Building the Future: Paul Graham & Amjad Masad on Vibe Coding and AI Agents
打开互动全文版(中英对照 + 朗读 + 问答)→Amjad Masad 与 Paul Graham 对谈,回顾 Replit 如何把用户手动复制粘贴 ChatGPT 代码的流程,变成能自主编写、调试并部署代码的 AI 智能体。
Amjad Masad joins Paul Graham to trace how Replit turned manual ChatGPT copy-paste loops into autonomous AI agents that write, debug, and deploy code.
我认为在 AI 这件事上,每个人都要经历否认期,然后是沮丧期,最终才会接受。Replit 就像一座冰山,水面上可见的部分是:你只要用英语告诉它要做什么,它就会帮你写代码;而水面下还有一大堆用于部署和沙箱的东西。
I think with AI, everyone will have to go through this period of denial and then depression and then eventually acceptance. Replit's like this iceberg where there's this visible bit above the water that's like, look, you just tell it in English what to do and it writes your code, and then underneath there's all this stuff for deployment and sandboxing.
我不认为我们会把自己描述为 vibe coding。现在我们认为 Replit 是通往自动驾驶式公司的转型平台。
I don't think we describe ourselves as vibe coding. Now we think Replit is the platform for the transformation to self-driving companies.
不过,比起其他公司,我觉得你们一直生活在未来。如果非要说的话,Replit 真正的 DNA 是什么?
More than other companies though, I think you guys have been living in the future. If you had to say, what's the real DNA of Replit?
那肯定是生活在未来。
It is living in the future for sure.
所以你们基本上发明了 vibe coding,对吧?你们在 2024 年做这件事的时候——听起来才两年前,但感觉已经过去很久了。
So you guys basically invented vibe coding, right? And when you did it, like in 2024, this was—which sounds like only two years ago, but it seems so long ago.
从技术上来说,感觉像是……
Seems like a tech—technologically.
你怎么知道这能行得通?因为现在很明显它成功了,对吧?但当时看起来挺疯狂的。至少对我来说很疯狂。
How did you even know this was going to work? Because it's obvious now that it did work, right? But at the time it seemed crazy. It seemed crazy to me.
嗯,对我们来说是渐进的。并不像——对其他人来说,这感觉是对做事方式的巨大背离,但我们一直——
Well, for us it was incremental. It wasn't like—for the rest of the world it felt like a big departure from how people did things, but we've always—
你们已经在写越来越多的代码了。
You were already writing more and more of the code.
嗯,不是,这更像是 YC 的理念:观察你的用户在做什么,这会给你灵感去构建什么。
Well, no, it was sort of like the YC ethos of looking at what your users are doing, and that will give you ideas on what to build.
因为你们已经在做自动补全了。你们在做自动补全,对吧?
Because you were already doing autocomplete. You're doing autocomplete, no?
是的,我们在做自动补全。
We're doing autocomplete, yes.
那么,在你们转向完全的 vibe coding——也就是用英语告诉它要做什么——之前,你们用自动补全生成了多少代码?当时你们已经生成了多少比例的代码?
So how much of the code were you writing before you went to full vibe coding where you tell the thing in English what to do, versus just generating code with autocomplete? How much of the code were you already generating then?
我的意思是,大概可以说 50% 到 60% 吧。
I mean, you could probably say like 50, 60%.
真的吗?这很有意思。甚至在 vibe coding 之前,你们就已经有 60% 是 vibe coding 了,对吧?这就是你们知道它能行得通的原因。
Really. So that's interesting. Even before vibe coding you were already 60% vibe coding, right? And that was how you knew it would work.
你觉得既然能做到 60%,为什么不做剩下的部分?嗯,差不多吧。我觉得真正有启发性的是观察我们的一些用户写代码。有些用户 100% 的代码都是用 AI 写的。但他们做的方式——
You thought we can make the 60%, why not make the rest? Well, kind of. I think the thing that was very enlightening is watching some of our users code. And so some users were doing 100% of their code using AI. But the way they did—
不,这是——你是在说你们发布 vibe coding 之后的事。
No, this is—you're talking about once you released vibe coding.
不,在那之前。在那之前。我告诉你他们是怎么做的。
No, before that. Before that. I'll tell you how they do this.
所以他们会在一个窗口打开聊天,在另一个窗口打开 Replit。这是一个从未写过代码的人。
So they would open chat in one window. They would open Replit in another window. This is someone who's never coded before.
你怎么知道他们是从 ChatGPT 里粘贴代码的?
How could you tell that they were pasting code out of ChatGPT?
我邀请了一些用户来办公室。
I invited some users to the office.
嗯,你 literally 看着他们操作。
Well, you would literally watch them.
我们看着他们。
And we watched them.
好的。那你是怎么做的?
Okay. How do you do this?
基本上他们会复制代码,粘贴到 Replit 里。Replit 会产生一些错误。他们会复制错误,粘贴到 ChatGPT 里。
And basically they would like copy the code, put into Replit. Replit would like produce some error. They would like copy the error, put it into ChatGPT.
嗯。
Yeah.
然后复制修复方案,粘贴进去。整个过程非常手动。我看着它——
And then copy the fix, put it in. And it was such a manual process. And I looked at it—
我们只需要做一个 API,对吧?
All we have to do is make an API, right?
嗯,我们必须把整个流程自动化。这并不明显,也不容易,因为那时候 AI 模型只能做几次工具调用。
Well, we have to automate the entire process. It wasn't obvious. It wasn't easy because back then, AI models would like do a couple of tool calls.
哦,你知道不同之处在哪吗?不同之处在于,你不仅要自动化来回的交互,还要理解错误。
Oh, you know what's different? The difference is it's not just you have to automate the back and forth. You have to understand the errors.
嗯。
Mhm.
对。
Right.
没错。
Exactly.
是啊,这很难。
Yeah. That's hard.
是的。而且你必须循环,终止条件应该是程序能正常工作,而不是——
Yeah. And you have to loop and the terminal condition should be that the program is working as opposed to—
是的,现在仍然如此,对吧?而不是像我们只做一轮,对吧?因为 ChatGPT 差不多就是这样工作的。大多数 AI 工具都是这样。那么你能无限循环这个循环吗?
Yeah. That's still true, right? As opposed to kind of like we'll do one turn, right? Because that's how ChatGPT sort of worked. That's how most AI tools worked. And so can you loop this loop indefinitely?
直到完成。
Go to completion.
是的。
Yes.
对。这就是新东西。Claude 就是这么做的。对。
Right. Which is the new thing. That's what Claude does. Right.
是的。这就是我们展示的。这就是 Replit 首次展示的。所以我们展示了几件事。一是你可以让 LLM 运行到完成。
Yeah. That's what we demonstrated. That's what Replit first demonstrated. So we demonstrated a couple things. One is that you can run LLMs to completion.
到——
To—
我敢打赌这对模型公司来说是个新闻。
Which I bet was news to the model companies.
确实。他们实际上联系了我们,说:“我们很惊讶 Replit 能行得通。你们是怎么做到的?”而疯狂的是,Paul,2024 年左右的 harness 比今天的 harness 复杂 10 倍。
They did. They actually reached out to us and they were like, "We're so surprised that Replit works. How did you make it work?" And the crazy thing, Paul, is that the harness circa 2024 is 10 times more complex than the harness today.
因为模型能力较弱。
Because the models were less capable.
是的。
Yeah.
所以,随着模型能力越来越强,让它们成为智能体所需的工程工作越来越少。那么这是否意味着人们更容易与你们竞争?
So, as models get more capable, like what kind of engineering you have to do to make them become agents is less and less. So does that mean it's easier for people to compete with you?
是的。但我们必须做——嗯,这就是当今 AI 业务的特点:你必须接受你所做的一切都在不断被商品化,你必须做下一件事。
Yes. But we have to do—well, that's the thing about the AI business today is that you have to just accept that everything that you're doing is getting commoditized constantly and you have to be doing the next thing.
做下一件事,对。但我们展示的另一个重要事情是,AI 可以配置基础设施。所以在那之前——
Doing the next thing, right. But the other big thing that we demonstrated is that AI can provision infrastructure. So up until that point—
而且你确实有基础设施可以配置。
And you actually have infrastructure to provision.
是的。这就是你和所有这些竞争对手的不同之处。对。
Yes. That's the thing that's different between you and all these competitors. Right.
没错。所以我们构建了像数据库这样的东西。
Exactly. So we had built like database.
你花了这么多年受苦做系统软件。对。对。
You spent all these years suffering doing system software. Right. Right.
对。所以这就像字面意义上的工具调用:启动一个数据库,你知道,启动一个 Redis,托管这个,等等。所以当我们发布时,人们感到非常惊讶——我是说你 2024 年 7 月看到了产品。
Right. So it was a literal like tool call spin up a database, you know, spin up a Redis, like host this, you know. And so what people found so surprising when we launched—I mean you saw the product July 2024.
你们什么时候发布的?24 年 9 月。
And when did you launch it? September '24.
好的。
Okay.
所以当人们看到它时,就像 Andrej Karpathy 说的,这就像一个 AGI 时刻,一个完整的 AGI 时刻,因为每个人都在想:“天哪,LLM 能做到这个。”这让我感到害怕。
And so when people saw it, like Andrew Karpathy said this was like an AGI moment, like a full AGI moment, because everyone was like, "Holy crap, like LLMs could do that." It freaked me out.
是的。
Yeah.
我一点也不喜欢。
I didn't like it at all.
你现在感觉如何?因为从美学上讲,这对你来说非常冒犯。
How do you feel about it right now? Because aesthetically it was very offensive for you.
是的,我知道。我知道。因为我编程的方式与此完全相反。我喜欢——我编程,你知道,我花三年写一千行代码,如果下个月能降到 980 行,那就更好了。
Yeah, I know. I know. Because the way I program is the exact opposite of this. I like—I program, you know, I spend three years writing a thousand lines of code and if I can get it down to 980 in the next month, so much the better.
但你给了我这一切的原则,那就是当你第一次投资 Replit 时,当你第一次支持 Replit 时,你说编程的容易程度与你能获得的程序员数量之间存在超线性关系。
But you give me the principle for all of this, which is when you first invested in Replit, when you first backed Replit, you said there's a super linear relationship between how easy it is to program and how many programmers you get.
是的,没错。这不仅适用于编程,也适用于任何消费者使用的东西。如果你让它容易 10%,你不仅会多 10% 的人,你会得到两倍的人。
Yes, that's true. That's not just true for programming. That's true for anything consumers use. If you make it 10% easier, you don't just get 10% more people, you get twice as many.
是的。
Yes.
是的,我记得。我不是说让它这么容易。
Yeah, I remember that. I didn't mean make it this easy.
就像你走得太远了。
It's like you went too far.
是的,我知道。我的意思是,只要用英语说,它就会吐出所有这些代码。
Yeah, I know. I mean, just talk in English and it spits out all this code.
嗯,所以我意识到,要让 Replit 真正成功。
Well, so the realization I had is for Replit to be truly successful.
我们得把整个故事讲完,因为你确实是早期预测 AI 的人之一,对吧?你当时是不是在 2015 或 2017 年左右就预测了 vibe coding?
We needed to go the whole way because you did know you were one of the early people predicting AI, right? And like were you predicting you were essentially predicting vibe coding in something like 2015 or 2017 or something?
是的,2015、2016 年在我们最初的演示文稿里,我们谈到如果你建立一个平台,有数百万人在上面编程,你就可以训练一个机器学习模型来进行预测。
Yeah, 2015, 2016 in our original deck we talked about how if you build a platform where there's like millions of people coding you can like train a machine learning model that predicts.
我是说,这超前得离谱。
I mean that's crazy ahead of its time.
是的。
Yeah.
对。这还是在 Transformer 论文发表之前。
Right. This was before the Transformers paper.
那是在 Transformer 之前。
That was before Transformers.
我开始觉得那不是有远见,那只是你疯了。如果你那么早就想到的话。
I'm starting to think that's not the visionary that's just your nuts. if you're that early.
嗯,我觉得我一直挺擅长预测未来的走向。你知道,甚至在我小时候,我记得我第一次用诺基亚手机时,我就想:“天哪,人们以后就会用这个,这就像新的电脑。”那其实基本上就是个翻盖手机。那种经历也某种程度上带到了 Replit,Replit 是第一个有移动应用的编程平台,大家都笑话我们。那个想法就是
Well, I think I've been fairly good at like predicting where the future is headed. Well, you know, even when I was a kid, like I remember when I used my first Nokia phone, uh I was like, "Oh my god, people are just going to be this is like the new computer and it was like literally like a flip phone essentially." And some of that experience kind of also um came to Replit which is Replit was the first programming platform that had a mobile app and everyone laughed at us. Uh the idea that
在移动端编程
programming on mobile
在手机上编程
mobile programming on your phone
但我为此写了一份 RFC。
but I had an RFC for it.
是的。
Yes.
不是 RFC,是创业公司的 RFS。
Not an RFC, an RFS for startup.
你实际上告诉过我,你会用手机上的 Hacker News 列表道具来修 bug。
You actually you told me you would like fix bugs using the like hacker news uh like list prop on your phone.
没错。是的,不是自愿的。那只是我们唯一能上网的方式。
That's right. Yeah. not voluntarily. It was just the only way we could get internet access.
但没错,既然这个播客是关于构建未来的,我觉得人们低估了自己预测未来走向并朝着它构建的能力。你只需要坐下来思考。我不认为很多人会坐下来思考。
But yes, since this like podcast is about like building the future, I think that people underestimate that the their ability to potentially predict the where the future is headed and like build build towards it. You just have to like think like sit down and think. I don't think many people like sit down and think.
不,这就是专家们想的。专家们以为我所做的就是坐下来思考,然后就有了想法,对吧?你知道,其他人也试着坐下来思考,但他们得不到秘诀,因为这是因为你在这个想法上已经琢磨了 20 年。你第一次开始想做一个像 Replit 那样可以在线编程的东西是什么时候?
No, this is what experts think. Experts think all I do is I just sit down and think and I have the ideas, right? you know, and like everybody else tries sitting in sitting down and thinking and they don't have the secret because it's it's because you've been doing this idea for like 20 years. When did you first start thinking about making something like Replit where you could sort of program online?
嗯,就像我当时有塞班智能手机,我在上课,我就想,你知道,我希望我能编程。
Well, it's like when I was like I had my Symbian smartphone and I was in class and I was like, you know, I wish I could program.
那是多久以前的事?
So, how long ago was this?
大概是 20,嗯,2008 年。
This was like 20 um 2008.
好。所以差不多 20 年前。这就是为什么当你坐下来思考时,你能想出未来,对吧?因为你的整个大脑已经思考这件事 20 年了。
Okay. So almost 20 years ago. This is why when you just sit down and think, you come up with the future, right? Because your whole brain has been thinking about this for 20 years.
对。
Right.
对。痴迷地。是的。
Right. Obsessively. Yes.
你一直在尝试,在它叫 Replit 之前你就试图构建它,你知道,年复一年。对。这一直是件大事,
You've been trying to you were trying to build Replit before it was called Replit for like, you know, years and years and years. Right. This has always been the big thing,
对吧?
right?
所以你当然比其他人先想到这些东西。
So of course you thought of stuff before everybody else.
那么,回到那个著名的演示,对吧,2024 年夏天
So, you knew back to that famous demo, right, in the summer of 2024
当时你叫我邦德反派。
when you called me a Bond villain.
嗯,你看起来就像邦德反派。你知道,
Well, you seemed like a Bond villain. You know,
那就是我的感觉,我被展示了一个可怕的东西,因为我记得,我仍然记得你说的话。别看那个。
that's what I felt like that I was being shown this frightening thing cuz I remember I still remember what you said. Don't look at that.
别看代码。
Don't look at the code.
那是目标代码,
That's object code,
对吧?我就想,“哦,不。我做的所有东西,我花了几十年做的所有东西,都只是目标代码。”
right? And I'm like, "Oh, no. The stuff I the whole stuff that I make, the whole stuff I've spent decades making is just object code."
现在大家都意识到这一点了。嗯,我觉得,是的,我见过一些工程师也对那感到反感。我记得你的推文很有争议,当他发那条推文时。
And now everyone's realizing that. Uh like I think uh yeah, I've been seeing like like engineers that were also repulsed by that. I remember your tweet was very controversial when he tweeted that that thing.
那个用 AI 生成了大量代码的人。是的。
That guy who who generated vast amounts of code using AI. Yeah.
是的。嗯,甚至那条关于我是邦德反派、关于目标代码的推文也很有争议。但现在所有工程师都,你知道,我认为有了 AI,每个人都必须经历这个阶段,你知道,否认,然后,然后后来,你知道,沮丧,然后最终接受。
Yeah. Well, even that tweet about me being the bond villain on object code and all of that was also controversial. But now all engineers are you know I think with AI everyone will have to go through this period of you know denial and then and then sort of uh later on you know depression and then you know eventually acceptance.
是的。我在 2020 年左右就经历过了,你知道,真的
Yeah. I've had it like in like 2020, you know, really
所以,所以这对我来说更容易,因为我经历过那种悲伤,我所热爱的技艺消失了。我已经哀悼过了
and so and so it just makes it easier for me like I had the grief of the craft that I love is gone. I've already grieved
你也有那种感觉。
you felt that too.
是的。是的。是的。是的。
Yeah. Yeah. Yeah. Yeah.
所以你知道我当时的感觉。
So you knew what I was feeling.
我知道你的感受
I knew what you were feeling
当你给我看那个演示时。
when you showed me that demo.
是的。嗯,但你必须感受它才能放下。我觉得去年年底大家都被说服了。2025 年底是每个人都意识到的时候,好吧,AI 将编写所有代码。
Yes. Uh but you had to feel it so that you could let go. I think everybody got convinced at the end of last year. At the end of 2025 was when everybody realized, okay, AI is going to write all the code.
总是让我惊讶的是这类事情的触发点是什么,因为对我来说这似乎很明显,而且嗯
It's always surprising to me what is the trigger for these kind of things because again it's it seemed obvious to me and um
嗯,对于 Replit 用户来说
well for replet users
我想他们早就知道了
I think they already knew
对我们来说,说服世界其他地方的东西可能是 Claude,对吧,你不觉得吗?
for us there's a thing that convinced the rest of the world might have been cla right don't you think?
我觉得是 OpenClaw。
I think openclaw.
好吧。我觉得 Claude Code 部分原因,但 OpenClaw 我觉得是人们第一次体验到这种想法:语言模型运行到完成,你给它一个任务,它能够一直做下去直到它
Okay. I think Claude Code partly but open claw I think was the first time people experienced this idea of LMS running to completion where you give it a task and it's able to like go all the way until it's
那时你已经做了一年半了
which you had been doing for a year and a half at that point
但我认为 OpenClaw 的形态也是,比如你可以给它发短信之类的,我觉得我有点低估了形态的力量,尤其是对普通人来说。它影响了你对 Replit 的做法吗?
but I think the form factor of open claw as well being like you could text it and and all of that I think I I sort of underestimate the power of like a form factor especially for like everyday people. Did it influence what you've done with Replit?
你从中学到了什么吗?
Did you learn anything from it?
还没有。我只是学到了那个教训。是的。我觉得它会影响未来的事情,就像我假设我出去提出一个愿景,构建一个产品,大家都会说,“哦,是的,那很明显。”但我觉得现实中,你需要大量调整人们体验它的方式。
Not yet. I've just learned that lesson. Yeah. I I think it will influence things in the future, which is like I just assume that I go out and like put out a vision and like build a product and everyone will be like, "Oh, yeah, that's that's obvious." But I think in reality, I think you need to tinker a lot with the with how people experience it.
你还会观察用户使用 Replit 吗?看看他们在做什么,并对此感到惊讶吗?
Do you still watch users using Replit and and sort of see what they're doing and are surprised by it?
嗯,是的。其实我不喜欢那样做,但 Haya 早期强迫我去做,我会又踢又叫,因为当用户对软件感到困惑或遇到 bug 时,我会感到强烈的痛苦。是的,
Um, yes. I actually didn't like to do that, but Haya kind of forced me to do it early on and I would like be kicking and screaming because I would like feel like intense pain when like users are confused by the software or hit a bug or whatever. Yeah,
我会犹豫。就像我真的会,你知道,身体上。
I would be like hesitant. Like I would actually be be, you know, physically.
看用户使用你的东西是令人难堪的。真的。但你知道为什么难堪吗?因为直到现在你都在否认每天在外面发生的恐怖事情,
It's mortifying to watch users use your stuff. It really is. But you know why it's mortifying? Cuz till now you were in denial about the horrors that were being perpetrated daily out there,
对吧?
right?
所以,但是,但是我们的公司文化变得非常注重贴近用户。嗯,主要感谢 Haya 有那种
And and so but but but our our company culture became a lot about being very close to our users. uh primarily thanks to ya kind of having that
那种心态,你需要不断和他们交流。
that mentality of you just need to constantly talk to them.
她还让你看起来很酷。
She also made you very cool looking.
那是 Replit 早期非常不寻常的一点。令人惊叹、颠覆认知的平面设计,对吧?比一个书呆子式的在线编程公司真正需要的要好得多。
That was one of the really unusual things about Replit early on. Amazing, mind-bending graphic design, right? Like way better than some nerdy program online kind of company really needs to have.
这对我们最初的受众帮助很大,他们很多是青少年黑客。
And it helped a lot with our initial audience, which were a lot of teenage hackers.
我觉得那真的很酷。
I thought it was really cool.
嗯。
Mhm.
它让 Replit 与众不同。
It made Replit different.
是的。
Yes.
你仍然有 Replit,它仍然有非常独特的东西。你知道它有一种非常 Replit 的——像许多非常成功的公司一样,它有一种非常强烈的自身特质。
You still have Replit, still there's something very distinctive about it. You know it has a very Replit—like a lot of very successful companies it has a very strong character of itself.
这听起来没什么意义,但你懂我的意思。这就像一个非常强的溶液——不是化学意义上的弱溶液,Replit 是一个强溶液。
Which sounds meaningless, you know what I mean though. It's like a very strong solution—it's not a weak, in the chemical sense, it's not a weak solution. Replit is a strong solution.
我们对所构建的东西和产品的概念性理解是,我们总是想到引擎和坚硬的机械东西,尽管它并不是。
Our conceptual understanding of what we build and our product is we always think about engines and hard mechanical things, although it is not.
那平面设计很大程度上就是关于这个的,不是吗?
That's a lot of what that graphic design was about, wasn't it?
是的。是的。是的。它就像有一种硬核的东西,但也矛盾的是,我们拼命让用户体验对用户来说如此简单,但在后端却发生着非常丑陋的工程。
Yeah. Yeah. Yeah. It's like there's something that's sort of hardcore about it, but it's also paradoxical in that we slave away so much to make the UX so simple for users, but on the back end just really ugly engineering is happening.
系统编程,真正的系统编程,对吧?那是你们 DNA 的一部分。
Systems programming, like real systems programming, right? That's part of your DNA.
那是我们 DNA 的一部分。但这一直有点挑战性,因为很多程序员会看着 Replit 并因为它看起来很简单而轻视它。
That's part of our DNA. But it's been always sort of challenging because a lot of programmers will look at Replit and sort of dismiss it because it looks very simple.
是的。
Yeah.
但后来甚至那些最终来 Replit 工作的人也会说,天哪,我不知道幕后有这么多东西。
But then even people who end up coming to work at Replit was like, oh my god, I didn't know there's so much stuff behind the scenes.
是的。
Yes.
是的。
Yeah.
你为人们的罪过做系统管理。什么?
You do system administration for people's sins. What?
是的。我的意思是我们做这个是为了他们不必做。
Yeah. I mean we do it so they don't have to do it.
是的。你认为这会是帮助你战胜竞争对手的东西吗?因为那是一个差异——你构建的东西的深度以及你对如何在线部署程序的理解深度。
Yes. Do you think that will be the thing that helps you prevail over competitors? Because that is one difference—the depth of stuff that you've built and the depth of understanding you have of how to deploy programs online.
是的。所以这帮助我们在这一类别中赢得企业客户,因为他们对可靠性、安全性等所有这些东西更加敏感。很难让消费者关心安全性。例如,我们的许多竞争对手在构建不安全的应用程序等方面做了非常鲁莽的事情。
Yeah. So that is helping us win enterprise in our category because they are a lot more sensitive to reliability, security, all of that stuff. It's really hard to get consumers to care about security. For example, a lot of our competitors are like do really reckless things with regards to building apps that are unsecure and so on.
因为如果你有智能体写所有这些代码,谁知道它在做什么?
Because if you have agents writing all this code, who knows what it's doing?
是的。是的。所以我们设置了很多护栏,这使我们的产品有时更慢,有时更贵,但企业真的很喜欢。所以这些新来的机会主义者没有像 Replit 那样的基础设施深度,就像冰山一样,水面上有可见的部分,就像你只要用英语告诉它做什么,对吧,它就会写你的代码,然后下面有所有用于部署和沙箱的东西。我们有内核操作系统工程师,我们像从头构建了自己的文件系统,使其具有事务性和完全可逆性。
Yes. Yeah. And so we put a lot of guard rails which makes our product sometimes slower, sometimes more expensive, but enterprises really like that. And so these newly arrived opportunists who don't have the same sort of depth of infrastructure like Replit's like this iceberg right where there's this visible bit above the water that's like look you just tell it in English what to do right and it writes your code and then underneath there's all this stuff for deployment and sandboxing. We have kernel operating system engineers, we have like we built our own file system from scratch to be like transactional and fully reversible.
我记得在 vibe coding 之前的日子里听过所有这些故事。你所有的更新都是关于这个的。就像我们制作了全新的文件系统,对吧?我希望有人欣赏它。
I remember hearing all these stories back in the days before vibe coding. That's what all the updates from you would be about. Like we made our whole new file system, right? I hope somebody appreciates it.
是的。我的意思是,当你想让编程协作时,你必须发明很多核心原语。
Yeah. I mean, when you want to make coding collaborative like you had to invent a lot of core primitives.
所以这是一个有趣的问题。你在 Replit 第一版中所做的事情,只是在线编程,你让它协作——你为使其协作所做的东西是否也帮助了 vibe coding?
So that's an interesting question. Did the stuff that you did in version one of Replit where it's just online programming and you made it collaborative—did the stuff that you did to make it collaborative also help with vibe coding?
100%。我的意思是,它是多么即插即用,有点超现实。所以当你把一个智能体放入一个 Repl(我们这么称呼它)时,就像你在与智能体协作。
100%. I mean, it's sort of surreal how plug and play it was. So when you put an agent into a Repl, as we call it, it's like you're collaborating with the agent.
是的。所以智能体是另一个用户。
Yes. So the agent is another user.
所以最初的版本基本上是智能体只是同一系统上的另一个用户。你已经构建了其他所有东西,为它准备好了。
So the initial version was basically the agent was just another user on the same system. And you had everything else built, ready for it.
是的。甚至你和智能体可以在同一个项目上工作,它就会正常工作。
Yeah. And even you and the agent could work on the same project and it will just work.
所以难怪你先发现了 vibe coding,因为你的位置完美。你拥有所有其他成分,对吧?
So it's no wonder you discovered vibe coding first 'cause you were perfectly positioned. You had all the other ingredients, right?
是的。我的意思是这个超现实的事情——甚至公司组织的方式也是如此完美。所以我们有处理代码执行的基础设施团队,对吧?这就像沙箱团队之类的。然后我们有部署和云服务团队。这就像你可以启动数据库和部署虚拟机等等。然后我们有构建基础设施和 IDE 之间粘合剂的团队,然后我们有 IDE 团队。那么 AI 适合所有这些的哪里?所以我们最终创建了一个新的智能体团队,我们把它放在中间,所有这些人现在都在为智能体构建。所以基础设施团队正在为智能体暴露工具。IDE 团队不再是为人构建 IDE,现在他们正在为智能体构建编辑工具。
Yeah. I mean this surreal thing—even the way a company was organized was so perfect. So we had the infrastructure team that handles code execution, right? This is like sandboxing team whatever. And then we had the deployment and cloud services team. This is like you can spin up databases and deploy VMs and whatever. And then we had the team that was building the glue between the infrastructure and the IDE, and then we had the IDE team. And so where does AI fit into all of that? So we ended up creating a new agent team and we put it in the middle, and all these people are building now for the agent. So the infrastructure team is exposing tools for the agents. The IDE team instead of building IDE for humans now they're building edit tools for the agents.
真的,智能体甚至需要什么样的 IDE?他们——我以为他们只是生成代码。
Really what sort of IDE do agents even need? Do they—I thought they just generated code.
嗯,最初在 2024 年,编辑代码实际上是一个难题。
Well initially in 2024 editing code was actually a hard problem.
你是说让它回去修复错误?
You mean having it go back and fix mistakes?
是的。所以不,没有原生编辑工具。现在每个基础模型都给你原生编辑工具,但那时我们必须创建自己的编辑工具。当你给它一段文本,在哪里替换——你不能依赖行号,因为它们会幻觉行号。有所有这些不同的东西。在早期你必须让它极其健壮,因为 LLMs 太差了。
Yeah. So no, there was no native edit tool. Now there's the native edit tool every foundation model gives you, but back then we had to create our own edit tool. When you give it a chunk of text, where to replace—you couldn't rely on line numbers because they would hallucinate line numbers. There's all of these different things. You had to make it ridiculously robust in the early days because the LLMs were so bad.
是的。它们也会懒惰。所以它们会写一个函数,但中间变得懒惰,像以后完成它,这是不可执行的。
Yeah. They would also be lazy. So they would write like a function but like get lazy in the middle like to do finish it later which is not executable.
是的。那曾经是——尽管现在也是——那太疯狂了。OpenAI 和 Anthropic 都有同样的 bug。我认为这是因为聊天模型。我认为它们总是希望聊天早点停止,不像产生大量文本。所以那是它的一个产物,所以我们不得不检测并循环。所以甚至编辑文本也是很多工作,现在有像原生编辑工具。
Yeah. That was—although it is now—that was insane. And that was both OpenAI and Anthropic had the same bug. And I think it's because of the chat model. I think they always wanted the chat to kind of halt early like not produce mountains of text. So that was like an artifact of that and so we had to kind of detect that and loop over. So even editing text was a lot of work and now there's like a native edit tool.
不过,有没有类似智能体 IDE 的东西?所以早期有很多研究是关于——
Is there anything like an IDE for agents though? So in the initial days there was a lot of research on—
不,我是说现在智能体的工作方式,好吧,现在智能体只是生成代码,然后就这样——
No, I mean now the way agents work, okay, now the agents just generate the code and that's—
现在你直接给它们一个 shell 就行了。
Now you can just give them a shell.
好吧。
Okay.
你只要给它们一个 shell,它们就会写 shell 命令。它们用 cat 和你知道的基本 shell 命令就能做所有事。
You just give them a shell and they'll just write shell commands. They can do everything with like cat and you know basic shell commands.
它们非常擅长这个。
They're very good at it.
但早期你需要 IDE,因为那时候智能体还很差。
But you needed to have the IDE early on because the agents were bad.
是的。
Yes.
所以那是让你能这么早起步的因素之一。
And so that was one of the things that enabled you to be so early.
是的。
Yes.
你能更早做这件事吗?
Could you have done it any earlier?
嗯,不能。
Um, no.
因为你正好赶在它勉强能用的那个时间点。
Because you got it just at the point where it barely worked at all.
是的。而且它勉强能用。比如我们发布时还是 beta 版,我会说 50% 的——
Yes. And it barely worked. And like when we launched it, it was in beta and I would say 50% of—
我记得我去你家,你给我演示一些东西,它们都跑不起来。
I remember I would go over to your house and you would show me things and they would not work.
是的。说 50% 可能太苛刻了。我会说大概 30% 的时候它跑不起来。
Yeah. Like 50% is maybe too harsh. I would say like 30% of the time it would not work.
好吧。那用户为什么会容忍这个?
Okay. So why would users tolerate that?
因为那是一种如此神奇的体验。而且,大概他们在 ChatGPT 之间来回剪切粘贴的时候,那要更——
Because it was such a magical experience. And also, presumably when they were cutting and pasting stuff back and forth between ChatGPT, that was way more—
那也一样跑不起来,对吧?但如果他们完全不会编程,那么 70% 的时候能编程就是巨大的进步。
That also wouldn't work either, right? But like if they couldn't program at all, then programming 70% of the time was a huge improvement.
嗯,最初开始付给我们很多钱的客户,是那些原本会去外包网站找外包程序员的人。所以现在你要么花大约 10 万美元请某个地方的工程师做一个应用,要么付给 Replit 大约 1000 美元。所以这就像两个数量级的差距,今天依然如此,在请工程师和让 Replit 做之间。
Well, the initial set of customers that started paying us a lot of money were people who would otherwise go on outsourcing sites and get outsourced programmers. So now you either pay like $100,000 to get an app done by an engineer somewhere, or you pay Replit like $1,000. So it was like this two order of magnitude difference, still is today, between getting an engineer and having Replit do it.
是啊。可以说外包公司的存在本身就证明了那里有市场。
Yeah. Arguably the whole existence of the outsourcing shops was evidence of a market there.
是的。没错。
Yes. Right.
因为我以前觉得它们很糟糕。比如谁会想要这个?谁会付钱给这些人,花大笔钱做这种糟糕的软件?
Because I thought of them as just terrible. Like who would ever want this? Who would pay money to these people to make vast amounts of money to make this terrible software?
但相反,我应该想,哇,那是需求的信号。
But instead I should have thought, wow, that's a sign of demand.
是的。是啊。我从没想过——我看着就像——
Yes. Yeah. I never thought— I looked at like—
我们尝试用几种方式满足这个需求。最初当我们平台上都是这些青少年黑客时,我们做了一个叫 Replit Bounties 的东西,我不知道你——
We try to satisfy this demand in a few ways. Initially when we had all these teenage hackers on the platform, we built something called Replit Bounties, I don't know if you—
我记得,我记得 Replit Bounties。
I remember, I remember Replit Bounties.
它就是想解决这个。我当时想,好吧,AI 还没到那一步。
And it was trying to get at that. I was like, okay, AI is not there yet.
它们会成为新的外包公司,就像零散的一堆团队那样。
They were going to become the new outsourcing shops, just like piecemeal as a bunch of teams.
是的,但那里的洞察是,它们由我们当时在 Replit 上的初级 AI 驱动,但这本身就已经是足够的优势,让它们——
Yeah, but the insight there was that they're powered by the rudimentary AI we had on Replit back then, but that in itself was enough of an edge that they have—
做 Replit Bounties 的人已经在用 AI 写东西了。
The people doing the Replit Bounties were already using AI to write the stuff.
是的,没错。它增长得非常快。但后来我看着它想,好吧,如果未来在这里,那这就是个死掉的业务。所以我们关掉了——
Yeah, exactly. And it grew really fast. But then I looked at it and I was like, okay, if this is where the future's at, that's like a dead business. So we shut—
为什么会是死掉的?
Why would it be dead?
因为最终人们会直接向智能体要整个东西,甚至不用通过那些青少年替他们向智能体提问。
Because ultimately people would just ask agents for the entire thing and not even go through the teenage kids asking the agents on their behalf.
是啊。我是说,现在就是这样。比如我们走进大型跨国公司、银行之类的——
Yeah. I mean, that's the case right now. Like we go into big multinational corporations, banks, things like that—
他们有很多花 1000 万美元买的 SaaS 软件,或者旧软件,他们在公司内部找到几个采用了 Replit 的氛围编程者,然后他们就去替换所有这些不同的系统。而且是从内部的东西开始。
And they have a lot of SaaS software they pay $10 million for, or old software, and they find a few vibe coders inside the company that have adopted Replit, and they're going and replacing all these different systems. And it starts with internal stuff.
从内部的东西开始。
It starts with internal stuff.
是的。以前他们会告诉我们,我们其实已经把这个项目规划好了,本来要花 1000 万美元,结果我们在 Replit 上花了 25 万美元。所以这到处都在发生,我认为外包行业现在差不多就是这样。
Yes. And previously they would tell us that we've actually scoped this project out and it would have been $10 million, we spent $250,000 on Replit. So it's happening all over the place, and I think really the outsourcing industry is sort of that at this point.
所以跟你打交道的多半不是技术性很强的公司,比如不专门做技术、不会自称科技公司的公司。
So it's mostly not very technical companies that you're talking to, like companies that don't specialize in tech, wouldn't call themselves tech companies.
我们有很多科技公司客户。我认为那不是我们影响最大的地方,因为我觉得他们自己能找到办法。但银行、制造业或订单公司,他们需要像 Replit 这样的东西,否则他们会搬起石头砸自己的脚。而真正推动它的主要是内部有雄心的人,比如某个负责销售的家伙想为他的销售部门写个工具。所以这在 Replit 内部造成了一点身份认同问题,比如,嗯,企业业务,那不是我们使命的一部分。但我告诉我的团队,那些是创业者。那些是有雄心的、在建设业务的人。这感觉非常核心,正是我们所做的。
We have a lot of tech companies as customers. I think that's not where we have the biggest impact because I think they can fish for their own. But the banks, the manufacturing or ordering companies, they need something like Replit, otherwise they'll shoot themselves in the foot. And really it's driven mostly by people internally who are ambitious, like some guy who's in charge of sales and wants to write some tool for his sales department. So this created a bit of an identity question inside Replit, like, well, the enterprise, that's not part of our mission. But what I tell my team is that those are entrepreneurial people. Those are ambitious people building businesses. This feels like very core to what we do.
那些也是公司里敢闯敢拼的人。
Those are scrappy people within companies too.
没错。没错。而且即使公司不拼,人也拼。他们往往会得到晋升。很多把 Replit 引入组织的人会得到晋升,因为他们产生了巨大影响。
Exactly. Exactly. And even if the companies aren't scrappy, the people are. And they tend to get promoted. A lot of people who bring Replit into their organization will get promoted because they are having great impact.
也因为尤其是现在所有这些公司都在想,我们得用 AI。我一直读到 AI 如何接管世界。我们应该用 AI。我们公司里谁懂 AI,谁在用?而这里就有个人在用 Replit 做东西。是的。他们就会成为他们信任的人。
And also because especially now all these companies are thinking we got to use AI. I keep reading how AI has taken over the world. We ought to be using AI. Who within our company understands anything about AI, who's using it? And here's the person using Replit to make stuff. Yes. They're going to be the ones that they trust.
没错。所以那是它的延续,但有一点,你怎么说,就像苦差事——有苦差事盲区,或者就是苦差事——就像有很多无聊的工作要做到企业级就绪,或者你必须做。
Exactly. So it's a continuation of that, but there's a bit of, what do you call it, like schlep— there's schlep blindness or just schleps— just like there's a lot of boring work to be enterprise ready or that you have to do.
嗯,你有做这类事情的练习,对吧?就像系统编程。它像系统编程吗?
Well, you have practice doing that kind of stuff, right? It's just like systems programming. Is it like systems programming?
是的,有一点。我是说很多也是给管理员和 IT 很多控制权,控制人们能做什么,设置预算,你知道——
Yeah, a little bit. I mean a lot of it is also just giving admins and IT a lot of controls over what people could do and setting budgets and you know—
哦,他们想要有规则。
Oh, they want to have rules.
是的。规则。
Yeah. Rules.
嗯,这有点像对人类做沙盒。
Well, it's sort of like sandboxing for humans.
是的。
Yeah.
这就像是委托-代理问题,对吧?最终用户并不是购买软件的决策者,这就是为什么科技行业历史上企业软件大多很糟糕。但现在我认为人们对自己想用什么有了更多发言权,因此真正优秀的产品在企业中胜出。但那个问题仍然存在:有时 IT 和安全部门会迫使你让产品对用户更不友好。你觉得答案是给他们想要的吗?比如他们想做蠢事,就让他们做蠢事。还是说答案其实是,你暗中代表公司里那些有野心、有闯劲的人,偷偷给他们绕过官僚的办法?
It's like the principal-agent problem, right? The end users are not actually the decision makers for who buys software, which is why most of the history of the tech industry, enterprise software, has sucked. But now I think people have a lot more say in what they want to use, and therefore products that are really good are winning in the enterprise. But that problem still exists where sometimes IT and security will force you to make the product worse for the users. Do you think the answer is give them what they want? Like they want to do stupid things, let them do the stupid things. Or do you think the answer is secretly you're working on behalf of the ambitious, scrappy people in the company and you secretly want to give them a way to get around the bureaucrats.
是啊,这很难。举个例子,拿一个真实的情况来说。我不同意配给 token。公司现在的做法是,每个人每月给 100 美元的 token。
Yeah, that's a tough one. So for example, you take a real thing. I disagree with rationing tokens. So the way companies are kind of approaching this is saying everyone gets, you know, $100 of tokens a month.
他们应该关心的是 ROI。
What they should care about is ROI.
ROI。没错。我们从客户那里看到的是,公司内部也存在 ROI 的平行分布。一个人可能比另一个人产出高 100 倍。
ROI. Exactly. And what we see from our customers is there's a parallel of ROI inside companies. So one person could drive 100x the next person.
我敢说确实如此。绝对有。软件行业一直如此。
I bet there is. Absolutely. There always is in software.
尤其是有了智能体之后。过去是 10 倍,现在是 100 倍、1000 倍。
And especially with agents. It used to be 10x, now it's 100x, a 1000x.
是啊。
Yeah.
你想控制 token 支出。所以你需要帮助公司看到哪里在发生好事。
You want to control token spend. So what you need is to help the companies see where good stuff is happening.
是的。我认为我们可以在这方面做一些发明来帮助实现这一点。所以短期内,我觉得你不得不做他们要求的事。
Yes. And I think there is some invention we could do there to help with that. So in the short term, I think you kind of have to do what they're asking for.
因为那是你进入的方式。
Because that's how you get in.
那是你进入的方式,然后你想发明更好的东西,再给他们指路。
That's how you get in and then you want to invent something better and then show them the way.
所以有更多东西要构建。
So there's more stuff to build.
有更多东西要构建。这可能是非常有价值的东西。这难道不是很有价值吗?因为想想你在做什么:你试图弄清楚人们用 token 做的哪些项目最有价值。
There's more stuff to build. This could be very valuable stuff. Wouldn't this be valuable stuff? Because when you think about what you're doing when you're trying to figure out which of the projects that people are using tokens for are the most valuable.
你正在解决大公司内部最大的问题。这就是大公司之所以如此的原因。他们无法弄清楚个人所做事情的价值。
You're fixing the biggest problem inside big companies. This is why big companies are the way they are. They can't figure out the value of stuff done by individual people.
对。
Right.
所以如果你构建的东西能让他们在这方面哪怕好 10%,你就是在解决大公司的整个根本问题,对吧?
So if you build something that makes them even 10% better at that, you're fixing like the whole fundamental problem with big companies, right?
是的。我认为这远不止是 vibe coding。
Yeah. And I think that's a lot more than just vibe coding.
是的。是的。我不认为我们现在把自己描述为 vibe coding。我们写了一篇博客文章叫“自驾驶公司”。其背后的想法是,Replit——技术的早期采用者,尤其是 AI 的早期采用者——看到我们的工作方式不同了,工作方式发生了根本性转变。
Yeah. Yeah. I don't think we describe ourselves as vibe coding now. We wrote this blog post called the self-driving company. And so the idea behind it is Replit, where early adopters of technology broadly, but definitely AI, are seeing that we're working differently, like there's a fundamental shift in how we work.
你是指 Replit 作为一家公司?Replit 作为一家公司,也就是 Replit 的零号客户。
You mean Replit as a company? Replit as a company, which is customer zero of Replit.
是的。是的。是的。所以也许你应该先自己弄清楚。你如何识别做有价值事情的人?
Yeah. Yeah. Yeah. So maybe you should figure this out yourself first. How do you identify people doing valuable things?
是的。所以 Replit 作为零号客户就是实验室。我们试图弄清楚这一点。所以“自驾驶公司”是关于工作未来、公司组织方式未来的想法,我们认为 Replit 是自驾驶公司转型的平台。这就是 Replit 前进方向的扩展愿景,因为最终我没有真正意识到的是,我一直知道让软件创建更易访问会产生变革性影响。我没有真正意识到它会从根本上改变人们的工作方式,因为似乎我们是程序员。似乎程序员写程序,对吧?公司运作的核心并不包括编程。是的。但也许现在包括了。
Yes. And so Replit as customer zero is the lab. We're trying to figure out that. So the self-driving company is this idea of the future of work, future of how firms are organized, and we think Replit is the platform for the transformation for self-driving companies. And so that's the expanded vision of where Replit is headed because ultimately the thing that I didn't really appreciate and realize is I've always known that making software creation more accessible would have transformational impact. I didn't really appreciate that it would fundamentally change how people work, because it seemed we were programmers. It seems like what programmers write programs, right? It's not like the core of the way companies work doesn't consist of programming. Yes. But maybe now it does.
确实包括。是的。
It does. Yes.
哇。顺便说一句,这是最优秀初创公司会发生的情况的一个例子。最优秀的初创公司,你提出某个想法,你以为是你提出的,你向他们建议,而他们已经做了,对吧?你已经计划好这个大的新转型,因为这也是最优秀初创公司的另一个特点。最优秀的初创公司,它们已经很大了。
Wow. You know, by the way, incidentally, this is an instance of what happens with the very best startups. The very best startups, you come up with something and you think you're coming up with some idea. You suggested to them and they've already done it, right? You've already got this big new transformation planned because that's also another thing about the best startups. The best startups, they're already big.
嗯。然后他们把这个新事物当作一家全新的公司来想,对吧?然后他们变得大 100 倍,你知道?
Mhm. And then they think of this new thing as if they were a brand new company, right? And like they get 100x bigger, you know?
嗯,我认为现在这样做实际上是必要的,因为 AI,就像我们之前谈到的,是一股巨大的商品化力量。所以我认为过去自我重塑是可选的,现在几乎是强制性的。
Well, I think it's now actually essential to do that because AI, like we talked about earlier, is a massive commoditization force. So I think where it used to be optional to reinvent yourself, it's now mandatory almost.
而你是 AI 流入公司的方式。
And you're the way AI flows into companies.
是的。
Yes.
我是说,这太巨大了。
I mean that's gigantic.
太巨大了。
That's gigantic.
是的。
Yeah.
你只需要让这句话不模糊,因为任何人都能这么说。你知道,任何听这个节目的人,可能 IBM 的某个人在想,‘是啊,听起来不错。’我们会说我们是 AI 流入公司的方式。
You just have to make that be not vague because anybody can say that. You know, anybody listening to this, there's probably someone at IBM thinking, 'Yeah, that sounds good.' We'll say we're the way AI flows into companies.
我们推广的一切最终都被复制了。比如你推广了“从想法到应用快速实现”这个理念。几乎每家初创公司都像布丁一样,那就是他们的特色。
Everything we've kind of popularized ended up being copied. Like you popularize this idea of idea to app fast. It's like literally every startup that does like pudding as like that's their thing.
但他们在复制旧版本的你。
But they're copying the old version of you.
是的。你是一部电影,他们在复制 6 个月前的这个画面,对吧?
Yes. You're a movie and they're copying this frame from like 6 months ago, right?
是的。事情就该如此。你需要不断这样做。但这意味着每家公司都有 Replit。
Yeah. And that's how it needs to be. You need to be constantly doing that. But this means Replits in every company.
嗯。每家拥有竞争对手、不能坐以待毙的公司。
Mhm. Every company that has competitors that can't afford to just sit.
是的。这就是目标。
Yes. That's the goal.
这太惊人了。
That's astounding.
是的。
Yeah.
那么你怎么做呢?我想答案是逐步进行。
So how do you do this? I suppose gradually is the answer.
是的。我认为创始人总是有这种诱惑:想出一个大战略,大肆宣传,然后说‘哦,这就是我们现在在做的事。’但不对,让客户来讲故事。
Yeah. I think there's always this temptation for founders to come up with this big strategy, make a big news about it, and just like, 'Oh, this is what we're doing now.' It's like no, let the customers tell the story.
对吧?更像是你有一个想要达到的总体目标,然后你逐步实现,而增量是什么?
Right? It's more like you have this general goal of where you want to end up, but then you do it incrementally and what are the increments?
是与客户一起出现的个别事情,但你知道你在寻找什么,对吧?你知道,你在寻找某个客户在做某类事情。好,让我们帮助他们,然后每个人都能做到。
Individual things that come up with customers, but you know what you're looking for, right? You know, you're looking for a customer doing a certain sort of thing. Okay, let's help them out and then everybody will be able to do it.
没错。然后你就进去了。这就是妙处。这就是做过所有那些系统编程给你带来的好处。它让你进入客户内部,这样你就能看到新事物的增量,你正在
Exactly. And you're in. That's the great thing. This is what having done all that systems programming buys you. It gets you into the customers so you can see the increments of the new things you're
是的。
Yes.
同样地,智能体取代智能体对外界来说是一个不连续的时刻,但对我们内部来说在很多方面是渐进的。我认为我们正在做的很多事情,在某个时候我们会觉得公司运营中存在这种连续的时刻。
And similarly, how agent replaced agent for the outside world was this discontinuous moment for us internally was incremental in many ways. I think a lot of things that we're doing at some point we were like there's this continuous moment in how companies operate.
这很有趣,如果你是专家,如果你是内部人士,那些看起来像阶跃函数的事情。
It's so interesting if you're the experts, if you're the insiders, things that seem like step functions.
是的。
Yes.
对于其他人来说,你看到了阶跃之前和之后。所以对你来说更平滑。
For the rest of the world, you saw before the step and you see after. So, it's smoother for you.
更平滑。我认为这不全是发明,但我想发明的历史,如果你放大并阅读那些故事,比如史蒂夫·乔布斯在 80 年代试图做 iPad,你知道,就像
It's smoother. And I think that's not all invention, but I think the history of invention, if you zoom in and kind of read the stories like, you know, Steve Jobs tried to do an iPad in the 80s, you know, it's like
因为他知道那样的东西会很好。只是技术还没到位。
cuz he knew that something like that would be good. The technology just wasn't there.
是的。所以当你读苹果的故事时,会觉得 iPhone 是完全显而易见的,尤其是之后
Yeah. And so when you read the story of Apple, it was like, oh, the iPhone was totally obvious, especially, you know, after it like
是啊。是啊。如果那么明显,为什么没有其他人做呢?对吧。对吧。
Yeah. Yeah. If it was so obvious, how come nobody else did it? Right. Right.
呃,但但但就像在苹果内部是渐进的,但对世界上的每个人来说,那是一个完全变革性的时刻
Uh but but but like it was sort of incremental inside Apple but for everyone in the world that was like a totally transformational moment
但如果你也知道 iPad 是如何发生的故事
but also if you know the story of how the iPad the iPad happened
嗯,或者 iPhone,嗯,最初他们在做这个平板电脑,他已经知道想要什么。他一直在思考这个,所以他一看到它就说好的就是这个。让我们把它做对,然后你
um or the iPhone um originally they were making this tablet and he already knew what he wanted. He'd been thinking about this and so as soon as he saw it like okay this is it. let's make this right and you
每当公司内部发生什么事情时,那一定是你已经思考了 10 年的东西,你会说好的这就是要构建的东西
whenever something happens whenever something happens within a company it's going to be something that you've been thinking about for 10 years you're like okay this is the thing to build
但比其他公司更甚,我认为你们一直生活在未来
more than other companies though I think you guys are have been living in the future
对吧,你不觉得这是真的吗
right don't you think that's true
我认为这是真的
I think that's true
这几乎就是如果你必须说 Replit 的真正 DNA 是什么
that's almost if you had to say like what's the real DNA of replet
那肯定是生活在未来
it is living in the future for sure
甚至比 AI 更具体地说,这经常令人沮丧,我觉得嗯
more even than AI specifically It's it's often frustrating like I think that um
因为东西不起作用
because stuff doesn't work
嗯,东西不起作用,而且人们改变得如此缓慢。例如云开发
well stuff doesn't work and and people change so slowly. So for example cloud developments
人们现在相信这个了吗
are people convinced about that now
仍然,他们将如何使用 AI?
still how are they going to use AI?
这太疯狂了。确实如此。所以呃我们内部的预测是,嗯,你会从每个人使用一个智能体发展到 10 个智能体、100 个智能体,到某个时候 10,000 个智能体。你无法在 MacBook 上运行 10,000 个智能体。这就像你不可能做到。
It's crazy. Exactly. So uh our prediction internally is that um you go from everyone using one agent to 10 agents to 100 agents to 10,000 agents at some point. You can't run 10,000 agents on your MacBook. This is like there's no way you could do it.
是的。罗伯特在 1980 年代就证明了这一点。
Yeah. Robert proved that back in the 1980s.
是的。所以最终我们将摆脱一般的本地工作。但但但但就像人们仍然主要在他们的办公桌上工作。
Yes. So eventually we're going to get away from local work in general. But but but but there's like the people still predominantly do work on their on their desk.
也许这最终会说服他们。
Maybe this is what will finally convince them.
是的。
Yes.
对。那些东西必须活在云中。
Right. That stuff has to live in the cloud.
是的。但这也很有趣。我觉得在 Replit 工作最终会成为一种奇怪的存在。就像你进入这家公司,你会想,“天哪,我”
Yeah. But it's also fun. I think it is it, you know, working at Replit just ends up being being a sort of a weird sort of existence. Like you go into this company, you're like, "Oh my god, I'm
嗯,奇怪的是你知道每个人的未来会是什么样子,这就是为什么当你演示后来被称为氛围编程的东西时,你看起来像邦德反派,加上你的笑声,对吧?你在演示后来被称为氛围编程的东西,而我坐在那里沮丧地看着它,你说,“是的,邦德先生。这是这是不,”
Well, the weird thing is you know what everyone's future is going to be like cuz that was why you seemed like a Bond villain when you were demonstrating what later and you plus that laugh, right? You were you were demonstrating what later came to be called vibe coding and here I was sitting there looking at it with dismay and you're like, "Yes, Mr. Bond. This is This is No,
确切地说。这是这是未来,邦德先生。
exactly. This is This is the future, Mr. Bond.
我觉得你甚至可能穿着黑色高领毛衣。
I think you might even have been wearing a black turtleneck.
我没穿高领毛衣。是一件黑衬衫。但
I was not turtleneck. It was a black shirt. But
好吧,黑衬衫,秃头的我坐在那里想,“哦不,我心爱的编程发生了什么?”而你说,“我早就知道了。我已经思考这个多年了,邦德先生。”
okay, black shirt, the bald me sitting there like, "Oh no, what's happened to my beloved programming?" And you're like, "I already knew. I've thought been thinking about this for years, Mr. B."
是的。我我确实认为我有一种能力,在某种程度上杀死圣牛。就像我一直有这种,当计算机在某件事上变得非常擅长时,好吧,完成了。让我们转向下一件事。我不那么执着于我所做的事情。我可以扔掉东西重新开始,在我的职业生涯中我已经重新开始了很多次。所以
Yeah. I I do I do think I have an ability to like kill the sacred cow uh in a way. Like I think I've always had this sort of when computers got really good at something like all right it's done. Let's like let's move to the to the next thing. Like I'm not like so tied to to what I do. Like I'm I'm fine with like throwing away things and like starting over like in my career I've like started over many times. And so
我记得你总是在 Replit 内部重写东西。
I remember you were always rewriting things within rep.
是的,重写东西。但但我觉得大多数人他们只是有点僵化于“这是我的身份。这就是我做的。”我的意思是他们应该读读你的文章《保持身份渺小》。但但我你知道我我觉得嗯我觉得这个时刻需要大量的适应性。很多人问我哦你如何抚养孩子,你教他们什么?而主要的事情似乎是,
Yeah, rewriting things. And so but but I think most people they just like there's kind of like rigid about this is my identity. This is what I do. I mean they should read keep your identity small uh one of your essays. But but I you know I I think um I think this moment requires a ton of adaptability. A lot of people ask me oh how do you raise children and what do you teach them? And it feels like the main thing,
你教他们关于 AI 的什么
what do you teach them about AI
是灵活性和适应性,以及成为终身学习者。
is like flexibility and adaptability and like being a lifelong learner.
我告诉他们的是,如果你有海啸逼近海滩,最安全的地方是冲浪,
What I tell them, I mean, is that if you have this tidal wave approaching the beach, the safest place to be is surfing on it,
对吧?而不是站在海滩上抓住棕榈树。
right? Not standing on the beach holding on to a palm tree.
它会被冲走。所以这就是你的用户正在做的,
It'll just get washed away. And so that's what your users are doing,
对吧?你的用户,在任何人能看到未来并免受坏事发生的程度上,你的用户就是,因为他们都在使用 AI,就像如果你想使用 AI,Replit 就是使用它的方式,对吧?而且使用它是避免被压扁的最好方法
right? your users like to the extent anybody can see the future and be protected against bad things happening your users are it because they're all using AI like if you want to use AI you like replet is the way to use it right and if you're and using it is the best way not to be squashed by it
对,这就是为什么我们觉得我们的使命几乎有一个道德角度,你想要帮助这些公司转型
right and this is why we feel our mission is there's like a moral angle to it almost where you want to help these companies transform
让那些人离开海滩,让他们都上冲浪板
get those people off the beach and get them all on the surfboard
没错。确实如此。确实如此。我们做了一个黑客松,嗯,与一家非常大的,嗯,你知道,电信公司,他们做了这件事,他们在使用 Replit 之前做了一个调查,使用 Replit 之后做了一个调查,调查问他们,比如你对 AI 有多焦虑?嗯,你知道有多担心等等,所以在使用 Replit 之前仅仅一个小时,80% 的人焦虑,担心他们的工作或其他,而使用 Replit 之后,情况反转了,80% 的人乐观,20% 的人
That's right. Exactly. Exactly. We did this hackathon um with um with a very large um you know uh telecommunication company and they did this thing where they did a survey before they used Replit and a survey after they used Replit and the survey asked them like how anxious are you about AI? um you know how worried and so on and so pre using replet just for like an hour like 80% were were anxious and were worried about their jobs or whatever and after they used replet it flipped like uh 80% were optimistic and 20% were
一个小时
for one hour
一个小时,是的,我认为硅谷的人不断低估世界其他地方的人还没有经历我们所经历的事情,而且人们需要很长时间
1 hour yeah and I think people in Silicon Valley do not u constantly underestimate how um how like the rest of the world have not experienced what we've experienced just yet and and It's just going to take a long time for people
因为事情发展得如此之快,
because things are moving so fast,
对吧?
right?
这里的事情发展得如此之快。
Things are moving so fast here.
嗯,不,但这就是我说的。如果他们只是落后两年,那就是永远落后,对吧?因为两年前,VIP 编程还不存在。
Well, no, but that's what I'm saying. If they're just two years behind, that's forever behind, right? Cuz two years ago, VIP coding didn't exist.
是的。我的意思是,大多数银行基本上在使用像 2023 年的 ChatGPT。就像大多数银行推出的东西,除了工程部门,我觉得像
Yeah. I mean, most banks are using like chat GPT like 2023 basically. Like the thing that is rolled out at most banks like other than engineering, I think like
CHBT3 或类似的东西。
CHBT3 or something like that.
是的。
Yeah.
它只是一个 LLM。它几乎不连接任何东西。所以它没那么有用。因此人们的使用方式存在真正的滞后。硅谷已经转向这些为你做事的智能体。但大多数公司仍在试图弄清楚如何处理他们拥有的所有遗留软件,他们想要重写等等。
It's just an LLM. It's barely connected to anything. And so it's not that useful. And so there's a real lag in how people use it. Silicon Valley already went to these agents that are doing things for you. But most companies are still trying to figure out what to do with all the legacy software that they have that they want to rewrite and so on.
他们不能直接忽略它,对吧?这就是现状。这就是公司内部的软件现状。也许所有这些 COBOL 代码最终会被重写。
And they can't just ignore it, right? That's the landscape. That's the software landscape within the company. Maybe all this COBOL code will finally get rewritten.
是的。它开始产生收入了。它终于要消失了。
It is. It is starting to get revenue. It's finally going away.
COBOL 会消耗大量 token。
COBOL would eat up a lot of tokens.
强大。在我们 AI 旅程的早期,很多公司会说,‘哦,这个 Replit 东西很酷,但你为什么不用它来重写我们的 COBOL?’我说,‘我们不会为你写代码。我们会教你如何使用这个东西。’但现在有很多 AI 公司做这个,他们会来重写你的 COBOL。
Powerful. Early on in our AI journey, a lot of companies would be like, 'Oh, this Replit thing is cool, but why don't you use it and come rewrite our COBOL?' I was like, 'We're not going to write your code for you. We'll teach you how to use this thing.' But now there's a bunch of AI companies that do that, that will come and rewrite your COBOL.
是的。所以我想问你最后一个问题。整个 Replit 在其整个历史中,之所以能成功,是因为你在某种程度上生活在未来,就像你个人,对吧,生活在未来。你是怎么做到的?你为什么生活在未来?
Yeah. So there is one last question I want to ask you. This whole Replit for its entire history, the reason it has worked is because you're living in the future to such an extent, like you personally, right, are living in the future. How do you do that? Why are you living in the future?
是的。我的意思是,我在我们对话的早期就开始回答这个问题的某个版本,我说过,你有一个很好的观点,这不仅仅是坐着思考,但我认为至少有一部分是我积极尝试预测事物,我积极修正我的预测并审视它们,我认为这是一种你可以擅长的技能,有一些你可以遵循的原则。早期我读了雷·库兹韦尔。雷·库兹韦尔的核心观点是,如果你绘制世界上所有计算 FLOP 的对数线性图,它是一条该死的直线,实际上 AI 并没有弯曲它,这很疯狂。它仍然是一条直线。
Yeah. I mean, I sort of started answering some version of that early on in our conversation and I was saying, and you had a good point that it's not just about sitting and thinking about it, but I think a component of it at least is that I actively try to predict things and I actively revise my predictions and look at them and I think it's a skill you can get good at it and there are some certain principles that you can follow. Early on I read Ray Kurzweil. Ray Kurzweil's whole thing is if you chart a log-linear chart of all compute flops in the world, it's this freaking straight line and actually AI didn't bend it, which is crazy. It's still a straight line.
它需要 AI 发生。它需要 AI 发生。这几乎是一种宗教般的事情。就像它无论如何都会发生。
It needed AI to happen. It needed AI to happen. It's almost a religious sort of thing. It's like it'll happen anyways.
是的。我看到有人说,你知道,发推文说。他说,‘不是 AI 带来了算力。是算力带来了 AI。’这是真的。
Yeah. I saw someone say, you know, tweet this. He said, 'It's not that AI brought on the compute. The compute brought on AI.' That's true.
这是一件迷人的事情。就像没有算力就不会有深度学习。
Which is a fascinating thing. Like deep learning would not have happened without the compute.
不。在 80 年代,每个人都认为神经网络不起作用。这就是神经网络的问题。它们不起作用。但事实证明,如果你给它们 100 亿倍的资源,它们就开始起作用了。是的。
No. Everybody thought neural networks didn't work back in the 80s. That was the thing about neural networks. They didn't work. But it turns out if you give them 10 billion times as much resources, then they start to work. Yeah.
谁知道呢?所以那条线,通过更广泛地真正理解那条线,你可以做出很多预测。
Who knew? So that line, there's a lot of predictions you can make by truly understanding that line more broadly.
好的。所以你如此擅长生活在未来的原因是你已经将那条线正常化了。
Okay. So the reason you're so good at living in the future is because you have normalized that line.
嗯。
Mhm.
你假设你不是生活在一个静态的世界里。事实上,你生活在一个指数级的世界里。
You assume that you're not living in a static world. In fact that you're living in an exponential world.
是的。指数级世界。你在文章中谈论的很多内容都是理解指数级。我认为一旦你真正理解了指数级,就像你可以看一个 LLM,我可以看 GPT-2 并比大多数人更好地推断,只是说我想象聊天,我想象智能体,我想象个人助理,所有这一切在 2019 年我第一次看到 GPT-2 时。所以我认为但你必须相信,你必须因为大多数人有这种几乎你知道的暂停怀疑,就像暂停,你必须真正相信,你必须真正拥抱它并违背你的本能,因为我认为有一种人类本能,啊,有所有这些理由为什么这不会起作用以及所有这些
Yes. Exponential world. And a lot of what you talk about in your essays is understanding the exponential. And I think once you truly understand the exponential, like you can look at an LLM, I can look at GPT-2 and extrapolate a lot better than most people and just say I imagine chats, I imagine agents, I imagine personal assistant, all of that back in 2019 when I first saw GPT-2. So I think but you just have to like believe and you have to because most people have like this almost like you know there's suspension of disbelief like suspension of like you you have to actually believe you have to truly embrace that and go against your instinct to because like I think there's a human instinct ah there's all these reasons why this would not work and all these
人们隐含地将事物视为静态的。
people implicitly treat things as static.
是的。与你的不同之处在于你的默认假设是事物将以指数级变化。这就是你的不寻常之处。
Yeah. And the difference with you is that your default assumption is that things will change exponentially. That's what's unusual about you.
是的。所以从小接受这一点,比如阅读科幻小说和阅读库兹韦尔以及所有这一切,我认为这在我身上创造了两件事。一是接受变化并对变化感到满意的能力。计算机将越来越多地做我的技能的事情。
Yes. So having embraced that as a kid like reading sci-fi and reading Kurzweil as well and all of that, I think just you created two things in me. One is an ability to accept change and be okay with change. Computers are going to do more and more of the things of my skill.
只是接受,期待。
Just accept, expect.
期待,然后期待并朝着那个方向建设,滑向冰球将要去的地方。
Expect and then expect and build towards that and skate where the puck is going.
冰球移动得越来越快。
And the puck is moving faster and faster.
它移动得越来越快,这就是它的意思。
It is moving faster and faster, which that's what it means.
但我现在感觉更自在,因为
But I feel more in my element right now because
嗯,因为没人能跟上你。这就是为什么一个假设剧烈变化的人。想象一下谁会是这个世界的完美人选,每 6 个月一切都在我们脚下变化。假设剧烈变化的人,对吧?你就像完美适合这个时刻。这就是为什么 Replit 做得这么好。
well cuz nobody else can keep up with you. That's why someone who's like assumes dramatic change. Imagine who would be perfect for this world where everything is changing under our feet every 6 months. The person who assumes dramatic change, right? You're just like perfectly suited to this moment. That's why Replit's doing so well.