of it is I need to hire a partner who's here. Or like, you know, we hired our first partner who's based in Cambridge full-time, Ankit Gupta, and he's incredible.

好的，希望有一天能看到。

All right. Well, hopefully we'll see that one day.

有个问题：我如何避免我的初创公司只是另一个云技能？

There's a question: How can I avoid my startup being just another cloud skill?

我觉得关键在于要有客户。而且，我是说，再说一次，这个星球上 99.9% 的人。你知道，每个人都知道，也许我上网太多了，但我总是想到那个梗，就是他们不知道，就像一屋子人都在跳舞。然后你知道，角落里有个家伙，他说，他们不知道一切即将改变。这就是我现在对这一切的感受。

I mean, the great thing about it is like have customers. And like, I mean, again, there's like 99.9% of the people on this planet. Like, you know, everyone knows, maybe I'm online too much, but like I always think of that meme where it's like they don't know, and it's like a room full of people all dancing. And you know, there's one guy in the corner and he's like, they don't know that everything's about to change. And that's like how I feel about all of this right now.

而且你们非常强调快速获取客户。即使在 13 周的项目里也是这样。

And you guys are really big on getting customers fast. Like, even in the 13-week program.

是的，我觉得答案就像我之前说的，是同理心。奇怪的是，如果你能设身处地地站在别人的角度，完全创造出那个东西，并理解他们想要什么。

Yeah, I mean, so the answer I think is like going back to what I said earlier, which is empathy. Weirdly, it's like just literally if you can put your feet into the shoes of another person and totally create that thing and understand what they want.

好的，这是最后一个问题。

Okay, this is the last question.

最后一个问题。

Last question.

除了运营 Y Combinator，你还创办过公司。你见过所有这些公司。如果明天加里·陈被迫成为一名创始人，你会对什么想法最着迷？

You've started companies in addition to running Y Combinator. You see all these companies. If Garry Tan were forced to be a founder tomorrow, what ideas would you be most enthralled by?

其实我最近就做过类似的事。我在创办，我是说，我发布了一些开源项目。除此之外，我还有 Garry's List，网址是 garryslist.org，现在看起来像个博客，但实际上有点像 WordPress 和 Meetup 生了个 AI 宝宝。所以，我觉得，它的好处是我们会利用所有这些。其中一个有趣的事情是，这有点像我觉得现在可以做的功能形态。我们第一次发布时有 400 人。他们都得填写短文。但然后，在结束时，我们没有直接散场去酒吧。我让人们，我做了定制介绍，介绍给房间里他们还不认识但需要认识的人。我给他们所有人的简介，然后你问这个人这个问题。然后我们基本上让每个人都带着至少三四个新朋友离开，他们可能可以一起吃饭，然后心灵相通。所以，你知道，我觉得这就是一个具体的例子。这可能是百万分之一。像现代匹配技能和 Meetup 类的东西。我觉得这是世界现在需要的百万种不同事情之一。就像我们如何构建软件，我的意思是，人们想要什么？他们想要连接，想要意义，想要社区。我想，我不知道。这就是我想要的。这就是 YC。你知道，在 YC 建立 Bookface，举办这些活动，这就是我们整天做的。这就是我们整天做的。而且我们需要的不只是一点点。我觉得我们的社会需要 2000 倍更多这样的东西。如果我们的社会有这些……

I mean, I sort of did that recently. Like, I'm starting, I mean, I released a bunch of open source. And then aside from that, I have Garry's List, which is garryslist.org, which is right now it looks like a blog, but it's actually kind of like if WordPress and Meetup had an AI baby. And so, I think that I mean, the great thing about it is like we're going to use all of that. Like one of the fun things that, and this is sort of the shape of like the kind of features I feel like you can do now. We had our first launch. It was 400 people. They all had to fill out the essay. But then, at the end, we didn't just break and go to the bar. I had people, I did custom introductions to the other people in the room that they didn't know yet, but they needed to. And I had like here's all of their bios and here you ask this other person this question. And then we basically made, yeah, everyone walked away with at least three or four new people who they could probably go have dinner with them and just mind meld with them. And so, you know, I think that that's like one concrete example of what I mean. This is one of like probably a million, you know. Like modern matching skills and Meetup type things. I think it's like that's one of a million different things that the world sort of needs right now. It's like how do we build the software that, I mean, what do people want? They want connection, they want meaning, they want community. I was like, I don't know. Like that's what I want. That's what YC is. That, you know, building Bookface at YC, like having these events is like that's what we do all day. That is what we do all day. And it's like we need not a little bit of it. Like I need like 2,000 X more of that in our society. And if we had that in our society...

好吧，我在这里没花时间，因为我知道我们想专注于加里谈论我的书，但书名叫《追逐梦想》，讲的是如何找到并热爱你真正喜欢的工作。我之前和加里谈过一件事，他肯定处于那种状态，运营 Y Combinator。而且 Y Combinator 交到了可靠的人手里，因为是他运营的。所以，非常感谢。再见。比尔的书。

Well, I didn't take any time up here because I knew we wanted to focus on Garry talking about my book, but it's called Running Down a Dream and it's about how to find and thrive in a job that you actually love. And one thing that I talked to Garry about before is he is in that place for sure, running Y Combinator. And Y Combinator is in good hands because he's the one running it. So, thank you very much. Bye, bye. Bill's book.