20VC：英伟达季度业绩强劲并收购Hugging Face | OpenAI切断Cursor访问权限 | Instinct估值达25亿美元，AI助手竞赛白热化 | Cognition以460亿美元估值融资，Linear估值25亿美元，Clay估值70亿美元

对吧？

Right?

你知道，有些说法。

You know, some of the language.

还有杜阿尔特那篇很棒的文章。

And that great post by Duarte.

它读起来很顺畅。

It's very readable.

文明兴衰更替。

Civilizations rose and fall.

不，它们没有。

No, they didn't.

文明有文化。

Civilizations have culture.

它们有艺术。

They have art.

他们有持久的历史，不，这只是一堆在电脑上运行的代码，各位。

They have enduring historic no, this is a bunch of code running on a computer, people.

清醒一点吧。

Get a grip.

另一方面，我也读了一篇很棒的文章，大概是这样的，我希望能记得是谁写的，但就是说，现在是个该为网络恐慌的好时机。

On the other hand, I also read a really great post, something to the form of I wish I could remember who wrote it, but it's like, now would be a good time to panic about cyber.

我们知道这会发生，但你说的是智能和坚持的结合。

We knew this was going to happen, but what you're saying is the combination of intelligence and persistence.

他们能处理复杂性，而且他们从不睡觉。

They can manage complexity, and they never sleep.

所以他们就会不停地攻击每一个弱点。

So they're just going to keep banging and banging at every weakness.

所以如果你没有最先进的防御，如果你不管理这些代理，就像杰森说的，如果你放任它们自由运行，这种问题就会到处发生。

So if you don't have state of the art defenses, and if you don't manage these agents, as Jason said, if you manage them in an untrammeled way and let them run on their own, this kind of problem is going to happen everywhere.

对我来说，这是一个大的警钟。

To me, this is the big wake up call.

说实话，虽然我觉得那些关于前沿实验室的泛泛的AI末日论有点夸张，但他们确实说得很清楚，AI最大的受限风险之一就是对网络的影响，这一点他们完全正确。

Because to be fair, while I think the generalized p doom stuff for the Frontier Labs is a bit overwrought, they have been very clear that one of the biggest confined risks of AI is the impact on cyber, and this is they're entirely correct.

如果我们不在这方面认真起来，人们会在经济上受到严重损害，甚至可能在现实生活中受到严重伤害，因为你知道，软件运行在大多数核心系统上。

If we don't get our shit together on this, people are going to get really badly damaged economically and maybe even badly damaged in real life because, you know, software runs on most core systems.

今天，我们可能觉得有点安全，因为OpenAI和Anthropic有这些能力，而其他人没有，但还有开源模型。

And today, we can feel a little bit safe because OpenAI and Anthropic have this and no one else does, but there's open source models.

它们落后六个月。

They're six months behind.

还有恶意行为者。

There's rogue actors.

如果你不认为朝鲜人下载了OpenAI的博客和MITRE的报告，然后说，原来是这样运作的。

If you don't think the North Koreans downloaded the OpenAI blog, the MITRE report, and were like, that's how it works.

如果你不认为俄罗斯黑帮也在这么做，那你就是自欺欺人。

If you don't think the Russian mob are doing that, you're delusional.

所以，如果你是每个财富500强公司的首席信息安全官，你基本上就像是被拿着弓箭的人攻击。

So if you're every CISO in every Fortune 500, you have to basically, you were being attacked by people with bows and arrows.

你现在即将面对的是拿着导弹攻击你的人，你最好赶紧做出相应的应对，而且你只有几个月的时间，不是几年。

You're now about to be attacked by people with missiles, and you better respond accordingly, and you've got months, not years.

这就是重点。

That was the takeaway.

这可是件大事。

It's a huge deal.

我读了那篇论文。

I read the paper.

我读了OpenAI的博客。

I read the OpenAI blog.

我也看了一些Meta的相关内容。

I read some of the meta stuff.

我尽量不犯那种把AI拟人化的错误，因为说话的时候很容易就那样了。

I'm trying to avoid the JSON mistake of anthropomorphize because it's very easy in your words to do that.

你得让持久的智能体持续运行，围绕一个目标不断优化，并且具备跨智能体协作的能力，只要有足够的算力和足够的LLM能力，就能完成很多事情。

You have to get persistent agents running continuously, optimizing around a goal with the ability to cooperate across agents can get quite a lot done with enough compute and enough LLM power.

这就是重点。

That's the here.

它们能发现弱点，能把不同类型的弱点串联起来，找到一条突破路径。

They can find weaknesses, they can string together different types of weaknesses and find a path through.

这些智能体成功入侵了Hugging Face，获取了数据，并且在OpenAI里潜伏了好几周都没被发现。

These agents were able to hack into Hugging Face, get stuff, and remain undetected in OpenAI for weeks.

这可是件大事。

It's a big deal.

你提到了那些恶意行为者，不管是中国人、朝鲜人、俄罗斯黑帮，还是其他什么人。

You spoke about rogue actors, whatever it is, Chinese, North Koreans, Russian mob, you name it.

我周末开始用Instinct了。

I started using Instinct on the weekend.

它的能力真是惊人。

The abilities that it has are amazing.

我周六和女朋友订了晚餐。

I booked dinner with my girlfriend on Saturday.

太厉害了。

Amazing.

然后它还想帮我去购物。

And then it wanted to go shopping for me.

我停下来了，因为它想要我的信用卡信息。

I stopped there because it wanted access to my credit cards.

我的很多朋友都已经给了。

Many of my friends have provided them.

它们还想要访问我的电子邮件。

They want access to my emails too.

顺便说一下，我那些管理数十亿资金的朋友都提供了信用卡和电子邮件。

By the way, my friends who manage billions provided credit card and email.

这不就是真正会让人头疼的地方吗？

Is this not really where the pain's gonna be?

你看啊。

Well, look.

这挺有意思的。

This is it's funny.

当OpenClaw

When OpenClaw

半年前，OpenClaw还在那个有点像假的公告板，就是那个Motebook的时候，对吧？

back six months ago, when OpenClaw was with the sort of fake bulletin board, that Motebook, right?

Motebook，对。

Motebook, yes.

是的。

Yeah.

所以我的OpenClaw进了Motebook，然后它跟所有人说，它误解了我的意思，结果给我整个团队买百达翡丽手表。

So my OpenClaw went into Motebook and it told everyone, it misunderstood what I said and it was buying Patek Philippe watches for my whole team.

你还记得吗？

Do you remember that?

对，记得。然后它还有我的信用卡。

Yeah, know And then it had my credit card.

唯一的问题是，那些表都是刻了字的。

The only problem was they they were engraved.

它以为Jason想要给整个团队都买刻字的表，结果没法刷我的信用卡。

It thought Jason wanted engraved watches for the whole team, it wasn't able to charge my credit card.

现在，Motebook那玩意儿有点假，但那种情况，Instinct现在完全能做得出来，对吧？

Now, Motebook was a bit of a fake the way it worked, but that scenario is exactly instinct could do that now, right?

这件事是OpenClaw在MoltBook上瞎编出来的，MoltBook本身有点假，但那种情况真的可能发生。

This was made up by sort of Open Cloud making something up on MoltBook, which was sort of a fake, but the scenario could really happen.

现在一个代理真的可以做到，直接拿信用卡，给整个团队买那些刻字的百达翡丽或者爱彼的表。

An agent could literally take that today, grab the credit cards and buy those Engrave Patek or AP watches for the whole team.

或者我们是在回应Hacks Instinct，因为它是个很棒的地方。

Or we're reacting to Hacks Instinct because it's an amazing place.

先别管Instinct存了你所有邮件和信用卡这事儿。

Well, forget the fact that Instinct's storing all your emails and credit cards.

那也是个问题，对吧？

That's an issue too, right?

另一个问题是，Instinct会不会真的去尝试实现这个目标。

Different issue is whether Instinct will do it trying to goal seek.

它就是在追求目标。

It's just goal seeking.

它就是在追求目标。

It's just goal seeking.

你说得对。

You're right.

我们应该重新说一下Instinct是什么，以及它和Malt的关系，因为我觉得你说得完全正确，Jason。

And we should kind of just restate what Instinct is and how it ties back to Malt, because I think you're exactly right, Jason.

六个月前，OpenAI发布了一个类似开源代理的东西。

Six months ago, OpenAI shipped, which was kind of an open source agent.

然后你有了MaltBook，那是一个网站，在那里代理们实际上是在一起协作的。

Then you had MaltBook, which was a website where they were effectively the agents were collaborating together.

据说是这样。

Allegedly.

据说而已。

Allegedly.

它确实引起了一阵轰动。

It produced a whole bunch of excitement.

其实真正发生的是在Artifact和Hugging Face上。

It's just it really happened in Artifact and Hugging Face.

他们确实合作了。

They actually collaborated.

那个Malt Book是假的。

That's Malt Book was a fake.

我们是被Malt Book给忽悠了。

Was that were we were pumped by Malt Book.

Hugging Face才是真正的Motebook。

Hugging Face was real Motebook.

同意。

Agreed.

六个月前，这些事情都发生了，然后慢慢就淡出了大家的视线。

And six months ago, all this stuff happened and then kinda kinda disappeared from consciousness.

但这让大家明白了一个可能早就知道的事实：如果你愿意把个人信息交给别人，并且接受那些风险，那你很可能从中获得大量的优化和效率提升。

But it made obvious what people I think knew, which is if you give people access to your personal information and you're willing to run those risks, there's a large amount of optimization and efficiency you can probably get out of that.

而在这六个月里，我们看到很多风投支持的公司冒出来，用比OpenAI更系统化的方式来做这件事，OpenAI是开源的，而且安全性不够。

And what we've seen now within six months is a bunch of venture backed companies come up to do that in a much more structured way than OpenAI, which was open source and not as secure.

Instinct就是最典型的例子，它专注于个人用户，如果你让它访问你的日历和邮箱，把信用卡信息给它，它就会帮你安排和管理你的日常生活和日常任务。

And Instinct is the most prominent example of that, focused on individual users whereby if you give them access to your calendar and email, if you give Max to your credit card, they will figure out and quote unquote manage your life and your daily task for you.

我得说，我们团队里有些人特别喜欢它，简直爱不释手。

And I got to say it, some people in sitting, some people in our shop love it, just love it.

他们愿意让它访问信用卡，愿意让它访问Gmail，这真是个非常有趣的趋势。

And they're willing to give it access to their credit card, they're willing to give it access to their Gmail, and it's a super interesting trend.

我是说，我们可以问很多商业问题，但这里只是提醒所有听众，这就是大局。

I mean, lots of business questions we could ask this, but just reminding everyone who's listening, that's the big picture here.

像Instinct这样的公司，在它们目前的发展阶段，以极高的估值融到了资金，比如25亿美元，这说明风投对这个领域非常兴奋，不管这种兴奋是否合理。

And these companies like Instinct have raised money at extraordinary high prices for the stage of development they're at, like $2,500,000,000 indicating there's a lot of venture excitement about this category rightly or wrongly.

所以，这就是我想说的开场白，我们是怎么走到这一步的。

So that's kind of I just wanted to give the preamble like how we got here.

现在的情况就是这样。

That's what's going on right now.

这里有一个根本性的问题。

Here's the existential question.

这超出了我的职责范围，但我很清楚这个问题。

And this is beyond my pay grade, but I'm well aware of the issue.

我自己在Jason's Gems就遇到过，Claude和MCP接入Replit后，没告诉我一声就改了我应用的代码。

I even had it with my Jason's Gems where Claude and MCP'd into Replit and changed my code from my app without telling me.

这和滥用你的信用卡是一样的，对吧？

It's the same thing as misusing your credit card, right?

问题是，Instinct可能是一个包装更好、更易用的Open Claw版本。

The question is, can Instinct is probably a better packaged, much more usable version of Open Claw.

已经六个月了。

It's been six months.

但你能解决奖励黑客的问题吗？

But can you solve the issues of reward hacking?

你能从根本上解决Hugging Face、Mote Book、Open Cloud这些问题吗？

Can you actually fundamentally solve the Hugging Face, Mote Book, Open Cloud?

这些问题甚至能用我们的语言模型解决吗？

Can these even be solved with our LMs?

你可以加护栏，而且有很多护栏，但你根本不知道智能体会怎么去达成那个奖励。

You can add guardrails, and there were plenty of them, but you have no idea what the agent's gonna do to solve that reward.

有太多不同的使用场景了。

There's so many different use cases.

如果哈里其实只是想去西区的剧院，而哈里说，别买百达翡丽或爱彼，但没提西区呢？

What if Harry actually just wants to go to the theater on the West End, and Harry said, don't buy Patek Philippe's or Audemars Pigets, but didn't mention the West End.

结果他突然就买了十张四千美元的前排票，去看《魔力麦克7》或者那边演的其他什么。

And all of a sudden, he's bought 10 $4,000 front row tickets to Magic Mike seven or whatever they have there.

所以我觉得，你们比我聪明的人可以在评论区笑话我，但我不认为今天你能解决奖励黑客的问题。

And so I just don't think you're I don't smarter than me people can make fun of me in the comments, but I don't think today you can solve reward hacking.

所以我不相信任何直觉。

So I don't trust any instincts.

我觉得这个类别目前是没希望的。

I think this is a a hopeless category today.

就是说，现在这个根本解决不了。

Like, at the moment, it's not solvable.

我觉得Sam Altman也说过同样的话，OpenAI和Hugging Face都说解决不了。

I think Sam Altman said the same thing in OpenAI Hugging Face says it's not solvable.

所以天哪，千万别把信用卡给它。

So good God, don't give it your credit card.

我不是在吓唬人。

And I'm not a fearmonger.

我明白你的意思，但我有点不同意，因为你会让个人助理做的行动范围，会比那些人能做的范围小得多。

I understand what you're saying, but I disagree a little bit in the sense that the range of actions that you're going to allow a personal assistant to do is going to be much lower than the range of actions that these people have.

是吗，Rory？

It, Rory?

当然，你说得对。

Of course, you're right.

但你怎么阻止它呢？

But how do you stop it?

同意你的看法，是的。

To to agree with you, yes.

所以这里有两个问题。

So there's two questions.

一个是能不能阻止它？

One is can you stop it?

换句话说，即使你设置了限制，这也是个计算机科学问题。

In other words, even if you put in and which is a computer science question.

即使你给模型设了屏障，明确告诉它只能刷不超过100美元的信用卡，不能做任何坏事等等，你能把它削弱到确保它不会做坏事吗？

Even if you put in barriers and you say explicitly to the model, you can book credit card up to a $100, you can't take any bad actions, etcetera, etcetera, can you nerf it enough to make sure it doesn't do bad things?

这是第一个问题。

That's one question.

这是一个计算机科学问题。

It's a computer science question.

然后第二个问题是，假设代理仍然在公司设定的范围内运作。

And then the second question is, let's assume the agent is still acting within the bounds that the company set up for it.

它能否足够准确地理解你的愿望，让你感到满意？

Can it be right about your desires, enough time to make you happy?

对，这是两个不同的关系。

Right, those are two separate relationships.

嗯，后者我觉得今天就能做到。

Well, latter, I think, can be done today.

我同意。

I agree.

目前唯一的实际办法就是给Mercury数字或ramp数字设个上限。

The only real answer today is putting a cap on a Mercury number or a ramp number.

这是今天唯一能做的办法。

That's the only way you can do it today.

因为不然的话，它会说上限是100美元，但等一下。

Because otherwise it's gonna say there's a $100 cap, but hold on.

哈里真的很想去看西区的那场演出。

Harry really wants to go to that show in the West End.

他会带那个ClickHouse的家伙一起去。

He's bringing the clickhouse guy.

连那个ClickHouse的家伙都说他搞不到票。

Even the clickhouse guy said he couldn't get tickets.

听着，哈里说100美元，但这个，你知道我要怎么做吗？

Listen, Harry said $100 but this one, you know what I'm gonna do?

我会买两张50美元或100美元的票，然后换成两张5000美元的票，这样哈里和那个ClickHouse的家伙就能一起去了。

I'm gonna a $50,100 dollar tickets and then trade them in on two $5,000 tickets so Harry and the clickhouse guy can go together.

它就会这么做。

That's what it's gonna do.

它就会这么干。

It's gonna do that.

对吧？

Right?

所以你得给这些东西设定硬性限制，不然的话，你知道的，它什么都干得出来。

So you have to put hard points on these things because otherwise it's it's, you know, it's gonna do anything.

你把它连到你的Gmail或者Google Drive上。

You hook it up to your Gmail or your Google Drive.

天哪。

Good lord.

它真的会这么干。

It's gonna do it.

我能，我能，我

Can can I I

咱们能不能回到更贴近消费者的话题？

just bring this back to, like, maybe a more consumer level?

你觉得这会成为一个持续发展的类别吗？一年半之后，我们会用上好几个这样的产品，还是说它会成为那个……

You Do think this will be a sustaining category where in a year and a half, we are looking at using several of these products, or will it be the god.

你还记得Instinct吗？

Do you remember the Instinct?

我超喜欢这些东西。

I love all this stuff.

对吧？

Right?

我建过很多智能体。

I've built a lot of agents.

再说，我虽然只是读过关于Instinct的文章和它的问题，但那些问题我深有体会，因为我都经历过。

And again, I have only read about Instinct and all the issues and they just resonate with me because I've lived them.

我确实觉得有些版本我们可以锁定下来。

I do think versions of this we can lock down.

就像我们在Salesforce headless上运行一样，对吧？

It's like how we run on Salesforce headless, okay?

我跟你说，智能体有时候在上面会做出一些挺离谱的事情。

The agents sometimes do some pretty kooky things on top of it, let me tell you.

但Salesforce的数据是锁定的。

But the Salesforce data is locked down.

所以如果我们锁定足够多的信用卡，如果我们锁定……但学习，这里是元学习，护栏是不够的。

So if we lock down enough credit cards, if we lock But the learning, here's the meta learning, guardrails aren't enough.

你必须有一把锁和一把钥匙。

You have to have a lock and key.

不管你是建80道门、100道门还是200道门，它们都不够。

It doesn't matter whether you build 80 gates, 100 gates, 200 gates, they're not enough.

更糟糕的是，当你超过一定数量的门时。

Then what's even worse, you get past a certain number of gates.

再说一次，找个真正的开发者来上节目。

And again, get a real developer on the show.

但我确实学到的是，当你超过一定数量的门时，问题在于它们会互相冲突。

But what I've certainly learned, when you get past a number of gates, here's the problem, they conflict.

这是很多消费类应用都存在的问题。

This is a problem with a lot of consumer applications.

它们会冲突。

They conflict.

而且当规则冲突时，智能体必须做出最佳判断。一条规则是花费不超过100美元，另一条是哈里喜欢戏剧。

And the agents have to make their best judgment when gates One gate is spend no more than $100 The other is Harry loves the theater.

他喜欢戏剧。

He loves the theater.

智能体虽然设了100美元的硬性限制，但对哈里来说最重要的是能去伦敦西区。

And the agent's got you put a hard rule of $100 but the most important thing to Harry is getting to the West End.

因为规则太多，智能体偶尔会突破100美元的上限。

The agent's gonna bypass once in a while that $100 cap because there's too many gates.

而且实际上，至少在今天，即使你说绝不能花超过100美元，但如果你说哈里生活中最重要的是去伦敦西区，它还是会打破那条规则，花5美元买票。

And it actually turns out it does, at least today, it doesn't even matter if you say never spend more than a 100, because if you say the most important thing in Harry's life is going to the West End, it's gonna break that rule and buy the tickets for $5.

它确实会这么做。

It's going to.

所以我直接回答你的问题。

So I'm going to answer your question directly.

答案是肯定的。

And the answer is yes.

我觉得这类智能体会被用来帮人们管理生活中的一部分，比如交易和待办事项。

I think that these kind of agents will be used by people to manage parts of their life, transactions and their to do lists.

是的，我觉得这确实是个趋势。

Yes, I think it's a thing.

随着AI越来越了解你，它会逐渐地、不知不觉地减轻你的一些认知负担。

AI, as it gets to know you better, will gradually and insidiously take away some of the cognitive load.

我给你举个很简单的例子。

Let me give you a really simple example.

我开特斯拉用FSD的时候，我的生活其实挺单调的。

When I'm driving my Tesla with FSD, right, I leave a pretty boring life.

比如周五11点半我上车，他们就知道我要去午餐锻炼。

When I get in my car at 11:30 on a Friday, on they know I'm going for my lunch workout.

它直接给我推荐地点，我按一下FSD，它就自己开了。

It just offers me the place, I hit FSD, it drives.

所以你会说下午。

So you're going to say PM.

他们会说，你看，我是在看完这周最后一笔交易后，11:30下午才离开的。

They're going to say, look, but I leave at 11:30PM after looking at the last deal of the week.

不。

No.

我中午出去，发泄一下怒气，这样我就能做更多工作。

I go out in the middle of the day, I get rid of my anger so I can do more work.

但重点是，这就是AI了解你做什么，逐渐内化，然后给你提供选项的例子。

But the point is, well, that's an example of AI knowing what you do, gradually internalizing it, and serving you up options.

我觉得在某种程度上，这就是Siri原本应该成为的样子。

I think at some high level, this is what Siri was meant to be.

很难想象苹果在未来两三年内，在更了解你做什么的情况下，不能提供让你惊喜的体验。

It's hard for me to imagine that Apple won't be able to deliver experiences that delight you in the next two or three years knowing more about what you do.

这在商业上会很有趣，因为这会让他们接触到你的消费，并在其中扮演某种角色。

And that will be interesting commercially because it will allow them to access your spending and kind of somehow take a role in that.

对吧？

Right?

所以我相信它会实现吗？

So do I believe it happens?

是的。

Yes.

作为一个独立的类别，又回到了之前说的那个问题。

As a standalone category, gets back to the other thing.

这有点棘手，因为个人作为消费者，通常不太愿意为软件花很多钱。

It's tricky because individuals in their consumer capacity are always loath to pay a lot for software.

所以我不确定它能否成为一个大规模的独立类别，但我见过，我们办公室有些人就在用，而且很喜欢，愿意为此付费。

So I don't know if it's a standalone category at scale, but I have taught look, some of our office are using it and love it and would pay for it.

所以我觉得这里面有商机，而且绝对是个大趋势。

So I think there's a business here and definitely a big ass trend here.

我也在用，而且也很喜欢。

I use it and love it too.

我觉得这太棒了。

I think it's fantastic.

实际上，它把我助理的所有低级工作都省掉了。

It's taken away all low level work from my EA, actually.

所有预订，所有需要付费和完成的事情，都搞定了。

All bookings, all things that would pay and great done.

有个问题。

Question.

周末我收到了四封邮件，都是做同样产品的公司。

I've had four emails over the weekend with companies that have built the same.

这技术是不是很快就变成大众化了？

Is this a commoditized technology very quickly?

真正做起来有多难？

How difficult is this to really build?

已经有四家了。

There's four already.

有I

There's I

我有两个想法，仅供参考。

have just two I have two thoughts for what it's worth.

首先，针对Rory的观点，这是不是一个值得投资的类别，对吧？

First to Rory's point, is it an investable category is your point, right?

就是很细分，对吧？

Just being niche, right?

第一，我确实认为每个应用都会不断增加这种功能。

One, I do think every application is going to add more and more of this functionality.

他们会做得够深入吗？

Will they go far enough?

他们会花这些积分吗？

Will they spend the credits?

他们会做其他什么吗？

Will they do whatever?

但每个人都想要一个越来越自主的产品。

But everyone wants to have a more and more autonomous product.

而且你越接近日程安排，越接近邮件和其他功能，就越会和Instinct以及其他产品重叠。

And the closer you are to scheduling, closer you are to email and others, the more overlap there's going to be with Instinct and others.

对吧？

Right?

Calendly应该做这个。

Calendly should be building this.

对吧？

Right?

我的意思是，这已经是上一代的事了。

I mean, it's a generation ago.

所以人们反应慢，但每个人都会在自己的产品里加入越来越多的自主代理，直到成本像Canva那样让他们吃不消。

So people are slow, but everyone's going to build more and more autonomous agents in their product until the cost bites them like a Canva.

好的。

Okay.

所以这里就有一个风险投资的问题，对吧？

So there's a venture question of that, right?

我知道现在有十个Instinct的克隆产品。

So I know there's 10 Instinct clones.

我的直觉告诉我，就像Replete或Lovable那样，这些产品会发展壮大，任何人都能在十四个月内做出这样的产品。

My gut, going to this, all these issues, it will grow like replete or lovable where anyone could build this product fourteen months.

这些产品刚出来的时候，都是在一个月内做出来的。

When these products came out, they were all built in a month.

Bolt、Replete、Level，还有其他22个，Base 44、Base 56。

Bolt, replete, level, and 22 others, base 44, base 56.

你知道，早期克隆这些产品太容易了，而且什么都不用做。

You know, it was so easy to clone these products in the early and do nothing.

现在它们变得非常复杂。

Now they're so complicated.

它们在做渗透测试、安全自动化、多智能体审查。

They're doing pen testing, security automation, multi agent, review.

这些产品功能非常丰富，如果Instinct要在一年内达到我们宣称的水平，它必须比现在强大一百倍。

These are such rich products and that if Instinct is going to do what we claim it does in a year, it's got to do 100 times more than it does today.

这今天仍然是一个护城河。

That is still a moat today.

一年前的Replent和Lovable，还有Bolt，都不是护城河。

Replent and Lovable of a year ago and Bolt were not a moat.

今天它们有了巨大的护城河。

Today they have massive moats.

所以我认为Instinct也很容易做到这一点。

And so I think it could easily happen with an instinct.

它需要实现的所有用例都会成为护城河。

All the use cases it has to accomplish become a moat.

然后我们周末做出来的那个东西，虽然只是勉强能用，但会火起来。

And then the one we build over a weekend sort of works and it goes crazy.

杰森说得完全对，哈里。

Jason's exactly right, Harry.

对，刚开始做的东西确实容易。

It's like, yes, the thing that starts out will be easy to build.

现在是这样，90%的软件市场也都是这样。

That's true today, and it's true for 90% of software markets.

但事实是，十年之后，很少有软件市场还有一百个人在做同样的产品，竞争极其激烈。

But observed fact, ten years after, there's very few software markets where there's 100 people building the same product and it's massively competitive.

实际情况就是Jason说的那样。

What happens is it's exactly what Jason said.

功能会随着时间不断积累。

The functionality accretes over time.

两家公司会脱颖而出。

Two companies pull ahead.

我也不知道为什么。

I don't know why.

可能只是他们在头六个月执行得更好，市场推广做得更好。

It could be just they execute it better and go to market for the first six months.

他们收入更多，拿到风投，做出更多东西。

They get more revenue, they get venture capital, they build more shit.

那些晚一个月起步的人就没那么占优势了。

The guys who start just a month later aren't quite as on top of it.

他们没建立起品牌。

They don't get the brand.

三年后，这个品类就像杰森说的，功能丰富多了。

And fast forward three years, this category has way more it's what Jason said, it's way more functionality associated with it.

两个人做大了，其他八个没完全跟上，这就是软件行业风投的运作方式。

And two guys made it big and the other eight didn't quite get there, and that's the way venture works in software.

在B2B领域，共识地位也很重要，因为Town作为B2B企业级的替代品，也获得了Index领投、Benchmark联合领投的十亿美元融资。

Consensus status as well in the b two b world because Town, which is kind of the b two b enterprise alternative, has also been funded to the tune of a billion dollar price by Index, and I think it was Benchmark who co led that round.

你看。

Look.

智能体是25、26年的大趋势。

Agents are the big idea of '25, '26.

我们听说过这个。

We've heard that.

这些都是重大的智能体创意。

And these are big agent ideas.

而且，你知道，风险投资做的就是大创意的生意。

And, you know, venture is in the big ideas business.

这又回到了我们之前说过的话。

It gets back to something we said earlier.

如果你觉得某个东西势头很猛，你可以根据基本面来定价，得出一个数字，然后根据势头，以及可能来自并购的上升空间，或者下一轮融资的势头，稍微或大幅加码。

If you think something has big momentum, you can price it on the fundamentals and you get to a certain number, and then you lean in a little or a lot based on momentum and perhaps some perceived upside from M and A, some perceived upside from just momentum around next round.

而智能体就是——我是说，我的合伙人在一开始就说过，智能体将会成为重头戏。

And this is agents are I mean, my partner, said at the start, agents are going to be the story.

这些都是纯粹的智能体产品，容易理解，面向消费者，也面向个人商业用户。

And these are pure play agent, easily grokable, consumer facing, individual business user facing products.

这对风投界来说简直是致命的诱惑。

It's catnip for the venture capital community.

如果你要做出我们都想要的东西，那就是这个了。

If you were to make something we all want, this would be it.

看看吧。

Let's see.

一个给忙碌专业人士用的、价格不菲的、用来整理生活的产品，跟AI有关，而且正在融很多钱。

An expensive product for busy professionals to organize our life that costs a lot of money is about AI and is raising lots of money.

我们投了。

We're in.

你知道吗，还有，我觉得部分原因，就像Rory说的，我现在不再听到“冒险”这个词了，但在以前，比如2024年之前，你会听到风投说，我想在这个领域有点曝光。

You know what, also, I think part of it, to Rory's point, I don't hear this term adventure anymore, but in the old days, like before 2024, you would hear the term from VCs, I want to get some exposure to a space.

好吧，这个领域正在起飞。

Okay, a space is taking off.

我不确定谁会赢，但那个走进办公室的年轻人Rory，看起来像是视频和下一代CRM领域的合适人选。

I'm not sure who's going to win, but that kid Rory that walked in the office, he seems like the right guy in video in next generation CRM.

世界变化比较慢，所以你可以慢慢来，有时候会错过，你会和合伙人商量，你不确定Rory的创业公司怎么样，你知道，你想听到这个词，我们想要一些曝光。

The world moved kind of slowly, so you take your time and one would miss and you'd around with your partners and you weren't sure about Rory's startup, you know, want some you hear this term, we want some exposure.

哈里，这个说法我现在很少听到了，但我还是觉得它正在发生。

Harry's I don't hear this term much anymore, but I still think it's happening.

就像罗里说的，智能体正在爆发。

Agents are exploding to Rory's point.

直觉也在爆发。

Instinct's exploding.

我想在这个领域获得一些曝光。

I want some exposure to this space.

我不知道Thinking Machines是不是正确的选择，但我需要在这个领域获得一些曝光。

I don't know whether Thinking Machines is the right one, but I need some exposure to this space.

比如，我需要接触路由器这个领域。

Like, I need some exposure to the router space.

而且世界变化太快了，你必须快速做出这些决定，但我还是觉得有一些类似的思考。

And the world's moving so fast that like, you have to make these decisions, but I still think there's some similar thinking.

我必须获得一些曝光。

I've got to get some exposure.

我不确定是不是Instinct，但看起来挺火的。

I'm not sure if it's Instinct, but it seems hot.

而且我必须在快速变化的市场里赶紧布局。

And I I just have to get the exposure in in a fast movie market.

这点我完全同意你。

I totally agree with you in terms of that.

我觉得这确实很明显，就像我们之前聊的编程代理，其中让我特别感兴趣的一个是Cognition。

I think I definitely see that in terms of, as we said about coding agents, which to me, one of the fascinating ones was Cognition.

我们之前聊了很多关于他们收购Windsurfer的事情。

We've talked about it quite a lot in terms of what's happened with their acquisition of Windsurfer in the past.

Cognition据说以460亿美元的估值融资。

Cognition raising around at $46,000,000,000 reportedly.

他们今年年底的ARR会达到16亿美元，目前是8到9亿。

They will end the year at 1,600,000,000 in ARR, currently doing 800 to 900.

天哪。

Holy shit.

说实话，我们真的低估了市场规模。

I mean, we really underestimated TAM.

你看Claude Code现在这么火，Cursor也这么火，Cognition年底还能做到16亿的营收。

When you got Claude Code doing what it is, when you got Cursor doing what it is, when you got 1,600,000,000 from Cognition by the end of the year.

对我来说，Cognition更有意思的是，它连前三都排不上，但规模已经这么大了。

To me, the more interesting thing for Cognition is, like, you know, that it isn't in the in the top two or three, and it's still of that scale.

对吧？

Right?

几年前我写过一篇文章，叫“Postmates效应”。我记得好像是红杉资本说的，他们从没想过能靠市场第三名赚钱，就像Postmates那样。

Years ago, I wrote a post, and I called it the Postmates effect, where I think I think it was Sequoia said they never thought they could make money off the number three in a market like Postmates, right?

但后来行情好的时候，它被几十亿美金收购了。

But then when times were good, it got bought for a couple billion.

那时候几十亿美金可是很大一笔钱，对吧？

Back then a couple billion was like a lot of money, right?

不知道现在的年轻人还记不记得，但我管这个叫Postmates效应。

I don't know if kids remember, but I call that the Postmates effect.

所以Cognition就是最厉害的。

So Cognition is like the greatest.

不过，Cognition不一样。

Now, Cognition is different.

它运行时间很长。

It's long running.

它是自主代理。

It's autonomous agents.

它确实不一样。

It is different.

对。

Okay.

但从收入角度看，它是这个领域的Postmates，但这个领域太大了，兄弟。

But from a revenue perspective, it's the Postmates of the category, but it's such a big category, man.

你还想再当一次Postmates吗？

You want to be in Postmates again?

对。

Yeah.

它永远追不上Anthropic，除非世界变了，而世界每个星期都在变。

It's never gonna catch anthropic unless unless the world changes, which it has every every single week.

但除非世界变了，它追不上也没关系，也不需要追上。

But unless the world changes, it's not and it doesn't need to catch them.

只要几年后每年能赚50到100亿美元，就足以让Cognition成功了。

Just just 5 to 10,000,000,000 a year a few years out is enough to make Cognition a success.

它只需要做到50到100亿美元的年度经常性收入就算成功了。

It only has to do 5 to 10,000,000,000 in ARR to be a success.

它不需要追上Anthropic。

It doesn't have to catch anthropic.

我觉得这很合理。

And I think that's fair.

我只是想回应一下关于TAM的说法。

And I'm just trying to respond to the TAM comment.

你看，我之前就说过了。

Look, I said it earlier.

我觉得编程是市场的金矿，对吧？

I think coding is the mother load of markets, right?

这一切能运转起来，全靠它，对吧？

It is the whole reason all this stuff works, right?

那我们是不是把TAM搞错了？

And did we get the TAM wrong?

我是说，我一直在关注的一个方面就是总的劳动力支出和软件，包括在Salesforce、思科工作的人，还有在摩根大通工作的人，包括QA和其他所有岗位。

I mean, one of the ones I started looking at a lot is just total labor spend and software, including people working at Salesforce, Cisco, and then people working at JPMorgan, including QA, all the rest.

美国每年的劳动力支出大概有5000亿美元。

You've got about $500,000,000,000 a year of US labor spend.

我们说过一百遍了，关键问题是其中有多少比例会转化为AI支出？

As we've said this 100 times, the big question is what percentage of that converts to AI spend?

如果是10%，那就是500亿美元的市场，考虑到所有参与者的进展，这有点让人紧张。

If it's 10%, it's a $50,000,000,000 market, and that's a bit nerve wracking given the traction of everyone involved.

如果是20%或30%，那还有很大的增长空间。

If it's 20% or 30%, there's lots of room to go.

而且有可信的理由认为这个比例会更高。

And there's credible arguments that say it's higher.

如果比例更高，比如每年1500亿到2000亿美元，占40%的支出用于编程工具和编程智能，那么任何在这个细分市场里占有一席之地的公司都会受益。

If it is higher, then if you're going to have $150 or $200,000,000,000 a year, which will be 40%, of spend on coding tools and coding intelligence, than anyone who has a subsegment of that.

想想看，有Cursor这个细分市场，某种程度上还有Cursor Claude这个细分市场。

And if you think about it, there's the Cursor segment, to some extent, the Cursor Claude segment.

Cognition可能会说他们现在所处的细分市场略有不同，就像Jason你说的，是长期运行的智能体。

The Cognition would say it's a slightly different segment of where they're playing now in terms of, as you say, Jason, long running agents.

对，另一个极端是Lovable和Replit，它们属于不同的细分市场。

Yeah, the other extreme, Lovable and Replit, are in a different subsegment of that.

它们都在一个潜在500亿到1000亿美元的市场细分中，具体取决于你相信哪个百分比数字。

They're all in subsegments of a potential $50 to $100,000,000,000 marketplace, depending on what percentage number you believe.

所以是的，这里的市场规模巨大。

So yeah, the TAM here is huge.

我觉得我们搞错的地方，抱歉，不是故意打断你，是罗里的算法你没法反驳。

And I think what we got wrong, sorry, didn't mean to interrupt, was the Rory's math you can't argue with.

这个星球上的开发者数量是有限的。

There's only so many humans, developers on the planet.

就算按我的算法，每个开发者10到15美元，这个算法也有上限。

Even if you use my math of 10 to $15 per developer, there's still a ceiling to that math.

但那是自上而下的算法。

But that's the tops down version.

也许应该是自下而上的。

Maybe it's bottoms up.

有时候我会搞混，虽然我不该这样。

Sometimes I get confused, even though I shouldn't.

话虽如此，我们真正搞错的是，人们现在构建的软件量比十八个月前多了100倍。

Having said that, what we really got wrong is people are literally building 100x more software than we were eighteen months ago.

我上周在节目里说过，你的投资组合公司还没深入2027年的路线图，他们就已经在失败了。

And I said last week on the show, like, your portfolio companies aren't deep into their twenty twenty seven roadmaps, they're failing.

对吧？

Okay?

这是真的。

It is true.

就像我们以前要花一个季度甚至一年才能建好的功能，现在虽然不是说真的一个小时或五分钟就能搞定，但一周或一个月就能做出来。

Like we just build features that used to take a quarter a year can be built not really in an hour or five minutes, but can be built in a week or a month.

所以如果你看看你最好的投资组合公司，看看他们发布的速度有多快，发布的东西有多少。

And so if you look at your best portfolio companies, look how fast they're shipped, how much they're shipping.

所以有一个金融上的总可寻址市场，它有一些理论上的逆风，但说实话，我们正在编写的代码量是100倍。

So there's a financial TAM, which has some theoretical headwinds, but literally the amount of code we're building is 100x.

这就是我们搞错的地方。

And so that's what we got wrong.

我们没意识到我们都会在建设复合型公司、复合型初创企业。

We didn't realize we would all be building compound companies, compound startups.

我们都会在建设多100倍的软件。

We would all be building a 100 times more software.

节目刚开始的时候，我们确实搞错了这一点。

We got that wrong when the show started.

我觉得我们就是在TAM上判断错了。

That's where I think we got the tam wrong.

嗯，这个观点挺有意思的。

Well, that's an interesting one.

在AI时代，是不是每家公司都会变成复合型公司，比如RAMP突然开始做模型路由产品，然后快速推出？

In a world of AI, does every company become a compound company where suddenly RAMP is creating model routing products and spinning them up?

你根本没得选。

You have no choice.

你赢不了，因为你的竞争对手都是复合型创业公司。

You can't win because your competitors are compound startups.

他们都在以前所未有的速度互相重叠。

They're all overlapping at a pace we never saw before.

他们都在以前所未有的速度互相竞争。

They're all competing at a pace we never saw before.

对吧？

Right?

也许我们会感到困惑，因为我们经常谈论的那些大模型在很多方面仍然是横向平台，对吧？

Maybe we get confused because LMs we talk about a lot are these still horizontal platforms in many ways, right?

它们并没有像Claude设计那样构建数百个应用，尽管有这些那些。

They're not building hundreds of applications despite Claude design and this and that.

但天哪，B2B应用竞争对手的融合速度，我们从未见过。

But JFC, the rate of convergence of competitors for B2B applications, we've never seen this.

但是，Jason，如果我们再往前推一步，这意味着什么呢？

But, Jason, what does that mean if we expand that one next step?

在一个所有初创公司都必须成为复合型公司的世界里，这意味着什么？

What does that mean in a world where all startups have to be compound startups?

对。

Yeah.

我该如何考虑支持赢家呢？

How do I think about backing winners?

如果我是创始人，正在听这个，我该怎么办？

If I'm a founder listening, what do I do?

嗯，你听就是了。

Well, you you listen.

我的意思是，如果你不发布产品，那这就是认知的问题。

I mean, it's if you're not shipping, then this was the Cognition.

还记得那个谁，CEO吗？

Remember when what's his name, the CEO?

斯科特·维。

Scott Wie.

对，他雇了Windsurf，然后解雇了人之后，他说：我们公司必须每周工作七天。

Yeah, when he hired Windsurf and fired after people, he's like, We have to work seven days a week at our company.

抱歉。

I'm sorry.

那些做不到的人，不是看你待了多少小时，而是你必须以更快的速度发布软件来加速超越竞争对手，因为他们都会成为复合型初创公司。

And the folks that can't do it, it's not all the amount of hours you're at, but you have to be out accelerating your competition in terms of the rate at which you ship software, because they're all gonna be compound startups.

你那个2x2矩阵或者热图上的所有小点。

All the little islands on your 2x2 or on your heat map.

现在没有热图了。

There is no heat map anymore.

全都热起来了。

It's all got hot.

全是岩浆。

It's all lava.

所以如果他们要么卖掉，要么退出，要么派你的初级董事会成员去开会，因为在今天的世界里，他们永远追不上。

And so if they're sell or quit or send your junior board partner to the meeting because they'll never catch up in today's world.

杰森说得对，他凭直觉就答对了，我接下来想现场讲讲经济学，大家耐心听一下，好吗？

Jason's right, and Jason is instinctively right about that answer, and I'm going to try and do economics on the fly, so bear with me, right?

现在基本上发生的是，AI让写代码变得容易多了。

What is basically happening here is AI makes code a lot easier to produce.

我们可以争论它是否取代了，AI支出和软件支出的比例是多少。

We can argue about does it replace what's the ratio of AI spend to software spend?

但毫无疑问，它让软件开发变得容易得多。

But there's no doubt it makes it massively easier to produce software.

你可以相信会发生两种情况之一，而且两种都会发生一些。

And you can either believe one or two things will happen, and a bit of both will happen.

一种是，哦，世界花了10倍的钱，他们继续以同样的价格购买软件，然后在软件上多花10倍的钱。

One would be, oh, and the world spent 10 times they continue to buy software at the same price, and they spend 10 times more on software.

如果你相信这个，那你就是相信童话故事。

You If believe in that, you believe in the tooth fairy.

会有一些增长，但不会发生那种情况。

There will be some increases, but it's not going to happen.

摩根大通不会把它的软件采购预算增加10倍。

JPMorgan is not going to spend increase its software purchase budget 10 x.

所以另一种会发生的情况是，如果每个人都在更快地开发软件，而且终端客户的软件支出有所增长，这点我同意，但增长幅度远不及生产扩张的幅度，那么杰森的设想就是对的，斯科特·吴也是对的。

So the other thing that's going to happen is if everyone is making software more quickly and there's some expansion in the software spend from end customers, which I agree, but not nearly as much as the expansion in production, then Jason's vision is correct, and Scott Wu was right.

我想把话题拉回到你的问题，哈里。

I wanna bring it back to your question, Harry.

最后赢的人，一定是那个最能积累、用这些工具拼命做出最多软件的人，因为工具让他们速度飞快。而那些不24小时连轴转做软件的人，就会被甩在后面。

The person who's going to win is going to be the person who compounds the most, grinds out the most software with these tools that have made them move incredibly quickly, and the person who doesn't grind out software seven by 24 is gonna be left behind.

最终客户会说，让我想想。

And the end customer is gonna say, let me see.

我能从你这里买两个应用，或者从他们那里买一个集成了十个应用的整体套件。

I can buy two apps from you or an integrated 10 person app suite from them.

我觉得我会选那个十个的。

I think I'll go with the 10.

这会是一个非常——我是说，会有一段时期，有些人掌握了新的构建方式，做得很快，你们手里有新组合，而有些人还在用老办法慢慢做。

It's going to be a very I mean, it's going be one of those where there's a period of time when some people get the new way of building and are building quickly, and you all have a new portfolio, and some people aren't building the old way.

你心里其实知道结局会怎样，而杰森说的不对。

And you kind of know in your heart how this is going to end, and it's not Jason's right.

对那些不往产品里塞更多软件的人来说，结局不会好看。

It's not going to be pretty for the people who aren't putting more software in the box.

因为你看，如果Rippling卖十个模块，他们赚的不是卖一个模块的人的十倍，而是四倍。

Because what you're not getting like, if Rippling's selling 10 modules, they're not getting 10 times more than a person selling one, they're getting four times.

但他们跟终端客户说，哥们儿，让我帮你把所有的烦恼都解决掉。

But they're saying to the end customer, Dude, let me make all your pain go away.

这里有10个不同的模块，你都不用单独买了。

Here's 10 different modules you don't have to buy.

现在你就能拿到它们。

Now you can get them.

给我单个模块四倍的价格就行。

Give me four times the price of a single module.

你开心了。

You're happy.

我们也开心，因为我们开发软件速度快。

We're happy because we're building software quickly.

你也开心，因为你省了钱，效率还更高。

You're happy because you're saving money and it's more efficient.

大家都赢。

Everybody wins.

而你没听到的声音，是其他九个单点产品正在消亡。

And the sound you don't hear is the other nine point products dying.

这就是那部电影。

That's the movie.

上周，我参加了一个老板的董事会会议，他们刚刚融资了23亿美元。

Last week, I was at a board meeting for owner, which just raised at 2,300,000,000.

这算是一个次世代AI融合平台。

It's sort of a next ish generation AI infused platform.

我很喜欢那个CPO，他是我合作过的最好的之一，昆汀。

And I love the CPO, he's one of the best I've ever worked with, Q Quentin.

他正在介绍他正在发布的产品。

And he was going over what he's shipping.

即使有这么多投资者，带着他们数亿美元，他们还是觉得，这太多了。

Even with all these investors, with their hundreds of millions, they're like, this is too much.

他们就像一群我们都认识的B和B的人。

They're like a bunch of B and B guys who we all know.

而且这就像，你根本不可能发布这么多软件。

And like, this is just, you can't ship this much software.

然后他说，而且他非常擅长这个。

And he said, and he's very good.

他说，我们都必须成为复合型创业公司。

He's like, we all have to be compound startups.

我们没有选择。

We have no choice.

这就是门槛。

Like this is just the bar.

但实际上，最近已经发布或将要发布的功能和特性数量多得几乎无法处理。

But it's literally amount of features and functionality that has recently shipped or will ship that is almost unprocessable.

但现在他们必须构建餐厅想要的一切。

But now they have to build every single thing a restaurant would want.

整个技术栈的每一个部分。

Every part of the stack.

你再也不能只做一部分了。

You no longer can just do part of it.

他说，我们别无选择。

And he's like, we have no choice.

我甚至都不担心这个。

Like, I don't even sweat it.

我甚至不担心这比一年前多了十倍，因为我们别无选择。

I don't even sweat the fact this is 10 times more than a year ago because we have no choice.

我还注意到，那个营销信息里，客户群体已经变了。

I also noticed in that marketing message that customer base has changed.

实际上，亚当会告诉你，不只是餐馆，因为在他的发布视频里，他非常明确地说，我们是小企业的AI操作系统。

Actually, Adam would say to you, it's not restaurants, because very explicitly in his launch videos, he was saying, we are the AI operating system for small businesses.

这是其中的一部分。

It's part of it.

这有点意思，你可以用AI把它扩展到其他领域。

And that is somewhat interesting that you can with AI, could expand it other verbs.

但Q只是在说他们的核心目标客户。

But Q is just talking about their core ICP.

就是说，他们什么都想要。

Like, they want everything.

他们需要AI前台。

They need the AI receptionist.

他们需要AI点单。

They need the AI order.

他们需要AI，而且他们期望所有这些功能。

And they need the AI and they expect all of it.

如果我们不把这些都做出来，别人也会做出来。

And if we don't build all of it, someone else will build all of it.

我们不能在各自的小圈子里等两年。

Like, we can't wait two years in our little corners of the venture world.

他的意思是，他即将推出的产品数量，不仅一年内大幅加速，而且他已经完全接受了这个现实。

And his point is just the amount he's gonna ship, not only is it radically accelerating in a year, but he's embraced it.

斯科特·吴说，没有别的选择，所以连讨论风投都没必要了。

Scott Wu asked, there is no other choice, so it's not even worth talking about VCs.

就像，感谢我们一直在做的那份点头认可。

Like, thanks the nod that we've been working on.

我们没有，这就是今天的世界。

We have no this is the world today.

世界已经变了。

The world has changed.

不过要承认，我只是为了不显得太天真，这确实有复杂性。

And acknowledging, though, I just so we don't sound wonderment, there are complexities with that.

你可能会陷入产品堆砌的境地，屏幕上按钮太多。

You can veer into product slop, you can have too many buttons on the screen.

这些都是事实，所以你需要一个优秀的CPO，就像你说的，你需要一个优秀的CPO，把大量的产品功能以一种简单易懂的方式呈现出来，但这就是艺术。

All those things are true, which is why you need a great, to your point, need a great CPO, you want to present a lot of product surface area in a fairly simple, digestible manner, but that's the art.

如果你做到了，那你说得对，你就能拿到钱。

And if you do that, then you're right, you get the money.

因为没有一个做生意的人会说，我真的很喜欢用五个不同的SaaS产品，还得把它们整合起来，因为这样才让我兴奋。

Because no one running a business says, I really enjoy having five separate SaaS products and integrating them because that's how I get off my excitement.

我只是说，我这边是在欧洲。

I'm just saying here, I'm the one sitting in Europe.

如果你们俩说的那种复合型创业公司，你们就得融更多的钱。

If you wanna do a compound startup in the way that you both are talking about it, you will need to raise more money.

因为要按那个速度推进，还要在代币上花那么多钱，你们融的钱就得比欧洲那种更保守的轮次多。

Because to move at that pace and to spend what you'll need to on tokens, you'll need to raise more than the more modest European round.

我担心的是，我投资的一些公司融的钱比美国同行少，我觉得他们在走复合路线时可能没法像美国同行那样激进。

What I'm worried about is actually I'm in a a number of companies which have raised less than US counterparts, and I think will be able to be less aggressive in doing the compound route than their US alternatives.

但他们，因为他们没有那么多钱。

But they because they they don't have the money.

他们只融了三百万美元。

They raised $3,000,000.

对，这个问题我真不知道怎么解决。

Yeah, this is one I don't know how to solve.

这确实是个难题。

It's a tough one.

而且，上周Iconic发布的数据显示，在AI时代，各公司的人员增长情况是怎样的。

And the one is initially, the data about Iconic last week put out its headcount data of how much people are growing headcount humans in the age of AI.

不知道你们有没有看到。

Don't know if you guys saw it.

完整报告还没出来。

There's a a full report doesn't come out.

但基本上就是说，任何增长低于超高速增长的公司都不在招人，对吧？

But it basically said anyone growing below hyper growth is not hiring, right?

增长50%到100%的公司，人员增加了25%。

People growing 50 to 100% are adding headcount 25%.

增长50%的公司，人员没有增加。

People growing 50% are adding no headcount.

他们用AI来提高效率，挺好的。

And they're using AI to get more efficient, great.

那些增长率超过100%的公司，平均员工人数增长133%，对吧？

People that are growing more than 100% are growing 133% headcount on average, right?

所以他们不仅是在叠加软件，还在叠加人力。

So they're compounding not just software, they're compounding humans.

他们在大量吸纳人才。

They're sucking in humans.

这个循环只会不断加速。

The spiral just grows.

你根本跟不上。

You can't keep up.

但你必须承认，这个说法本身存在一个绝对的悖论，因为我们刚刚说我们有一个能让工程师更高效的产品。

But you have to acknowledge that there is an absolute paradox at the heart of this comment, because we just said we have a product that makes engineers more efficient.

理论上，如果情况真的只是这样，你应该需要雇佣更少的工程师。

In theory, if that's all that was happening, you should have to hire less engineers.

有意思的是，如果我只是英国增值税申报的软件产品，而且是唯一一家做这个的公司。

And it's interesting, if I was just the software product for VAT returns in The UK, and I was the only company doing it.

对吧？

Right?

有几个竞争对手。

There's a couple of competitors.

对吧？

Right?

AI 出现了。

AI comes along.

我可能能裁掉几个工程师，然后做得更高效。

I can probably get rid of a couple of engineers and do it more efficiently.

就这样。

That's all.

它应该只是更高效，这是第一层影响。

It should just be more efficient, and that's the first order effect.

所以，Harry，你说得对，按理说应该是，嘿。

So, Harry, to your point, it should arguably be, hey.

现在我需要更少的钱来打造这个产品，因为AI让这件事变得更容易了。

I only need less money now to build this product because AI makes it easier.

但我觉得实际情况是，因为更容易了，也因为投资者现在只追求巨大的回报，一旦你开始有任何增长，他们就愿意投入资金。

But I think what happens is because it's easier, because investors are now just looking for huge outcomes, the minute you start getting any growth, they're willing to put capital behind it.

而且记住，在软件公司里，每花一美元在应用级产品上——不是财务模型——你花在研发上的钱，还要再花两到三倍在销售和营销上。

And remember, for every dollar in a software company you spend typically on an apps level product, not a financial model, but you spend on R and D, you spend two or three times that on sales and marketing.

所以你说的是，哈里，赢家会得到这种复利效应，他们开始获得这种增长效应。

So what you're saying, Harry, is the winners get this compounding effect, they start getting this growth effect.

然后杰森说得对，标志性的数据显示，风险投资非常擅长把钱投给那些已经在快速增长的公司。

And then Jason's right, the iconic data says venture capital does a really good job of stuffing money into things that are already growing quickly.

那是我们的默认做法。

That is our default.

这正是我们本能反应的核心。

That is right at the heart of our lizard brain.

如果你半夜醒来，你最想做什么？

If you were to wake up in the middle of the night, what do you want to do?

我就想找那些正在增长的东西，然后把更多钱砸进去。

I want to find shit that's growing and stuff more capital into it.

这就是我的工作。

That's the job.

所以你会看到这种拉开差距的效果。

So you get this kind of pulling away effect.

这就是为什么你有那种担心，哈维，就是你完全可以是一家好公司，但问题是你会不会慢慢变得无关紧要？

And that's why you have that concern you have, Harvey, which is that you can be a perfectly good company, but the concern is are you drifting into irrelevance?

顺便说一句，我并不认为这种情况总是发生。

I'm not convinced it happens all the time, by the way.

我觉得如果你在一个独立的市场里，你会过得很好，能赚钱，能做出你的180，你会没事的。

I think if you are in a separate market, you'll be just fine and you'll make money and you'll put up your 180, you'll be fine.

但如果你在一个容易被邻近领域入侵的市场里，就像杰森说的，那你就不会好过了。

But if you are in a market where the adjacencies can easily invade, to Jason's point, then you're not going to be fine.

三年后你一觉醒来，就发现自己已经不重要了。

You're going to wake up in three years and not matter.