"CEO 必须兼任首席 AI 官"

对。

Mhmm.

那些善于与人建立深厚情感连接的人，能够揣摩他人内心的想法和感受。

That has to do with people that have very good emotional connection with people are able to simulate what the other person is thinking and what the other than others.

心理理论。

Theory of mind.

没错。

Exactly.

我觉得，创业者如果能理解客户没说出口的需求，就能把那些隐含的东西变成明示的——加里就说过这话。

And I think the founders, I get that and have the empathy to figure out what the customer is not verbalizing is what is make the I think Gary says this, make the implicit explicit.

完全同意。

A 100%.

那些需求到底都是些什么？

Of what are all those desires.

完全同意。

A 100%.

这些迹象通常都很微妙，因为它们往往只是低语。

And they're very subtle signs a lot of time because they're murmurs.

对。

Mhmm.

创始人在梳理这些线索时，会慢慢意识到：这还真是个问题？

As founders go through them and figure out the insights that, oh, is this really a thing?

但你怎么知道什么时候该深入追问呢？

But how do you know when to poke for it?

没错。

Exactly.

现在的问题是，过度依赖模型的话，我依然对创始人能创造大量价值充满信心，

And the problem with relying on models and right now, which is I'm still very optimistic that there's still a lot of job through founders

当然。

Oh, definitely.

因为你根本不知道该用什么话术或提示去问模型，连该问什么都不知道。

Is you don't even know what the right incantation or set of prompts to ask the model because they're they're you don't even know what to ask.

没错。

Exactly.

还有另一层更深层的含义。

There's like another meta layer.

是啊，就像那个光锥的概念一样，你得想清楚，你问的这个问题，到底宇宙是在回答哪一个。

Yes, the whole Lightcone thing of like, you know, which question is the universe the answer for, kind of.

当然，这些都是笼统的说法，对吧？

And of course, these are generalities, right?

但我认为关键在于，得记住大语言模型并不是魔法。

But I think what I've seen is you have to remember that LMs are not magic.

它们的训练数据非常特定，优化目标也是针对一组明确的基准和结果。

Like, LMs are trained on a very specific corpse of information, optimizing for a very specific set of benchmarks and outcomes.

我觉得大语言模型最大的问题在于，你根本不知道模型在你问的这个问题上到底见过多少训练数据。

And I think the biggest pitfall of LMs is you have no sense of how much training data the model has seen for the exact thing that you're asking it.

想象一下，每次你问大语言模型一个问题，它都能告诉你：‘这个答案在我的数据里出现的频率是x，而另一个答案只有0.00001x。’

So imagine imagine if, like, every time you ask an LLM a question, it gave you, like, yeah, like, I the the the the sampling frequency of this in my dataset was, I don't know, x, and on this other answer was 0.00001 x.

你所信任的那些东西，完全不一样。

Out of You would trust is very different.

对吧？

Right?

分布差别太大了。

The distribution's so different.

我愿意为这个付费。

Oh, I would pay for that.

这真是个不错的创业点子。

That's a that's a great startup idea.

没错。

Exactly.

那咱们就来做这个吧。

So we should do that.

我们得开发一个能做到这一点的模型。

We need to do a model that does that.

是的

Yeah.

百分之百

A 100%.

我愿意为此付费

I I I would pay for it.

对

Yeah.

因为这很有趣，一旦遇到分布之外的情况，你就可以去填补这些空白

Well, because it's fascinating because then, like, anything that's out of distribution, you just go and, like, fill that in

为了什么

for what

你习惯的那些情况

you're used to.

作为一名构建在大语言模型之上的应用工程师，确实如此

I mean, actually, as an applications engineer on top of the LLMs Mhmm.

这其实是个巨大的优势，百分之百。

That's actually a huge A 100%.

盲区。

Blind spot.

梅尔科和许多其他数据公司正是在做这件事。

And that's what Merkor and a lot of the other data companies are doing.

对他们来说，很多工作就是找出大语言模型的盲区在哪里。

Like, a lot of the jobs for them is to say, well, where what are the blind spots for LLMs?

说来有趣，现在很多数据标注公司正试图搞清楚模型在哪些地方理解不到位。

And and it's funny, like, I I think a lot of the data labeling companies right now trying to understand that bit falls in the models.

但问题是，要发现答案中的缺口，你得是专家才行。

But the problem is, in order to do that, you have to be an expert to know what the gaps are in the answers.

但作为创始人，当你在找创业点子时，你对这个领域其实一无所知。

But the problem as a founder when you're looking for an idea is you know nothing about it.

所以这就陷入了一个认知困境：你既不了解真正的知识边界，又容易误以为自己懂了，其实模型才真正懂。

So so the so there there is a there is a curse of knowledge and and a curse of not even knowing what the bounds of knowledge is, which I think can can can make you believe that you understand something that you actually, you know, the model actually understand.

我能坦白一件事吗？关于我做完 gbrain 之后。

Can I confess something weird about like, after creating gbrain?

现在我用 AI 的方式变了，因为我有了一个真正能用的检索系统。不管遇到什么问题，比如我最近在搞一个特别特别难的、要让人感觉像真人写的提示词。

Now I I do use AI in a different way where now that I have a retrieval system that is actually usable, if I have a problem or question about anything, like for instance, I was trying to work on a really, really, like, the last humanized prompt.

对。

Mhmm.

而且很多这类内容可能还没被广泛传播。

And, you know, a lot of that stuff probably isn't in distribution yet.

比如，维基百科上其实有一整篇关于 AI 写作风格特征的文章。

Like, there's a whole Wiki Wikipedia article about, like, you know, characteristics of AI writing.

但现在我直接问它就行。

But, you know, now I can just go tell it.

花一整天去查。

Like, go spend a day.

把每一篇论文、每一篇文章都翻一遍，通读所有内容，整理进我的代码库，之后就能随时调取了。

Like, deep research literally every single paper, article, like, read everything, put it into my Git repo, and then I'll be able to retrieve it

是的

Mhmm.

然后把它总结成真正能用的东西。

And summarize it into something that actually is usable.

这就相当于把100%的空白都填满了。

And so that's sort of like filling in A 100%.

那些超出常规范围的内容。

The things that are out of distribution.

对于100%的情况，我都可以塞进任何相关的背景信息。

A 100 I can sort of like pack it with whatever context.

其实，任何事情都可以这么做。

And it's like, you can do that with anything.

比如你对开餐厅感兴趣，完全可以去读五百本关于开餐厅的书，把所有顶尖的书都读一遍。

It's like, if you're interested in, you know, running a restaurant, literally, you could have, you could go and read like 500 books about like every the top 500 books about what it's like to run a restaurant.

这样你就拥有了关于这个主题的全部知识汇编。

And you would have like the compendium of all information about it.

对。

Yeah.

我觉得我们现在在Brex做的很多事情，比如构建一个客户全景模型，其实也是类似的想法。

And I think a lot of what like, for example, like, one of the things that we we do at Brex now is building this customer world model.

我们试图捕捉客户与我们互动的每一个接触点。

It's a similar idea where we're trying to get every single touch point that the customer has of us.

从他们在仪表盘上点击了多少次按钮，到他们在邮件里提到我们、打电话过来，或者发送语音消息，我们都会把这些数据收集起来加以利用。

Like, literally, like, what how many times they click a button in the dashboard all the way to where they tell someone on an email or where they say on the phone or they send a call and ingest that enhanced data.

那么，这个客户接下来最需要我们提供什么？

Okay, what should this customer need next from us?

这个客户应该考虑哪些问题？

What should this customer be thinking about?

他们还没遇到，但将来可能会碰到的困难有哪些？

Like, what are the issues that they will face but haven't faced?

归根结底，这还是个数据覆盖范围的问题。

And again, it's just a distribution problem.

这其实是在回答其中一个问题是，会不会有工作之类的。

This is actually an answer to one of the questions, which is like, will there be jobs or whatever.

只要内存还有限制，那就会有。

It's like, as long as there are limits on RAM Mhmm.

确实会有的。

Actually, like, there will be.

对。

Mhmm.

所以我也说不准。

So I don't know.

说实话，这个问题挺有意思的。

I mean, it's kind of an interesting one.

对吧？

Right?

我也这么觉得。

I think so.

实际上，不可能有一个模型能包含你所有可能需要的参数，覆盖所有分布场景。

Literally, you can't have a model that has enough parameters that could, like, have everything that you could possibly need in distribution.

宇宙里的原子根本不够用。

Like, there aren't enough atoms in the universe.

对吧？

Right?

这本质上是个建模问题。

It's like a modeling problem.

我觉得我们常常忘了，模型所依赖的世界观其实是被设计者塑造的，最终会影响模型的行为。

I think we forget that the world models in which the models are trans like, there is something that the designers of the models influence the way the model actually behaves in the end.

所以我们花很多时间思考的是，怎么让大语言模型适应和我们完全不同的人群。

So so, you know, one of the things that we spend a lot of time thinking is like, how to make LMs work for people that look very different from us.

比如怎么让模型适应美国普通财务人员的需求——如果你问一个问题，模型默认把AI资本支出当作分类答案，这例子就挺搞笑的。

How to make LMs work for, like, the average finance person in The US that if you're talking about an answer and, you know, the model defaults to, like, AI CapEx as a as a default category for, like like, for example, that's a really funny example.

我之前在试用AI做会计分类的时候就遇到过。

Like, I was playing with AI for accounting categorization.

第一个例子就是写财务分类，比如AI资本支出。

And then, like, the first example of, like, an example of an it's just, like, writing pros, and an example is, like, AI CapEx.

我就想，为啥它第一个蹦出来的就是AI资本支出？

And I'm like, oh, why is it AI Capex?

因为开发模型的人脑子里就只有AI资本支出这一个念头。

The first example it comes up with because the people that are building the models fucking only think about AI Capex.

对吧？

Right?

像这种问题其实挺值得想的，模型背后的思维模式其实比我们想象的要偏得多。

So so there there are things like that that I think is like a kind of interesting to think about that that the the mental models of the models, I think, are out of the box, are are more biased than we may give them credit for.

说到AI资本支出，你刚才说我们还处在特别早期的阶段。

I mean, speaking of AI CapEx, like earlier you're saying, like, you know, we're we're so early still.

说实话，我也说不准。

I don't know.

我觉得最有趣的是，我老是不自觉地用以前玩加密货币时的那些想法来想AI。

The funniest thing about AI to me is how often I find myself thinking crypto maxims.

是的。

Yes.

现在的模型表现是最差的。

This is the worst the models will ever be.

没错。

Yes.

我现在最喜欢跟那些讨厌AI写代码的人说：那你就好好享受每秒只跑一倍速度的编程吧。

My favorite now is telling people who hate AI coding, like, have fun coding at one x speed.

没错。

Exactly.

没错。

Exactly.

我跟一个朋友聊过，怎么理解长推理这个观点——其实就是说，实际需要的推理量会比大多数人想象的多得多。

I was telling a friend about, you know, how how how to how to be, you know, long inference that basically the thesis that there's gonna be a lot more inference than people think.

现在大家看公开市场、半导体供应链这些，都觉得推理需求会很大。

And people are expecting a lot of inference if you just look at public markets and, you know, semi supply chain, all that.

有人说会达到一万倍。

People are saying like 10,000 x.

没错。

Yeah.

但有点讽刺的是，我觉得现在只有一张图。

But but the the underwriting, which is kind of funny, is that I think there's one image.

两千五百个点，每个点代表全球三百二十万人。

2,500 dots, each dot is a 3,200,000 people on the planet.

基本上，全球百分之八十四的人从来没用过AI。

And basically, you know, 84% of the world never used AI.

百分之十六的人至少用过一次免费聊天机器人。

16% have used at least once a free chatbot.

而只有百分之零点三，也就是大概六七个格子的人，愿意每月花二十美元买AI服务。

Then 0.3%, which is, I guess, six or seven squares, pay 20 a month for AI.

在这两千五百个格子里，只有一个格子的人真正用上了AI代理。

And one box out of the 2,500 actually use agents in in whatever capacity.

所以这就是支持长期推理的理由。

So that's the that's the argument to be long inference.

我觉得这才刚刚开始。

And I think it's I think it's just it's just starting out.

我觉得有意思的是，这很可能会成为公司最大的开支，毫无悬念。

And I think a funny thing on on this is I think the you know, it will be the biggest expense in a company, like, easily.

对吧？

Right?

现在token确实利润空间很大，但大家总是想站在最前沿。

And yes, there's a lot of margin in tokens right now, but people always want to be the bleeding edge.

就算token成本降十倍，使用量也会涨十倍。

But even token costs decrease by 10x, they're going to have 10x more usage.

所以开销依然会非常大。

So it will be a still large cost.

我们花了很多时间思考，怎么帮企业更好地控制token支出。

And we're spending a lot of time thinking how to help companies actually manage token spend.

在Brex，我们最终自己开发了一个这样的内部系统。

On Brex, we we ended up building our internal version of this.

我们叫它Magpie，核心想法是，公司里每花一美元的token，都能追踪到它对应了哪个产品、哪个客户、或者内部员工使用的工具，帮助我们了解模型的使用情况等等。

We call it magpie, where the idea is you can effectively, you know, every dollar of token spend in the company, can attribute to product we have to customers, an internal tool that we use to serve or internal employee and understand model usage, etcetera.

我们现在正在琢磨怎么搭建数据分析，搞清楚我们到底用这些token在做什么，从而初步评估投入回报。

And and we're now figuring out how to build analytics on what are we trying to do with the tokens to start to get a sense of ROI.

不过说到底，这真是个很有趣的话题，现在才刚起步，跟未来相比，目前的探索还非常初级。

But, anyway, it's a fascinating topic that I think has a lot of, like, early early early work compared to what it will be one day.

你能分享一下Brex这边收集到的数据吗？比如整个行业里token的花费情况是怎样的？

Can you share any of the data that you've gotten from Brex about, like, what token spend is like in the economy?

在持续增长。

It's increasing.

不是。

No.

我觉得有两件事挺让人意外的。

Look, I think two things are surprising.

一方面，就像你之前提到的，我们该怎么看待令牌的使用上限呢？

One is, I think, to to your point earlier on how do we how do we look at token maxing.

我确实觉得，公司内部设定的成本边界，会直接影响令牌的消耗量，这很明显。

I do think there's such a thing as how much cost boundaries you create internally dictate token consumption, obviously.

但我觉得更有趣的是，看看我们现在所处的这十英里范围，如果算上纽约，令牌消耗量巨大，而且我们已经看到数据表明，这些地方的收入增长速度更快。

But to me, I think what's the most fascinating is when you look into the sort of 10 mile radius we're in now, and if you include New York, tons of token consumption, and you could probably argue, and we've seen the data, faster revenue growth.

真正有意思的是，这十英里圈内和圈外其他地方之间的巨大差距。

I think what's really interesting is the gap between anyone in these two ten mile radius and everything else.

这可不是小公司的情况。

And this is like not small companies.

你去看那些预算庞大的大公司，它们的令牌使用量可能已经到极限了。

So you look into, like, very large companies with very large budgets, and that that could be token maxing.

对它们来说，最合理的经济做法就是把令牌用到极致。

And and the economic thing for them to do would be to token max.

它们每个月的花费可能高达一万美元。

And they spend like, I don't know, 10,000 a month.

你就会想，你其实应该多花十倍、二十倍，甚至一百倍的钱。

And you're like, you should probably be spending 10 times more or 20 times more or a 100 times more.

这还是让人挺意外的。

That's still surprising.

我觉得原因跟一开始说的有点类似。

And and I think the reason is, again, sort of similar to the the point at the beginning.

两年前半前，我曾和公司里很多工程与产品负责人坐下来讨论过一个问题：如果我们现在从头开始做Brex，答案会完全不同。

Like, we did this exercise two and half years ago where I sat down with, you know, a lot of the the engineering product leaders in the company, we had this question, which is if we started Brex again in 2024, the answer would be even more different now.

我们会做出哪些不同的选择？

What would we do differently?

结果发现，每一件事都得改。

And turns out, like, everything.

我们开始往这个方向思考，结果简直让人抓狂，他们就说：好吧。

And we started going down this route, and it's like, it's kinda maddening because they're like, okay.

我们至今还在用一套完全过时的方式，去思考公司的根基、产品的构建方式，以及内部流程的设计。

We we have this, like, completely old way of, like, even thinking about the fabric of the company and and the way we build the product and the way we build our processes internally.

最好的答案是：我们真希望当初从现在开始做起。

The first best answer is yes, we wish we had started now.

次好的做法是，干脆行动起来，彻底改变我们的工作方式。

Second best answer is like, let's go do something about it and change the way we do things.

对吧？

Right?

我觉得我们在引入人工智能时，很多时候都忽略了停下来想一想：不仅解决问题的方法变了，连问题本身的意义都不同了，得退一步重新思考。

And I think a lot of our approach in terms of, like, adopting AI has also been, you know, how do you how do you pause and say, okay, like, there is a discontinuity in the not just in how we solve the problem, but on what the definition of the problem actually even is, and sort of take a step back and rethink it.

这种例子其实多得数不清。

And, know, like, there's there's, like, millions of examples of that.

举个挺有意思的例子，我们现在正在重新设计客户身份核实流程。

But, you know, one example, which is kind of funny, is, you know, we're redesigning our KYC process.

每次新客户入驻，我们都得做一连串的核实检查。

Like, whenever we onboard a customer, we have to do all these checks to KYC the customer.

过去，身份核实这事儿，80%可以自动化，剩下20%还得靠人工。

And KYC historically is something that you can automate, like 80% of it, 20% is manual.

当然，最初的想法是让一个智能代理来完成这件事。

And of course, the original impetus for anyone is that it's with an agent that does it.

没错，我们本来可以就这么做。

Yes, we can go do that.

但我们决定彻底重新设计整个流程，从头到尾 overhaul 一遍。

But what we decided to do is actually say, let's redesign the entire process end to end.

于是我们重新设计了整个客户入驻流程。

And then what we redesigned is the entire onboarding process.

当你重新设计整个入驻流程时，就会发现销售漏斗最前端有一个非常关键的环节——客户资格筛选。

And when you redesign the entire onboarding process, what you realize is there's a very important thing that happens in the beginning of the funnel, which is deal qualification.

这个客户真的有资格成为 Brex 的客户吗？

Like, is this customer even remotely qualified to be a Brex customer?

但既然 KYC 是免费的，我们就可以对潜在客户而不是正式客户做 KYC。

But when you have KYC for free, you can you can KYC a lead versus a customer.

这样一来，风险评估就提前到了销售漏斗的上游，这彻底改变了我们的目标客户群体，因为我们现在能提前知道谁最终能通过审核。

So you start to have risk orientation up in your funnel, and that changes who you even target because you know who's gonna qualify.

信贷方面，情况也差不多是这样。

And the same thing is true for credit to some degree.

所以现在问题的边界已经变了。

So now the bounds of the problem have changed.

很多人，包括我们的不少竞争对手，都采取这样的做法：觉得我这套老流程还在，那就直接在上面加点AI。

And and you can go in and say and and I think a lot of well, including a lot of our competitors had this approach of saying, oh, I have this entire old process.

或者在我们的产品上硬套个AI功能。

Let me go and, like, latch on AI on top of it or latch on AI on top of our product.

但真正让我们取得突破的，恰恰是反过来想：咱们先把旧方法搁一边，要是从零开始，为今天的公司重新设计会怎么样？

And I think the the biggest discontinuities in a positive way that we've had were when we said, hey, let's keep this old way here, keep putting it in a corner, and like how would we design it for sort of the company today from scratch?

然后就这么做了。

And then just doing that.

这需要创始人花点精力，但这是我们唯一看到真正能带来转折的做法。

It takes a little bit of founder energy to do that, but I think it's the it's the only thing we've seen working to to really sort of inflect.

这让我想起很久以前的一件事。

I think that reminds me a lot about this is sort of way back.

你有没有试过编译 Arc 版的 Linux 发行版？

I don't know if you ever tried to compile Arc distributions of Linux.

对。

Mhmm.

Arc Linux 和 Ubuntu 的用户群体文化差异很大。

The culture within power uses of Arc Linux versus Ubuntu is very different.

是的。

Mhmm.

差别确实很大。

Very different.

我觉得 Ubuntu 的用户更像是那些去试 ChatGPT 的人。

I think that Ubuntu people kinda feel more like people that try ChatGPT.

很多东西开箱即用，特别省心。

Stuff kinda just works out of the box.

有些东西你很快就能上手用起来。

There's some stuff that you can get up and running.

顺便说一句，用Linux的人其实还是不多。

There's still not a lot of people that use Linux, by the way.

对。

Mhmm.

我觉得这跟现在AI所处的状态挺像的。

Which I think it feels where AI is.

但用Arc的话，那你就是真·硬核玩家。

But with Arc, you're, like, super hardcore.

对。

Mhmm.

我觉得OpenClaw和Hermes也是这样，你得花大力气按自己的需求去定制，不断维护技能，还得管一堆配置文件。

And I think that's what OpenClaw and Hermes feel like because you have to really customize it to your own own use case, maintain your skills, have all the markdowns.

一旦搞定了，你真能做出特别牛的东西。

And if you get it working, you can build something awesome.

我见过的最让人惊叹的Arc项目之一，就是有人用它做了个类似Valve的SteamOS系统，就是那个让设备用起来像Nintendo Switch的操作系统，对。

One of the most impressive thing I've seen people build with Arc is actually I don't know if you know Valve, the Steam engine, the operating system that runs that makes it feel like a Nintendo Switch Mhmm.

实际上是基于 Arc 构建的。

Is actually built on top of Arc.

哦，这挺有意思的。

Oh, interesting.

他们自定义了所有驱动程序，还支持无线更新。

They customize all the drivers, over the air updates.

它能兼容所有游戏主机。

It works with all consoles.

各种硬件都能开箱即用。

It worked with all sorts of hardware out of the box.

但他们做了极其深入的定制。

But they super duper customized it.

我觉得现在发生的就是这种情况。

And I think this is kinda what's happening.

如果你能把 OpenClaw 配置得特别顺手，就能为自己打造一个专属的 Nintendo Switch，用来做任何你想做的事。

If you get your OpenCLOUT to work really well for you, you could kinda build your own custom Nintendo Switch for whatever you need to do.

是的。

Yeah.

我总是跟人说这么一件事，挺有意思的，就是想想两年前的你。

I I always have this thing that I tell people, which is which is funny, which is think about your time two years ago.

我觉得你现在比两年前忙多了。

Like, I feel like you're working a lot more now than two years ago.

对吧？

Right?

估计在座的各位也都差不多。

And probably same for everybody here.

那问题就来了， productivity 呢？

So then the argument is, but what what about the productivity?

生产力去哪儿了？

Where's the productivity?

对吧？

Right?

这周我跟一家大型上市公司的首席财务官聊了聊。

And and I was talking to the CFO of a very large public company this week.

她告诉我，我们看到大家用的计算资源越来越多，也在试着衡量产品迭代速度，发现代码提交量确实上去了。

And she was telling me that we see all the soaking consumption and and and, you know, we're trying to measure, like like, product velocity, and we're seeing, like, more lines of code pushed.

所以，没错，也许用这个来算投资回报率也行，但真的是因为大家花在令牌上的钱太多了吗？

So so, yes, maybe that's the way to measure the ROI, but but but is it really there because people are spending so much on tokens?

我觉得，当然，去分析令牌的投资回报率确实重要。

And and I think the I think I think this analysis, like, yes, of course, I think having a sense on ROI on tokens is important.

但你忽略了一点：我们现在正站在历史的时间线上，才刚过去六个月，电才刚被发明出来。

But I think it misses the point that you're standing in the timeline of history, and it's six months after electricity was invented.

想象一下，要是有人在19世纪80年代说：‘哎呀，我电费太高了，得省着点用。’

Like, thinking about like, imagine someone saying in in, like, I don't know, '18 the eighteen hundreds, like, oh, my electricity bill is so high now.

‘咱们别用电了，还是继续用蒸汽机吧，省点钱，反正迟个二十年再升级也行。’

Like, gosh, let's use a little less.

当然，从省钱的角度看，这么说好像有道理，但根本不是这么回事。

Let's keep let's push this steam engine to come, like, maybe twenty years later because the cost savings like like, yes, of course.

我的意思是，别为了用令牌把公司搞破产了。

Like, I'm like, don't bankrupt your company on tokens.

这个比喻太贴切了，你可能不知道，电力刚发明的时候其实根本不好用，投资回报率也很差。

It's actually a perfect analogy because I don't know if you knew this, but when electricity was first invented, it didn't work very well, and the ROI was actually bad.

所以如果电力刚出现不久，有会计做这样的分析，他们肯定会说，这电力玩意儿

And so if shortly after the invention of electricity, some accountants had done this analysis, they would have been like, this electricity thing is

根本不可能成气候。

like, isn't never gonna be a thing.

投资回报率太低了。

The ROI sucks.

人们干嘛还要坚持用它呢？

Why do people stick stick to it?

根本不是为了省钱。

And and it wasn't it wasn't the cost savings.

就是单纯因为大家对它感到好奇。

It was just because people were curious about it.

我觉得，我昨天一直玩到凌晨两点，折腾那些斜杠工作流和OPUS 4.8，根本不是为了赚钱，就算没收入我也一样会干，因为你看到的是可能性，看到它能怎么改变技术。

And I think it's I I think the point of, like, why, you know, like, I was yesterday until 2AM playing with slash workflows and OPUS 4.8 and and all of that is because I think I'll be doing the exact same thing if I wasn't making any money because you just see the possibilities and you see what it can do to technology.

这种东西真的会让人行为方式不一样。

And that just drives people to behave differently.

对我来说，这才是真正的试金石，也是最清楚的分水岭。

And and I think that to me is the the ultimate litmus test, and and it's a good separator.

当然，如果代币现在这么贵，长远来看，大概率会变成免费的。

And and sure, if tokens are so expensive, they're gonna be, I think, the fullness of time, probably free.

一个项目要是能在线上存在一百年，就像现在的电一样，我们日常根本不会去想电费这回事，除非你是在数据中心工作。

If a project is, I don't know, a hundred years online, almost compared to what electricity now we don't think of electricity costs in our day to days, but unless you're in a data center.

但我相信，这里头确实有相似的地方。

But but I think there's something there's something similar for sure.

我们接触过很多后期公司的创始人，他们都希望自己的公司能尽可能深度整合AI。

We talked to a lot of founders of later stage companies, who wish that their companies could be like as AI pilled as possible.

你现在运营着这么大的公司，有这么多员工，这还只是Brex这边，Howard那边也想要同样的东西。

And you run this like big company now with all of these employees, and that's only the Brex side, there's also the like, Howard wants that.

我想知道你是怎么让整个公司

I'm curious what you've done to,

跟上你的步伐的，

like, bring the rest of

如果你对其他人有什么建议的话，

the company along with you on this journey, and if you have advice for other people

尤其是其他首席执行官。

or other CEOs.

需要做的事情太多了。

There's a lot to do.

我觉得首席执行官就得亲自当首席人工智能官。

I think the CEO needs to be the chief AI officer.

这可不是工程团队的事。

Like, it's not a engineering team thing.

也不是产品团队的事。

It's not like a product team thing.

你得比任何人都更清楚这项技术的边界。

It's like, you have to understand the bounds of the technology better than anyone.

我觉得，除非你每天都能亲身体验到技术的极限，否则根本难真正理解它能做什么。

I would argue that unless you unless you really experience the limits of the technology every day, I think it's really hard to even understand what it can possibly do.

你知道为什么吗？

Oh, you know why?

因为除了董事会，没人敢对CEO说不，而董事会也不会深入到具体细节里。

It's because nobody can say no to the CEO except the board, and the board won't be in the weeds per se.

这完全没错。

That is a 100% true.

你想想我们之前说的KYC这个例子。

When you go think about, like, you know, the whole the whole example of KYC that we were saying.

KYC团队根本不会想到用KYC技术来给潜在客户打分。

Like, the KYC team would never think of using the KYC technology to score a lead.

只有当你掌握了全局的背景，才可能去思考整个系统的组织方式。

The only people that can think about the organization of the system itself is if you have the context of the whole.

对我来说，任何一个CEO最需要回答的问题是，别管竞争对手了。

And and to me, like, the the the single most important question that any CEO needs to answer is forget about the competitive landscape.

想象一下，如果你能把今天的技术水平直接带回你刚创业的那一刻。

Imagine you could get the state of the technology today and transport to the the moment you started your company.

机会还是那个机会，但打造一家公司的可能性却完全不一样了。

The opportunity was still the same, but but just the possibilities of the way to build a company are totally different.

你会怎么做？

How would you do it?

然后把现在的情况和那时的设想对比一下，先默默承受一阵子，因为你肯定会。

And then diff this versus what you have, and then first suffer in silence for a little bit because you will.

我每天都在经历这种事。

I I mean, I I do every day.

但接下来第二步是，那该怎么办？

But then the second thing is, okay, what do do about it?

如果你从零开始，你会怎么干？

And how would you do it if you were starting from scratch?

你会亲自去琢磨，怎么设计我们的入职流程，怎么打造增长引擎、用户获取方式，以及如何与用户沟通、整合数据。

You'll be the one figuring out, okay, how do we design our onboarding process or how we design our our growth engine and our customer acquisition and the way we talk to users and the way you synthesize the data.

所有这些都得从头重新设计。

And all of that would be would be redesigned from scratch.

所以我觉得，你本质上得重新定义公司的核心理念，重新构建各个职能和团队对成功的理解方式。

So so I think it's like it's almost like you have to sort of refound the very concept of what the what the the company self identity is and the way the functions and and and people's sense of success get structured.

在我们内部，AI其实是个总称，大致包含三个方面。

AI is is a umbrella that I think has like three things, the way we talk about it internally.

第一是产品型AI，也就是直接交付给客户的产品功能。

There's product AI, the product actually ships to customers.

第二是运营型AI，直接影响我们规模化服务客户的能力。

There's operational AI, which is things that directly affect our ability to serve customers at scale.

比如客户成功、风控、入职、运营这些环节。

Like, think of customer success, risk, onboarding, operations, etcetera.

第三是企业内部AI，也就是员工日常怎么用AI来工作。

And there's corporate AI, which is how people work internally.

这三个方向都很重要，而且重要性会随着公司所处的阶段而变化。

The three agendas matter, and they matter in different ways depending on the timing of the company.

有些人往往会把自己局限在其中某一个方向上。

And and I think people will sometimes sort of pigeonhole themselves in one of the three.

但事实上，你得退一步想想，就像我们刚才说的那样。

But in reality, I think you have to take a step back and be like, you know, the same same thing we're talking about earlier.

为什么不能用人工智能解决所有问题呢？

Like, why can't you solve everything with AI?

说到底，这正是核心问题。

Like, at a limit, that's the question.

从这个问题出发，再围绕它一步步找解决方案。

And and then sort of start from there and sort of problem solve around that question.

这其实是一种彻底的转型。

It's a turnaround almost.

我觉得，如果你是一家非原生AI的大公司，某种程度上你就是在进行一场转型。

I think you have to assume that if you're a big large company that's not AI native, you're doing a turnaround to some degree.

我们一直拿富士康的工厂开玩笑。

I guess we've been making fun of Foxconn factories for some time.

但另一方面，你仔细看看，它们简直就是高效到极致的典范。

But on the other hand, like, if you look at them, they're like this paragon of, like, very extreme efficiency.

是的。

Yeah.

而且它们的设计初衷就是如此——不停地重复生产同一种产品。

But they also are designed to be that, to like create like one thing perfectly back to back to back.

所以你得建这样的工厂。

And so you have to build a factory like that.

大多数公司都是按这种方式设计的。

And most companies are designed that way.

对吧？

Right?

我觉得，流程的设计本来就不希望轻易改变。

I think like processes are designed not to change.

对。

Yep.

总得打破一些东西才行。

There is a certain amount of broken glass required.

我觉得，CEO打破障碍比高管容易十倍。

The question is how like, I think it's 10 x easier for the CEO to break glass than an executive.

天啊。

Oh, my

高管比普通员工又容易得多。

easier for an executive than an employee.

经常有人来找我，说他们想用AI做点什么，但别人不答应，理由是这个场景还没测试过，那个场景也没验证过。

So, you know, a lot of times, like, someone comes to me and says, I'm trying to do this with AI, but someone is saying no because we haven't tested this in this use case or in that thing.

我就说，嗯。

And I'm like, okay.

你到底想做什么？

What are you trying to do?

你明白这其中的风险吗？

Like, do you understand the risks?

你清楚那些限制条件吗？

Do you understand the guardrails?

明白。

Yes.

好的。

Okay.

我花十秒钟就能解决这个问题，但别人得花十个小时去开一堆会、层层上报、搞清楚到底能不能用AI做这事。

It takes me literally ten seconds to solve that problem, And it would take someone ten hours to go in into the meetings and escalate and understand, okay, can we do this with AI?

或者根本就做不成。

Or maybe never.

我觉得结果大概率是做不成，因为大多数人会说：

And I think the conclusion is probably never because most people would say, you know what?

我还是老老实实用老办法做这个产品吧，毕竟为什么不呢？

I'm just gonna like build off this product in the old way because like, why wouldn't we?

这事儿就这么成了。

It just works.

我们知道它就在这儿。

We know it's here.

那 guy 肯定恨死我了，以后每天在食堂碰面都得躲着走。

Well, that's got that guy's gonna hate me, and then I have to look at that person in the lunch line every day.

可我又希望大家都开心、喜欢我，所以干脆就不做了。

And it's like, I want people to be happy and like me, so I'm just And not gonna do

我对别人说的是，公司里的上报流程得降低敏感度，因为公司天生会排斥任何破坏内部和谐的动静，就像免疫系统一样直接过滤掉。

what I tell people is that I I think I think the escalation paths need to be like desensitized in the system because the company builds antibodies against any sort of, you know, disturbance to the social cohesion of the company typically gets like rejected by the antibodies.

我觉得应该让上报更快一点，比如直接说：

And I think making escalations faster and being like, hey.

咱们试试这个新办法吧。

We're gonna go try this thing.

你懂的。

You know?

我明白这其中的风险。

I I understand the risks.

咱们还是冒这个险吧。

Let's let's take this risk.

因为最大的风险，其实是错失这个机会。

Because the biggest risk is not taking that.

这本质上就是错过了重新思考问题的机会——如果你今天从零开始创办这家公司，你会怎么做？

It is it's just literally missing the opportunity to rethink a problem from what would you do if you started the company today?

关于企业AI这一块，你认同杰克·多西的观点吗？就是每家公司其实都在试图打造自己的专属通用人工智能？

On the corporate AI sort of leg of that stool specifically, like, do you buy into sort of like the Jack Dorsey view of every company is essentially trying to, like, build its own little company AGI and

我认同，不过我的理解可能有点不一样。

I I do, but but maybe in a slightly different way.

我觉得领域特异性很重要。

I do think domain specificity matters.

我不觉得应该搞一个万能的公司模型，把所有数据都塞进去，没有任何筛选或视角。

So, like, I I don't believe in the, like, well, I'm gonna have, like, a single company model that has, like, every piece of data, like, a single like, with no judgment or lens into anything.

所以我更倾向于用虚拟员工这个比喻来想，就是怎么打造一个代理，能彻底理解每个客户真正关心的一切。

So and and the way I think about it more is, like, is more the sort of the the the virtual employee analogy, so to speak, which is, like, how do I build an agent or a virtual employee that is exceptional at understanding everything that matters about this customer?

这是一个定义清晰、边界明确的问题，有清楚的接口，知道谁依赖这些数据、谁和数据互动。

That is a well defined problem with clear boundaries of, like, clear APIs of who people who who depends on the data, who interact with the data.

这个系统是自成一体的。

That that is self contained.

然后还可以有另一个代理，负责处理另一件事。

Then there's another another agent that can be, okay.

基于我们所有客户及其面临的问题，该怎么规划产品路线图？

Given all the customers that we have and the problems they have, how do I manage my product road map?

这个可以是另一个独立的代理，但它建立在客户数据的基础之上。

That can be a separate agent, but that that builds on top of this customer world

地图。

map.

相当于一个虚拟的高管团队。

A virtual exec team, basically.

没错。

Exactly.

功能性和领域知识依然重要。

Functional and domain knowledge still matter.

对吧？

Right?

这些东西不会消失。

These things are not going go away.

我觉得知识的组织方式，现在还是对的。

And I think the the way knowledge is structured, I think still is still true.

对吧？

Right?

这一点其实没太大变化。

That doesn't necessarily change that much.

你应该把生成代码的代理和系统，跟直接跟客户沟通的系统、以及分析客户对话并转化为产品路线图的系统分开。

And and you should separate the agent that and the systems that are actually emitting code from the system that is talking to customers and the system that is reasoning about the conversations with customers and translating into a product roadmap.

这三样东西是完全分开的。

These three three separate things.

我们在AI领域就像特斯拉一样。

We're kinda like the Tesla for AI.

我不相信任何没有实际应用的东西。

We're like, I don't believe in anything that doesn't have real usage.

我是这么想的，就算我做出了个很棒的模型，我也得问，到底有多少人在用？

It's like, yeah, I build this great model, and I'm like, okay, how many people are using it?

它真能替代公司里雇人干活的需求吗？

Is it actually displacing the need to hire a person inside a company?

它真能省下成小时成小时的人力吗？

Is it actually displacing the need to, you know, spend literally hours?

这东西到底能节省多少工时？

Like, how many hours is this thing saving?

很多时候，大家只会说，哦，这模型挺酷的。

And and I think a lot of times people say, well, you know, it's a it's a cool model.

我就觉得，没错。

And I'm like, yeah.

但光这样可不够。

But, like, that's not that's not gonna cut it.

对吧？

Right?

一旦有了这个方向，我觉得就该以客户为中心了。

Once you have that orientation, I think customer role model.

明白了。

Okay.

比如说，我们的销售团队现在就是按照这个客户角色模型在运作。

Like, your or for example, our client sales team now runs on our customer role model.

不错。

Cool.

所以我清楚这招是管用的。

So I know it works.

我明天要和一个客户一起吃午饭，但对这个账户的实际情况，我了解得还不够充分。

I'm I'm I'm actually having lunch of a a customer tomorrow, and I don't know the state of that account as well as I probably should.

客户角色模型帮我解答了这个问题。

Customer role model answered the question for me.

现在我有一份报告，里面包含了团队之前不知道的信息，比如通过支持工单反映的问题，还有位高管在机场刷卡时遇到的麻烦。

And I now have a report with including things that the team didn't know about that came through support tickets and, you know, an executive that was traveling had an issue at an airport with their card.

所以这些信息都变得更有意识了。

So all these awareness.

全面的信息洞察。

Total information awareness.

对吧？

Right?

这是一个明确的问题。

That is a well defined problem.

而且它正在起作用。

That is working.

我可以把这个模块当作公司整体模型中可靠的一部分。

I can trust this building block as part of my company model as a whole.

而且你可以对它做评估。

And and you can have evals on it.

我们其实一直陷在评估这个事儿上，怎么都绕不开。

Like, we know, like, you know, you you I think a very we're stuck on evals.

这里面有很多经验，是怎么把评估真正融入公司的日常运作的。

There's a bunch of learnings on this and and how to build evals into the fabric of a company.

不过不管怎样，我觉得关键是得把问题拆开来看。

But but anyway, I I think it's more of like, you have to decompose a problem a little bit.

对。

Yeah.

我最喜欢的是用不同模态的评估结果互相验证。

My favorite thing about evals is just running cross modal evals against each other.

我们正在做的一件事特别有意思，就是当公司引入AI代理后，怎么让每一个员工的互动都变成一次评估。

So one of the things that we're doing that I I is related, but I think it's really fun, which is how do you have every single human interaction in the company becoming an eval when you have an AI agent?

比如，我们让入职机器人做一些事情。

So for example, we have the onboarding agents doing something.

然后会有团队专门查看模型识别出的KYC异常情况。

And then you have a team that actually goes in and looks at KYC exceptions that the model can figure out.

怎么让这个过程变成一次关键改进？

How to make that a breaking change?

这个手动操作就会变成一个评估案例。

And and, okay, like, this this manual interaction will become an eval case.

我们在Brex有个费用处理机器人。

We have an expense agent in Brex.

只要有人和机器人对话时，发现有问题、漏洞，或者感觉对话不顺畅，系统就会自动生成一个漏洞报告。

Whenever someone has a conversation with the agent that is the flags, an issue or a bug or something that feels like the conversation didn't go as smoothly, that creates a bug.

这个漏洞会触发另一个机器人，去修改代码、提示词等所有相关部分，确保这个评估能通过。

That bug triggers an agent that's gonna go and modify the code base and the prompts and everything to make that eval pass.

如果机器人自己解决不了，工程师就会介入，找出让它通过的方法。

And if that doesn't break, then an engineer is gonna go in and figure out how to make that pass.

因为最终的目标，我觉得，是让整个系统实现自我学习。

Because the goal at the end, I think, is to make the whole thing a self a self learning system.

对吧？

Right?

我看到很多公司花大量时间把一个代理跑起来，却从没想过怎么让它每天都能进步。

And I think the a lot of what I see with with companies is they spend a lot of time getting an agent working, but never thinking how to make the agent improve every day.

我觉得这才是最关键的突破点。

And I think that's, like, always the biggest unlock

理想循环。

dream cycle.

你需要一个理想循环，明白吗？

You need a dream cycle see

每晚都这样。

everything every night.

没错。

Exactly.

然后你就想，那边到底发生什么了？

And then it's like, oh, what's going on there?

我得把这个放这儿。

I need to put this over here.

没错。

Exactly.

那

What

到底发生了什么？

actually happened?

有没有什么规律？

Is there a pattern?

我该怎么去发现呢？

How do I Mhmm.

是什么导致了这种情况？

Cause this?

那怎么才能把梦境循环融入到产品、智能体，还有实际发布的东西里呢？

So how to how to how to bake the dream cycle into the products and into the agents and into things that ship?

我最近最感兴趣的是，正在给几个朋友开发三到四个智能体。

My favorite thing right now is I'm building like three or four agents for my friends.

哦，这挺有意思的。

Oh, interesting.

其实这些也算是我为G Brain做的用户调研，因为我自己用了一个，效果特别好。

And I mean, some of it is like, this is a user research for me for g brain, because it's like, I have one, it's working really well.

现在里面已经有三十五万篇Markdown文档了。

I have 350,000 markdown pages in there now.

简直难以置信，我当初还以为这不过是天马行空的幻想，没想到真会在我们有生之年实现。

What a crazy like, I thought it was this wild, you know, pie in the sky thing, and it's like, it's hap it's gonna happen in our lifetimes.

我记得Neuralink刚发布的时候，我老在想这事。

You know, I remember when when Neuralink came out, I used to think about it.

我当时根本搞不懂，这到底有什么意义。

Was like, I I don't get it.

我当时就想，嗯，概念上我明白，但这到底有什么特别的呢？

Like, was like, yeah, of course, I get it conceptually, but but why is it a thing?

可现在我开始用AI了，而你呢

And then now I use AI and you're

没错。

like, yeah.

对。

Yeah.

说得通。

Makes sense.

说得通。

Makes sense.

是的。

Yeah.

我成了瓶颈。

I'm the bottleneck.

是的。

Yeah.

对。

Yeah.

打字太慢了。

Typing is so slow.

我不知道你平时用很多自适应功能吗？

I don't know if you use a lot of adaptation.

我用得很多。

I I I use a lot.

我现在最常用的开发工具其实是用语音备忘录发给OpenClaw。

My most used developer developer UI right now is like voice memos to OpenClaw.

我以前见过这种做法。

I've seen this before.

说起来可能有点偶然，但我特别喜欢Telegram和

Like, it it was maybe accidental, but I actually just really love the fact that like Telegram works so well with

通过语音。

With audio.

用OpenClaw，因为它逼着我往里加更多内容，让智能体变得更聪明，这样就能通过语音备忘录完成更多事情。

OpenClaw, because it's forced me to just put more stuff, like make the agent more intelligent so that you can do more stuff via voice memos.

因为你得克服传统开发者那种本能反应——总觉得‘这个做不了’或者‘它不支持这个’，所以得去开发更多客户端、界面或者功能。

Because you have to sort of fight the natural instinct as like a traditional developer where you're like, oh, can't quite do this or it doesn't do this, so I need to go build more client or UI or more functionality for it.

富士康。

Foxconn.

没错。

Yeah.

就是这样。

Exactly.

不是。

No.

就让它做它想做的事吧。

Just let it do what it wants to do.

给它一些背景信息，它就会自己琢磨：诶，那这个怎么样？

Give it some context, and it'll just think about, oh, actually, what about this?

我觉得，正如你所说，很多工作在于怎么为模型组织好上下文。

I think a lot of the work, to your point, is the how do you organize the context for the model?

而且你可以用模型来帮忙，但真正的瓶颈就在于怎么组织这些上下文。

And and and you can use the model to help, but but that is that is that is the that is the bottleneck for for most things.

一旦把上下文搭好了，其实你能让它干出不少惊人的事情。

Once you have the context in there, it's actually you can do some pretty crazy stuff.

比如我最喜欢的新功能

Like, my favorite new feature

我看到你的大脑像吃了迷幻药一样。

I saw your brain LSD.

是的。

Yeah.

迷幻药。

LSD.

对。

Yeah.

横向突触漂移。

Lateral synaptic drift.

所以你是直接调高搜索的温度参数？

So you you just bump the temperature on the search?

不只是这样。

It's not just that.

你有这些向量。

So you have the vectors.

对吧？

Right?

是的。

Yeah.

如果你想想大家通常说的那些想法，比如两个想法放在一起，就像在一个圆锥体里那样。

And so, you know, if you think about what conventional ideas are, like, most people give you, like, oh, well, this idea with this idea, and it's, like, kinda like in this cone.

对

Mhmm.

LSD模式实际上要求你不能把那些彼此关联的概念组合在一起，它们必须是正交的，或者看起来完全随机。

LSD mode actually says you cannot combine concepts that are within the They actually must be orthogonal or just, like, feeling seemingly random.

然后它会随机尝试成百上千种这样的组合。

And then it'll try, like, you know, randomly hundreds of these combinations.

是的

Yeah.

接着它会把这些组合按连贯性排序，选出最合理的那些。

And then it'll rank order them into the ones that are actually the most coherent.

是的

Yeah.

如果你做一百次，通常前五个结果都会是特别炸裂的推文。

And then if you do like a 100 of them, actually, like the top five tend to be banger tweets.

你有没有觉得奇怪，他根本没跟Alfred明确说过要干练一点。

You know what's crazy is like, didn't tell Alfred explicitly to be dry.

我让ChatGPT帮我生成了一个灵魂文件。

I went I actually had like chat GPT generate like the soul file.

它就根据你对我的所有了解，还有我们在这里的全部互动，给我生成了一个属于我OpenClaw代理的灵魂MD文件。

It was just like, based on like everything you know about me, all the interactions here, like generate like a soul MD for like my OpenClaw agent.

结果出奇地准确，完全贴合我的风格。

And it was so unerringly, like, accurate.

这正是我期望一个代理该有的样子。

But, like, kinda what I would want from, like, an agent.

我当时就忍不住说：天啊。

And I was like, oh, damn.

这些模型对我们了解得太深了。

These models know a lot about us.

我给我的OpenClaw导入了60G的谷歌数据导出后，它突然变得特别有意思。

My OpenClaw got really interesting once I I I just ingested my 60 gig Google takeout.

不过你得写一堆Haiku代码，才能把那些真正重要的邮件筛选出来。

I mean, you have to write a bunch of Haiku code to, like, only get the the emails that are actually real.

但你知道吗，从60个G里头，它真能挑出大约4000封真正重要的邮件。

But, you know, there's like it it extract, like, 4,000 emails out of 60 gigs that actually matter.

而这些邮件里，其实藏着你很多思考，还有人生中那些关键的时刻。

But like those are like, oh, actually, like a lot of your thinking and, you know, the consequential moments of your life.

所以，佩德罗，非常感谢你今天来和我们交流。

So Pedro, thank you so much for being with us.

你绝对是那些最懂AI、走在最前沿，同时又特别务实的CEO之一，亲自上手玩这套东西，还真的把它做出来了。

I mean, you're by far one of the most AI pilled, farthest out on the edge, but also very practical CEOs who is, you know, playing with this stuff and actually building it yourself.

对那些正在看直播、想创业的朋友们，你有什么建议吗？

What would you say to people watching who are founders, who who wanna be founders?

我觉得你就是个活生生的例子，说明了人该怎么用AI当搭档，来创办和经营公司。

You know, I I think that you are sort of the model for the way people should start companies and run them with AI as your SLN buddy.

我一直在想那个电力的比喻，就是说，人类历史有两百年，其中有个时间点，电力被发明了。

I I really can't stop thinking about the electricity analogy, which is, you know, you're standing there's a two hundred year timeline of human history.

就在某个时刻，电力出现了。

There's a point in time where electricity was invented.

刚开始的时候很糟糕。

It's sucked in the beginning.

你处在那个时间点之后的六个月。

You're six months after that point.

你会有什么不一样的做法？

What do you do differently?

如果你早就知道电力未来会怎样发展，知道数据中心会消耗多少电，连AI也不例外。

Knowing everything that will be true about electricity, knowing that data centers when they will consume electricity and and even AI.

对吧？

Right?

我觉得你会做很多不一样的事情。

Well, you do a lot of things differently, I think.

我觉得这让人特别惊叹，我们此刻正站在这样一个关键的时间点上。

So I think that's one of just just marveling at the possibility of the exact moment in time we're now.

第二点是，可以在电脑上贴个便利贴，每天早上醒来，不管遇到什么问题，都问问自己：能不能用AI来解决？

I think the second is, like, have a Post it on your computer, which is you wake up, whatever problem you have in your life, why can't you solve it with AI?

那就从这儿开始吧。

And just, like, start there.

80%的情况你用聊天机器人就行，但那20%做不到的，得弄清楚为啥，然后自己动手做个工具来解决。

And and 80%, yeah, you can use a chatbot, but the 20% that you can't, figure out why and go build something that makes you solve that problem.

不是因为解决这些问题能立刻带来多大用处，而是因为只有亲身体验，才能真正感受到这项技术的潜力——天天不碰，根本体会不到。

Less so because of the because of the immediate usefulness that solving that thing at scale will have, because it gives you a texture and a feel for the possibilities of the technology, which are really hard if you're not playing with it every day.

第三个建议，可能是去追踪你的令牌使用量，看看自己到底在多大程度上推着公司往前走，从一个前提开始：好吧。

And and maybe the third thing is, like, I I think it's, like, just measure your token consumption and how much you're just pushing the the the limits of the company and starting at the premise of, okay.

为什么不能就一个人搞定？

Why can't it just be one person?

为什么不能就我一个人把整个东西做出来？

Like, why can't it just be me that that builds the whole thing?

你迟早会碰到一道墙，就是那些模型能做什么、不能做什么的界限。

And you're gonna probably face a wall of the, the elements of, you know, what what models can and cannot do.

但归根结底，我觉得关键问题是：作为创始人，你该怎么把时间花在只有你能做的事上？

But at a limit, I think the question is, you know, how do you spend your time on the things that only you can do as a founder?

对我来说，这些事第一是哪些问题值得解决，第二就是我们之前聊过的那个选择问题。

And these things to me are, number one, which problems are worth solving, and two, and the sort of the choice thing we talked about.

第二点是，既然有了这些选择，那大语言模型目前还做不到哪些事？

And the second thing is, okay, given that the given these choices, what are the limitations of an LLM that they still cannot do?

这些事我得亲自上手去做。

And I have to go in and do that those those things myself.

但某种程度上，你其实是在为大语言模型打工。

But almost, you know, to some degree, you're working for the LLM to some point.

如果你在大公司里，你几乎是在把大语言模型当创始人和CEO来用，整个公司都在围绕这个想法来设计。

And and and and if you're in a bigger company, you're in a turnaround to put the LLM as almost a founder and a CEO, and you're you're you're you're almost architecting the entire company around that idea.

但早期阶段，最重要的其实是选对方向、跟客户沟通、注入模型没有的洞察，就像在2026年，电力才刚出现六个月时，你得重新思考一切该怎么做。

But I think early on, so much of it is, you know, choosing what matters, talking to customers, injecting the signal that the models don't have, and just, you know, regolding it the way you would do it in 2026 with electricity being six months old.

谢谢，佩德罗。

Thanks, Pedro.

这太棒了。

This is awesome.

是的

Yeah.

谢谢您邀请我。

Thanks for having me.

非常感谢。

Appreciate it.

马上来。

Coming.