当人人都能构建软件：Supabase 的智能体驱动未来

我对很多人未来的可能性感到由衷兴奋。我特别喜欢 Superbase 的一点是，我因为这个平台看到了很多创业精神。我确实认为，世界上每个能接触到 AI 的人都会成为某种意义上的构建者。你可以构建任何东西，你可以定制它。如果你想对某个应用、某个 UI 做些调整，你就不必受制于人。实际上，我对 2027 年相当期待。

I'm genuinely excited about what the future could be for a lot of people. The thing that I really love about Superbase is that I see a lot of entrepreneurship because of the platform. I certainly think that it'll be the case that everyone in the world with access to AI will become a builder of some sort. You can build anything, you can customize it. If you want to make tweaks to some app, some UI, then you don't have to be beholden to someone. I'm pretty pumped about 2027 actually.

欢迎回到 Lightcone 的另一期节目。今天我们邀请到了 Paul Copplestone，Superbase 的联合创始人兼 CEO。Superbase 是一个开源的 Postgres 后端，已经成为 Claude Code 和 Coda 等工具中的默认数据库。Superbase 现在有 1300 万用户，每周创建超过一百万个数据库。Paul，欢迎来到 Lightcone。

Welcome back to another episode of the Lightcone. Today we're joined by Paul Copplestone, co-founder and CEO of Superbase, the open source Postgres backend that's become the default database inside tools like Claude Code and Coda. Superbase now has 13 million users and it's spinning up over a million databases a week. Paul, welcome to the Lightcone.

谢谢邀请。

Thanks for having me.

那么开始吧，你知道，YC 在第一天会给你一件 T 恤，上面写着“做人们想要的东西”，但越来越多人意识到，现在是“做智能体想要的东西”。你的公司真正体现了这一点。你知道，大约 70% 的新活动来自智能体。这难道不疯狂吗？

So to start off, uh you know, famously YC gives you a t-shirt on the first day of YC that says make something people want, but increasingly all of us are realizing it's make something agents want. And your company really exemplifies that. You know, something like 70% of the new activity is from agents. Isn't that crazy?

是的，太疯狂了。太疯狂了。我觉得尤其是今年，从数据上看，拐点就是向智能体的巨大转变。

Yeah, it's wild. It's wild. I think like this year especially what we've seen in terms of the numbers, the inflection is just a huge shift towards agents.

为了让我们了解你们增长了多少，因为这有点令人难以置信。是的，比如你关心哪些数字，去年这个时候你们在哪里，现在又在哪里？

Just so we can understand just how much you guys have grown cuz it's kind of mind-boggling. Yeah, like what are the numbers you care about and where were you like this time last year and where are you now?

我想去年这个时候，是的，我们现在有 1300 万注册用户。大概 12 个月前可能只有 300 万左右。今年营收将增长 4 倍。我们正在想我们

I think this time last year, so yeah, we're 13 million sign-ups now. Probably 12 months back we would have been around three perhaps. This year like revenue will grow 4x. We've got trying to think what we

比如目前每周创建一百万个数据库。

like a million databases being created per week at the moment.

是的，超过一百万。是的，是的，是的。

Yeah, over a million. Yeah, yeah, yeah.

去年这个时候是多少？

What was it this time last year?

实际上，现在管理的数据库接近 5000 万，而去年这个时候是 800 万，所以这是一个巨大的拐点。

I think actually if databases under management now are approaching 50 million and this time last year it was 8 million, so it's just a huge huge inflection.

是什么推动了这一增长？

What's been driving that?

首先，软件工程师们正在放手构建。其次，我认为有更多的构建者上线。所以，传统上需要软件工程背景的人，现在非工程师也能够在 Superbase 上构建东西了。

First, people are building with abandon, software engineers. Second, I think that there are a lot more builders coming online. So, what was traditionally someone with a software engineering background is now is now becoming possible for non-engineers to actually build things on top of Superbase.

所以，这是一个巨大的转变。你们实际做了什么来促进或鼓励这种转变？显然软件工程师会继续选择 Superbase，但你们最初是如何让智能体过来的？

So, this is a really big shift. What have you guys actually done to facilitate or encourage this shift from, you know, obviously software engineers are going to keep choosing Superbase, but you know, how do you get the agents to come over in the first place?

是的，我们在 Superbase 大致经历了三个篇章。我们最初是一家仪表盘公司，然后成为一家 API 公司，现在我们非常注重代码优先，所以我们尝试提供智能体真正想要构建的方式，即代码优先。它们大多数时候真的想使用 CLI。另一件事就是被选为 LLM 中的默认选项，我认为我们是 timing 的巨大受益者，我们开始成为数据库的默认选择。

Yeah, we largely have seen like three chapters in Superbase. We started as a dashboard company, then we went to an API company, and now we're very much a code-first company, so we try to have the modalities that the agents really want to build with, which is code-first. They really want to use CLIs most of the time. The other thing is just being chosen as a default in the LLMs, which I think we were a huge beneficiary of our of our the timing of which we started to become a default choice for databases.

你认为开源在其中起了多大作用？因为智能体喜欢开源的东西。

How much do you think being open source played a role in it, too? Cuz the agents love open source things.

你知道，很难知道反事实，但我认为这很关键。我的意思是，我们做的一切都是开源的，这似乎给我们带来了巨大的好处，我们以非常开源的方式来做。比如我们尝试支持现有工具。不仅仅是 Superbase 工具。我们支持像 Postgres 这样 Postgres 生态系统中非常著名的工具，我们帮助维护它，所以我们扩展到许多社区。

You know, it's hard to know the counterfactual, but I think it was critical. I mean, everything that we do is open source and it seems to have been a a huge benefit to us and we do it in a very open sourcey way. Like we try to support existing tools. It's not just Superbase tools. We we support like Postgres as a very famous tool for the Postgres ecosystem and we help to maintain that, so we kind of got sprawl across many communities.

我猜智能体实际上真的很喜欢开源，因为当出问题时，它们可以直接查看源代码。

I would guess that the agents actually really like open source because when something goes wrong, they can just look at the source.

是的。

Yeah.

是的，我的意思是，我注意到智能体真的很喜欢感觉强大，它们热爱数据。没有什么比撞墙然后不知道发生了什么且无法修复更令人沮丧的了。

Yeah, I mean, I noticed the agents really like feeling powerful and they love data. And uh there's nothing more frustrating than hitting a brick wall and then you don't know what's going on and you can't fix it.

我认为这也会越来越重要，因为现在安全如此重要，我认为很多企业会想要检查源代码以确保安全。如果你有闭源代码，那么你可能不会想采用它们作为供应商。

I think it's going to be increasingly important as well because now with you know, security being so important, I think a lot of the say enterprises are going to want to inspect the source code to know that it's secure. And if you've got closed source code, then you probably won't want to adopt them as a vendor.

开发者还是你们追求的核心用户吗？或者你知道，在 2026 年你们怎么考虑这个问题？

Is developer even still sort of the the core user that you're going after or how you know, how do you think about it now in 2026?

是的，我认为我们正在走向的世界，至少今年我们看到的是，构建者正在上线，上线的人数疯狂增长。今年疯狂的是，通常当我们的漏斗顶部，注册加速时，你通常会看到其他许多数字开始下降，激活、转化，但今年所有这些都在向右上走，这有点疯狂，因为模型在构建方面变得如此之好，我们开始看到传统上没有软件工程背景的人能够来构建东西。

Yeah, I think that the world where we're moving and at least what we're seeing this year is like uh kind of builders, builders are coming online and it's like a crazy amount of people coming online. The crazy thing about this year is usually when you're uh like our top of funnel our sign-ups are accelerating and usually when this happens, you see a lot of the other numbers start to decline, activation, conversion, but actually this year all of them are kind of going up into the right, which is a bit insane just because the models are getting so good at building and we're starting to see people who traditionally wouldn't have a software engineering background able to come and and build things.

这就是区别，构建者可能不是真正的软件工程师，他们可能只是需要构建某个东西的人，他们只是打开 Claude Code，然后发现文件只能带你走这么远。

That's the distinction is like builder as in they might not be actually software engineers so they might just be someone who needs to build a thing and they're just you know, popping open Claude Code and then it turns out files only get you so far.

嗯。是的。在某种程度上，这类人有很多想法，他们永远无法说服软件工程师帮他们构建，现在他们不必说服任何人。他们有所有这些想法在产出。软件工程师仍然占很大比例，但构建者部分的增长率相当大。而且我不知道这个的上限在哪里。我记得我们的早期投资者说 PLG，产品驱动增长，会在某个营收点停止，你必须开始进入这个细分市场，但到目前为止似乎没有上限，我看不到它在哪里停止。

Mhm. Yeah. And in some ways like this category of people have all these spinned up ideas that they could never convince a software engineer to like help them build and now they don't have to convince anyone. They've got all these ideas that they're churning out. Software engineers still make up a huge portion, but the rate of the growth of the builder segment is is pretty pretty big. Also like I don't know where the ceiling is on this one. Like I I remember our early investors were saying PLG, product like growth, it's going to tap out at this revenue and you'll have to start moving into this segment of the market and there just appears to be no ceiling so far and I I can't see where it stops.

我的意思是，这在两年前或一年前基本正确，但今年绝对不正确。

I mean, that was basically true like 2 years ago or 1 year ago, but definitely not true this year.

是的。

Yeah.

实际上我上周末刚和一个人聊过，如果属于构建者类别，他就像一位金融科技创始人，刚以超过十亿美元退出公司，他一直有点接近产品，在 YC 本科学习数学，能拼凑一些 Ruby 脚本，但他们总是需要工程师来构建软件，现在成了副业项目，他只是在构建自己的东西，他拥有如何构建金融科技公司的所有知识。

I was just talking to someone actually last weekend if it's the builder category, it's like a fintech founder just exited his company with like over a billion dollars and he's always been sort of product adjacent like studying math at YC as an undergrad and sort of like kind of could pull together some Ruby scripts, but they always need an engineer to like build the software and now it's side projects, he's just building his own stuff and he has all the knowledge from how to build a fintech company.

所以他实际上能构建相当复杂的软件，比如用于他个人财务的资产负债表和追踪软件，并且知道如何与银行集成。他从头编写所有集成，这正是那种甚至两年前都永远不会被构建出来的软件。

So he can actually build pretty sophisticated software, like for his own personal finances—balance sheets and tracking software—and knows how to integrate with banks. He's writing all the integrations from scratch, and it's just the kind of software that would never have been built even two years ago.

这些也是完美的构建者类型，因为像我这样的开发者有非常开发者导向的想法和想构建的东西。但然后你有所有这些领域专家，他们确切地知道细分领域——哦，这是我细分领域中的一个未解决的问题。所以现在他们实际上能够直接深入这些细分领域，而不必说服别人加入他们。我想你们在 YC 里讨论过这个，不是吗——比如甚至从你们引入 cohort 的人的资料都在转变。

These are the perfect type of builders as well, because developers like me have very developer-focused ideas and things that we want to build. But then you've got all these domain experts, and they know exactly the niches—oh, this is an unsolved problem in my niche. And so they're actually able now to dig right into those niches without having to convince someone else to join them. I think you guys were talking about this, weren't you, in YC—like the profile shifting even from who you're bringing into the cohort.

我认为如果你看 2022 到 2025 年这个时代，有一种真正的思维转变：嗯，AI 要来了，突然之间软件工程成为核心——无论如何你都需要一个技术联合创始人。那是一个巨大的想法，在那段时间里在 YC 真正凸显出来。这一直有点道理，但更像是加倍下注。然后真的在去年，我们有点说：“实际上，是的，构建者。”就像我们开始看到人们能够——我的意思是更难的事情是知道要构建什么。你知道，能够构建是现在任何人都可以随时拥有的东西。这实际上是一件非常酷的事情。就像你能想象所有那些永远不会被构建出来的东西吗，正是你所说的？就像，“哦，我们知道人们会买这个。我们知道这里有问题，但在边缘上那些人无法说服工程师去做。”

I think that if you look at the sort of era of what, 2022-2025, there's sort of this real shift thinking like, well, AI is coming and suddenly having software engineering as core—you need a technical co-founder no matter what. That was like a mega idea that really came to the fore at YC during that time period. It's always sort of true, but like sort of double down on that. And then really in the last year we sort of said, "Actually, yeah, builders." Like we're starting to see people who are able to—I mean the much harder thing to do is to know what to build. You know, being able to build is something that anyone can sort of have on tap now. And that's actually a really cool thing. Like can you imagine all the things that never would have been built, exactly for what you said? It's like, "Oh, we know that people would buy this. We know that there's a problem here, but at the margins those people couldn't convince an engineer to work on it."

你有例子吗，Paul，人们在 Supabase 上构建的应用？也许只是为他们自己构建的，那些以前永远不会被构建出来的东西？

And do you have some examples, Paul, of apps that people have built on Supabase? Maybe just for themselves that are the kinds of things that would have never been built before?

嗯，一个挺酷的例子是在我们的会议上。我们在这里开了会议，之后这个来自巴西的家伙走到我面前，他已经从他的手机上构建了一个完整的出租车应用。它有点像为他的本地市场做的小型 Uber。我当时想——他自己就是个出租车司机。我其实不知道他为什么构建它，但他能构建它，而且他走过来时非常感激，感激他能做到这一点。还有他那种能量——他大概 50 岁左右，但他突然有了青少年般的构建能量，我觉得这挺酷的，因为现在到了这样一个阶段，我觉得有点——关于工作的未来会是什么样子有太多不确定性，但在构建者这个类别中，有那种对未来的兴奋。这正是我喜欢 Supabase 的地方。就像人们可以进来，那些现在不确定的人正在发现这套技能和这些机会，而以前他们只是蓝领工人，这类人群。

Well, one kind of cool example was at our conference. We had the conference here and afterwards this guy from Brazil came up to me and he has built an entire taxi app from his phone. And it's kind of like a little mini Uber for his local market. And I just thought—he's a taxi driver himself. I don't actually know why he built it, but the fact that he could build it and he came up extremely grateful and that he could do this. And also like with the type of energy he—he was like 50 years old or something like that, but he had the energy of a teenager building suddenly, which I think is kind of cool because now it's kind of getting to that stage where I think it's a little bit—there's so much uncertainty around like what is the future of work going to look like, but in this category of builders there's that excitement of the future. And that's really what I love about Supabase. It's like something that people can come in and those who are uncertain now are discovering this skill set and these opportunities where previously they were just blue-collar workers, these type of profile.

好的，所以你最近一直在构建一些你自己的个人软件。你能谈谈你一直在捣鼓什么吗？

All right, so you've been building some of your own personal software recently. Can you talk about what you've been cooking up?

是的，有点像一站式的终点，但它只是从对 open core 上瘾中生长出来的。实际上，大概六个月前——我很尴尬我这么说过——我说过不会有 UI，一切都只是通过我的 open core 聊天。但事实证明——而且我也曾下定决心很长时间不用数据库。我只想让一切都在 markdown 里。嗯，但是的，事实证明你会到达一个点，数据库有用，你确实需要 UI。所以我一直在构建那个，特别是过去几周。所以就像——我的意思是，我称之为控制塔，但它就像我生活的每个部分，我发现我有一个 UI 来分类所有邮件，所以我不需要阅读收件箱。嗯，财务，追踪支出，健康。比如我追踪体重和锻炼，一切都在一个地方，但我喜欢它的魔力是，有时这是一个令人难以置信的想法，就像我想到我的一个应用中的财务部分，我认为至少与 Mint 或 Monarch 或所有这些功能相当，但在某些方面实际上更好，因为你可以开始做真正个性化的上下文。比如它可以通过查找我 G brain 中的名字来弄清楚交易是什么，那个人是谁。哦，就像我知道那是谁，并正确分类。那曾是一整个公司，好几家初创公司只是做个人财务，而我可以直接为自己构建，我不需要它适用于其他任何人，也不需要其他任何人对它有意见。就像我的东西，这挺棒的。

Yeah, it's sort of like the end of one stop, but it's just growing out of getting really addicted to open core. Actually, like maybe six months ago now—I'm embarrassed that I said that—I said that there was going to be no UIs and it was all just going to be like chat through my open core. But it turns out actually—and I also was really determined not to use a database for a long time. I just wanted everything to be in markdown. Um, but yeah, it turns out actually you hit a point where like a database is useful and you do need UI. And so I've just been building that over the last couple of weeks in particular. And so it's like—I mean I call it a control tower but it's just like sort of every section of my life I find like I have like a UI to like triage all my emails so I don't need to read my inbox. Um, finances, tracking like spending, health. Like both I track my weights and my workouts and everything is all in like one place but I like the magic of it is sometimes it's a mind-boggling thought actually is like I think about like the finance section of like my one app for myself is better—I think is at least feature parity to Mint or Monarch or all of these things but in some ways actually better because you can start doing really personalized context. Like it can figure out what a transaction is by looking up in my like the name of my G brain out who that person is. Oh, like that I know that's like and categorize it correctly. And that was like a whole company like several startups just doing personal finance and I can just like build it for yourself and I don't need it to work for anyone else or like have anyone else's opinion on it. It's just like kind of like my thing is like is kind of awesome.

我们看到很多这样的，比如财务和健康基本上是两个你想要自己数据的领域。所以我们看到很多人这样做，将设备上的数据同步到一个数据库是一种神奇的技能，你可以提示它构建跨所有设备的这些关联。

We see a lot of that one, like where it's finance and health basically the two ones where you want your own data. So we see a lot of people doing that and syncing on the devices to a single database is kind of a magic skill and you can like prompt it to like build these correlations across all of your devices.

那真的像——再次，如果你只拿健康例子来说，就像是的，所有数据都在我的 Supabase 数据库里，但它也可以访问我的日历，所以它知道一些事情。就像哦，你的体重——我知道你最近体重增加了，但你在旅行，所以没关系。就像你可以下周回到正轨，它说这样的上下文真的很有价值。

That are really like—again like if you just take the health example, it's like yeah all of that data is in like my Supabase database but then it also has access to my calendar so it knows things. Like it's like oh like your weight—I know your weight went up recently but like you were traveling so that's okay. Like you can just like get back on track next week and it says like this context that's really valuable.

我也发现，当我开始使用个人智能体时，我有点偏执，我把所有例程和一切，CRM 变得像被框在一个智能体里。所以现在我慢慢把它提取到一个中，这样如果我想在上面附加不同的智能体，我知道我可以使用任何智能体。所以 CRM 是另一个非常关键的东西。

I also found that as I started to use the personal agents, like I'm getting a little bit paranoid that I'm putting all of like all of the routines and everything and the CRM is becoming like boxed into one agent. And so now I'm slowly extracting it out into one so that if I want to attach different agents on top, then I know that like I can use any agent. So the CRM is another really critical one.

是的，这就是为什么你们如此有价值，实际上就像它是记录存储——数据库实际上是宝贵的东西。就像你不——你想拥有自己的数据库，然后在上面构建所有应用或智能体，而不是把所有东西都放在别人的数据库里。

Yeah, that's why you guys are so valuable is like actually like it's the store of record—what's the database is actually the thing that's valuable. Like you don't—you want to own your own database and then build all the applications or agents on top of it versus having that all in like someone else's database.

这是一个完全疯狂的想法。就像如果你只想想数据库的历史，过去的情况是，世界上只有相对少数由大公司运行的非常大的数据库。我记得当 Supabase 起步时，你有点这种想法，它们应该让每个人都能非常容易地创建数据库。

Which is a totally wild idea. Like if you just think about the history of databases, used to be the case that like the world had only a relatively small number of very large databases run by like large companies. And I remember like when Supabase got started, you kind of had this idea that they should be like really easy for everyone to like make a database.

当时几乎说不通为什么这么多人想这么做，因为除非你会写代码，否则你能用数据库做什么？而且实际上并没有那么多人能写代码并构建那样的个人 CRM。在 AI 之前，对大多数人来说工作量太大了。

And almost like it didn't make sense at the time why so many people would want to do that, because unless you can write code, what can you do with a database? And there are just not that many people who can actually write code and build a personal CRM like that. It's just too much work for most people pre-AI.

Supabase 对我真正有价值的另一个例子是文件上传部分。听起来很简单，但任何文件我都可以通过电子邮件转发给你，或者直接扔进这个 UI，然后我的智能体会弄清楚那是什么、它与我的生活有何关联以及如何处理它。但所有这些都存储在 Supabase 上。所以总的来说，Supabase 最终会把关于我生活的所有数据集中在一个地方。一年前我绝不会想到需要一个个人数据库，因为那毫无意义。但现在这实际上是最有价值的东西。

Another example where Supabase is really valuable for me is the file upload section. Sounds very simple, but any file I can send you forward via email or just dump it in this UI, and then my agent figures out what that is and how it's relevant to my life and what to do with it. But that's all being stored on Supabase. So in general, Supabase is actually going to end up having all of the data about my life in a single place. And I would never have thought about wanting a personal database a year ago because it wouldn't have made any sense. But now that's actually the most valuable thing.

是的，你有一个类似 S3 兼容的 Supabase 文件存储，我所有的东西都在里面。

Yeah, you have like an S3 compatible Supabase files thing and all my stuff is in that.

哦，真的吗？不错。是的。

Oh, really? Nice. Yeah.

这是 G brain 中的默认设置。

It's default in G brain.

太好了。谢谢。

Great. Thank you.

Paul，你认为这是未来吗？基本上世界上每个人，不是每个开发者，而是每个人，都会有一个 Supabase 数据库，里面存有关于他们生活、健康、财务、人际关系的内容？

Do you think this is the future, Paul, where basically everyone in the world, not every developer, every person, will have a Supabase database that has stuff about their life in it, their health, their finances, their relationships?

是的，我们很乐意。我当然认为，世界上每个能使用 AI 的人都会成为某种形式的构建者。你可以构建任何东西，可以定制它。你想对某个应用、某个 UI 做些调整，就不必受制于人。当然，我们把自己视为一家指数公司。当我们进入 YC 时，我们只关注 YC 公司。我们想赢得所有 YC 客户。那是因为我们知道，在企业起步的第一天我们就必须在那里。使命没有变，只是视野大大扩展了。现在，随着更多构建者上线，我们知道我们需要为他们服务，这实际上是一个完全不同的问题，但令人兴奋。

Yeah, we'd love that. I certainly think that it'll be the case that everyone in the world with access to AI will become a builder of some sort. You can build anything, you can customize it. You want to make tweaks to some app, some UI, then you don't have to be beholden to someone. Of course, we think of ourselves as an index company. When we came into YC, we just focused on YC companies. We wanted to win all YC customers. And that was because we knew that we had to be there on day one when businesses were getting started. The mission didn't change, just the aperture has increased significantly. Now, as more builders come online, we know that we need to service them, which is actually a whole different problem to solve, but kind of an exciting one.

大大扩展意味着像 1000 万倍。

Significantly meaning like 10 million X.

是的，我们现在每周发布一百万个数据库，但如果你做一下基本算术，假设用 ChatGPT，我想他们说有 12 亿周活跃用户。假设他们都想每月发布一些东西，那就是每月 12 亿个数据库。所以，这就是我们正在迈向的规模，我们需要拥有的架构，以便真正满足那种规模，这在数据库领域是前所未有的。

Yeah, so what, we launch now a million databases every single week, but if you do some basic arithmetic, let's say with ChatGPT, I think they said they've got 1.2 billion weekly active users. And let's say all of them want to launch something every month, that's 1.2 billion databases a month. And so, that's the scale that we're moving towards, architectures that we need to have so that we can actually cater for that scale, which is something that has never had to happen before on the database realm.

你甚至能用 Postgres 做到吗，还是你必须构建自己的数据库引擎来扩展到每周 10 亿个新数据库？

Can you even do that with Postgres, or are you going to have to build your own database engine to scale to 1 billion new databases per week?

我想是的。嗯，如果我们能更好地与 AWS 谈判，也许我们可以，但事实证明现在每个人都想要计算机。而且实际上，我不知道对于许多工作负载是否需要这样的法拉利。我们最近收购了 Turso。他们有一个与 Postgres 兼容的架构。实际上，它也与 SQLite 兼容。我认为 SQLite 是一个非凡的数据库。所以，我们将提供这个层级，我们称之为 Supabase Spark，基本上用于这些入门级工作负载。想法是一样的。你以相同的体验开始。你可以扩展到 PB 级别。所以，我们的口号是从原型到 PB 级。作为一家公司，我们需要确保能为每个人提供服务。如果他们确实扩展了并且业务持续增长，那么我们要确保他们有一个愉快的体验。

I think, yeah. Well, if we could negotiate better with AWS, then maybe we could, but it turns out that everyone wants computers at the moment. And actually, I don't know if you need such a Ferrari for a lot of these workloads. And we made an acquisition recently, Turso. They've got an architecture that's compatible with Postgres. And actually, it's compatible with SQLite as well. I think SQLite's a phenomenal database. And so, we'll provide this tier, which we call Supabase Spark, which is basically for these getting started workloads. The idea will be the same. You get started with the same experience. You can scale up to petabytes. So, our tagline's kind of prototype to petabytes. And we as a company need to make sure that we can provide for everyone. And if they do scale up and their business keeps growing, then we make sure that they have a great time.

是的，这就是我好奇你实际上如何思考这个问题的。就像经典的软件公司，你从早期初创公司开始，比如在 YC 批次中，然后你有点关注企业市场。我觉得特别是如果你是一家数据库公司，在 AI 之前，那种未来主义，就像你想成为 Oracle，所以你必须尽快进入企业市场，因为那里是所有钱所在。所以你开始构建企业功能。你倾向于 SQLite 是有点相反的方向。比如你如何思考这个问题？你是在同时关注两个方向吗？

Yeah, that's what I was curious about how you think about this actually. Like the classic sort of software company, you start with, you know, maybe early-stage startups, like in a YC batch, and then you sort of have one eye towards the enterprise. And I feel like especially if you're a database company, like pre-AI, the kind of futurism as well, like you want to be Oracle, so you got to get to the enterprise as quickly as you can because that's where all the money is. And so, you start building out like enterprise features. You're sort of leaning into SQLite is sort of going the opposite direction. Like how do you how do you think about it? Like are you sort of looking both ways?

是的。是的，我们不得不两者都做，这非常困难。基本上，我们有两个产品分别满足两端。所以，Multigress 会扩展到 PB 级别，而 Turso 会缩小到这些原型和 SQLite 类型的工作负载。所以，试图以这种增长速度做到这一点，同时不得不雇佣更多人，即使有 AI，也正在成为或一直是一个真正的挑战，但也是一个有趣的挑战。

Yeah. Yeah, we're having to do both, which is incredibly difficult. Basically, we've got two products that cater on the two ends. So, Multigress will scale up to petabytes, and then Turso will scale down to like these prototypal and SQLite type workloads. So, trying to do this at the rate of growth with having to employ more people is just even with AI is becoming or has been a real challenge, but a fun one.

所以你最近做了一个民意调查，显示从 10% 的工程师用编码智能体编写 80% 的代码，然后这个数字在 6 个月内最近上升到了 60%。这是怎么发生的？发生了什么？你知道，从这里往哪里走？

So, you ran a poll recently where people went from 10% of your engineers were writing 80% of their code with the coding agents, and then that number recently got up to 60% in a matter of 6 months. Like how did that happen? What's happening? Like, you know, where's this going from here?

所以，我们内部没有 AI 政策。我们只是让人们做他们想做的事，并让他们访问他们想使用的任何工具。我认为现在只是他们确信模型可以做得更好、更多。所以，采用率非常疯狂，而且他们在内部构建的东西不仅仅是五编码类型的东西。它们相当密集。随着我们转向这个 AI 编码领域，我们正在进行的测试量以及我们为测试采用的工具非常有用，以确保他们真正知道最终结果会是好的。

So, we don't have an AI policy internally. We just let people do whatever they want and we give them access to whatever tools they want to use. And I think now is just that they're convinced that the models can do so much better work and so much more. So, the adoption has been crazy and yeah, the things that they're building internally are not just five coded type stuff. They're pretty intense. And as we're shifting into this AI coded realm, just the amount of testing that we're doing and the tools that we're adopting for testing has been pretty useful to make sure that they actually know that the end result is going to be good.

是的，我们在 YC 的初创公司中看到了这一点，比如 Blacksmith、Depot、Ubi Cloud，它们都是垂直的，因为每个人都在做 CI。

Yeah, we're seeing that across YC the YC startups like uh Blacksmith Depot, Ubi Cloud, like they're all vertical cuz everyone's doing CI.

是的。

Yeah.

而且你拥有的测试数量简直爆表。

And the number of tests that you have are just like through the roof.

用于审查。

for like review.

所以，老实说，看到这个相当疯狂。是不是就像现在模型已经足够好了，人们花了足够的时间后说：“哦，实际上这相当不错。”

So, that's pretty wild, honestly, to see. Was it like a pretty like it's just the models are good enough now and like it's just people spend enough time where they're like, "Oh, actually, this is quite good."

人们需要被说服。尤其是工程师，我的意思是，他们可能在年初还不认为它非常好。

And people need to be convinced. Especially engineers, like I mean, they might not at the start of the year I've thought that it was very good.

模型确实变好了，但人们也开始意识到：“哦，其实用这些模型可以做成一些了不起的事。”还有 Navier-Stokes 方程那一类的东西，都在说服市场：没错，我们确实能开辟新天地。所以我觉得我们正在走向那个世界，而现在人们已经相信事情就会这样发展。

And the models have gotten better, but also they're starting to see, "Oh, actually, you can achieve some great things with these models." And all the stuff with Navier-Stokes and all this stuff, like convincing the market that actually, yeah, we can break new ground here. So I think that's the world that we're moving towards, and now people are convinced that this is the way that it's going to be.

你经历了 Supabase 的三个时代。第一个算是仪表盘公司，在 YC 期间用 100 天做出来的。然后是 API 公司。再到现在，算是智能体时代，就是“做智能体想要的东西”那个时刻。每个时代里有没有一两件事让你印象特别深，就是“哦，正是那件事成就了当时的我们”？

You've seen Supabase through three eras. First kind of like a dashboard company, built in 100 days during YC. Then the API company. And then now it's sort of this agent, you know, make something agents want sort of moment. Was there like one or two things in each era that really jump out at you, just like, "Oh, that was the thing that made us in that moment"?

我觉得那些转变确实发生了。第一个，其实，对，我想你是在说，我们进 YC 的时候，为什么会有人去做数据库公司？

I think those shifts happened. So the first one, actually, yeah, I think you're saying, like, why would a database company exist when we came into YC?

说实话，当时这想法显得特别反主流。

It seemed like such a contrarian idea at the time, actually.

对，对，对。我觉得我们看到了数据库有多难用，哪怕对开发者来说也是如此。所以第一个仪表盘就是：“哦，我们做一个长得像 Airtable 的界面，架在 Postgres 上。”因为我本人也很喜欢 Airtable。我当时想：“如果后端就是一个 Postgres 数据库，那岂不是很酷？”所以那一次，就是看到了连软件工程师都为此头疼。

Yeah. Yeah. Yeah. I think we saw how hard it was to use a database even for developers. And so that dashboard first was like, "Oh, let's build a UI that looks like Airtable on top of Postgres." Because I actually loved Airtable as well. I was like, "Wouldn't it be cool if that was just a Postgres database on the back end?" So that one was just seeing that even software engineers struggled.

然后当我们转向 AI 时，我记得我们在另一档播客里聊过。AI API 那次转变，是因为有个客户在跟我们做集成，他们说：“哦，我们想开很多数据库，需要一个 API 来做这些事。”那大概是 2021 年。结果证明，这恰好是 2024 年最正确的举动——那时 Bolt、Lovable 这些公司开始在我们之上构建产品。

And then when we moved to AI, I think we talked about this on the other podcast. The AI API shift was because we had someone integrating with us and they were like, "Oh, we want to launch lots of databases. We need an API to do all of this." And that was like 2021, maybe. And it turns out that was exactly the right move for 2024, when the likes of Bolt and Lovable and all of these people started building on top of us.

然后今年，我发现其实没人再去仪表盘了。就是没人——

Then this year, I just saw that actually no one's coming to the dashboard. Like no one—

没人再访问仪表盘了。

No one visits dashboards anymore.

那智能体要怎么构建呢？这一次，我给团队发了一份产品备忘录，说必须做到代码优先、原生支持。Supabase 必须转向代码优先。所以凡是仪表盘上有的功能，基本上都必须在代码里同步可用，本质上就是基础设施即代码。我们重新架构了大量 CLI 和相关东西，让它们能在沙箱里以代码方式运行。

So how are agents going to build? In this one, I just kind of put out a product memo to the team that said it has to be code first, get native. Supabase just has to shift to code first. So if it exists on the dashboard, it basically has to work in lockstep in code, infrastructure as code, essentially. And we re-architected a lot of the CLI and everything to work in sandboxes in code.

这可能是很久以前的事了，但你们有没有受到 Parse 的启发？它也是 YC 的公司。

Were you guys maybe too far back, but were you inspired by Parse, another YC company?

对，对。Parse 和 Firebase，这两家公司我都喜欢。这些浪潮很有意思，对吧？因为他们做的是后端，Firebase 被 Google 收购，Parse 被 Meta 收购。但某种程度上，我觉得他们为我们铺了路，让我们能真正做成一门正当的数据库生意。因为 2014 年，你不用 RDS 做数据库简直是荒唐。后来他们出现了，开始转向 Atlas，很多这类数据库公司证明了你不必只用三大超大规模云厂商之一，你确实可以有一家正当的数据库公司。所以那个时机对我们至关重要。

Yeah. Yeah. Parse and Firebase, I liked both of those companies. And it's funny, these waves, right? Because they did the backend, and Firebase got acquired by Google, Parse by Meta. But in some ways I think they blazed the trail for us to come along and actually have a legitimate database business. Because in 2014, it would have been ridiculous not to use RDS for your database. Then they came along, started moving to Atlas, and a lot of these database companies kind of showed that you don't need to just use one of the three hyperscalers. You can actually have a legitimate database company. And so that timing was crucial for us.

是啊，人们有点忘了这一点，但那确实是个做得非常好的产品。他们发布的时候，正好是我第一次做 iPhone 应用，当时能换成一个漂亮的 RESTful API，感觉相当不错。

Yeah, people have sort of forgotten that, but it was like a really well-built product. They launched like just when I was building iPhone apps for the first time, and it was quite nice to switch to a nice restful API as I was trying to deal with.

这些转变总是伴随着拐点发生，对吧？当时市场上发生了两件事。不知怎么的，MongoDB 说服了所有人，认为 NoSQL 是存储数据的绝佳方式。第二件事是移动端。所以我们很幸运，也赶上了几波这样的浪潮。而现在我们正处在浪潮中心。

These shifts always happen with inflection points, right? So at that point, there were two things that happened in the market. Somehow, MongoDB convinced everyone that NoSQL was supposed to be an amazing way to store your data. And the second was mobile. And so, you know, we're fortunate that we had a couple of these waves as well. And now we're right in the thick of it.

对，这正是我刚才问的，就是你怎么看待企业客户和早期客户。因为我觉得，再往前追溯，Heroku 算是第一个这样的：嘿，先从最好的开发者体验做起，然后再往上扩展。但实际上，随着客户成长，扩展起来可能相当难。我觉得 Heroku 有点——我不认为它能完全跨过那道坎，就是当应用变大、需要更高性能等等时，怎么留住客户。但历史上，每家公司都得想办法解决这个问题。而你们看起来也在往那个方向走，但你们也许不必如此。说实话，如果光是“拥有”这个市场，我不知道，就叫它 0 到 10 吧，不管怎样，如果它继续以这个速度扩张，那将会非常巨大。

Yeah, I mean, this is what I was asking about, sort of like how you think about the enterprise versus early, because I think, I mean, if you go even further back, this is sort of like Heroku was the first where it's like, "Hey, just start with the best developer experience and then scale up." But it actually can be quite hard to scale up with people. And I think Heroku sort of—I don't think could quite bridge that gap to how do you keep people on as the apps get bigger and need to be more performant, etc., etc. But historically, every company's had to figure that out. And I mean, you guys are looking that way, but you might not have to. Like honestly, if the market for just owning, I don't know, call it zero to 10, whatever, keeps expanding at this rate, it's going to be huge.

是啊，不过我们全赢了。所以——

Yeah, well, we won it all. So—

所以我们也许不必，但我们肯定会去做。也要向 Heroku 致敬，我觉得他们让 Postgres 流行了起来。但当时 Postgres 还相当不成熟，仍有一些 Scaling（规模扩张）上的限制。所以，对，我们不得不在 Postgres 外面包一层工具，让它能无限扩展。

So we might not have to, but we'll definitely do it. And kudos to Heroku. I think they popularized Postgres. But at the time, Postgres was still pretty nascent and it still got some scaling limitations. And so, yeah, we've had to wrap Postgres in some tooling that will allow it to scale indefinitely.

我猜对你们来说更好的类比是 AWS，它确实能早期获取客户并扩展。所以——对，有哪些经验教训？比如，你怎么确保自己能无限扩展？

I mean, I guess the better analogy for you guys is more like AWS was able to actually get people early and scale. And so, do—yeah, like what are the lessons maybe? Like, how do you think about making sure you can scale infinitely?

我觉得他们的经验是——我听到一个数据，有人告诉我，哦，我可能记错，你帮我核实一下，大概是 AWS 有 40% 的营收来自那些从他们起步、然后成长起来的初创公司。这正是我们想要的类别。你知道，我们知道要说服人们从一个数据库提供商迁移到另一个几乎是不可能的。所以我们只想在第一天就在场，确保他们体验很好，然后确保他们真的不想离开。

The lesson I think that they—I heard a stat and someone was telling me, oh, I might get it wrong, so fact-check me, but something like 40% of AWS's revenue comes from startups that started on them and then scaled up. And that's the category that we want. You know, we know that to convince people to migrate from one database provider to another is near impossible. And so we just want to be there on day one, make sure that they have a great time, and then make sure that they actually don't want to leave.

就像 Stripe 的商业模式。我的意思是，如果你能做到，这基本上是世界上最好的商业模式。就是第一天就抓住他们，然后随着世界上增长最快的公司无限扩展。

Business model as Stripe. I mean, it's basically the best business model on Earth if you can get it. Like, get them on day one and then just scale with the fastest-growing companies in the world infinitely.

你只要这样干一代人，然后你就成了一家世代级的企业。差不多就是这样。对，这对我们来说很难，因为“第一天”是什么样子？它在快速变化。过去，对我来说第一天永远是 YC 的客户。而现在，第一天变成了——好吧，人人都是构建者。他们会用什么？有哪些模态？他们会怎么选数据库？这些都是棘手的问题。

And you just do that for a generation, and then you become a generational business. That's pretty much it. Yeah, it's hard for us because what does day one look like? That's rapidly changing. So in the past, day one for me was always YC customers. And now, day one is becoming like—okay, everyone's a builder. What are they going to use? What are the modalities? How are they going to choose their databases? These are tricky questions.

YC 下一期正在接受申请。你心里有个创业点子吗？到 ycombinator.com/apply 申请。永远不嫌早，填写申请表本身就能让你的想法更上一层楼。好，回到视频。

YC's next batch is now taking applications. Got a startup in you? Apply at ycombinator.com/apply. It's never too early, and filling out the app will level up your idea. Okay, back to the video.

这算是不同的客户群体。就像最初你起步时，数据库并没有面向个人的市场。

It's kind of a different customer segment. Like initially, when you started, there was not an individual market for databases.

就像数据库公司做直接面向消费者的业务，这听起来就像语法错误一样。

Like the idea of a database company being a direct-to-consumer business would have just like sounded like a syntax error.

是的。

Yes.

但现在你实际上有两个细分市场。你有早期阶段的公司，它们有强大的 Stripe 商业模式，但你也在直接向消费者或他们的智能体销售。

But now you actually have two segments. You have the early-stage companies that have this powerful Stripe business model, but you're also selling directly to consumers or to their agents perhaps.

是的，是的，是的。

Yes, yes, yes.

这些用例大概不会像初创公司那样扩展，但他们可能会在余生继续使用你，因为你支撑着运行他们个人财务软件的应用程序，或者你的出租车公司之类的。

And those use cases presumably will not scale the way startups do, but they might just continue to use you for the rest of their lives because you're powering the app that runs their personal finance software or your taxi cab company or something like that.

是的。随着他们生活的发展，他们积累信息，这基本上是我们自身增长的一个指标。所以他们只是存储更多。人们在收集更多数据。这种产消者市场对我们来说本质上是一个良好的增长向量。

Yeah. And as their lives grow and they accumulate information, that is basically an index of our own growth. So they just store more. People are collecting more data. This kind of prosumer market is essentially a good growth vector for us.

那么你如何从那个角度思考个人智能体世界？如果编码智能体想要 Supabase，这很清楚。你认为下一阶段是假设我们都有个人数据库，他们需要做些什么来确保个人智能体容易与之对话和使用吗？这有什么不同，还是基本上一样？

How do you think about the personal agent world through that perspective then? If coding agents want Supabase, it's very clear. Do you think about the next phase as being assuming we all have personal databases, do they need to do anything to make sure that it's easy for personal agents to talk to and use that? Is that any different or is it basically the same thing?

是的，我现在非常看好个人智能体，因为我的 EA 离开了一周，我了解了它们能做什么。现在我只是痴迷于为个人智能体构建的想法。确实，我对我们公司说过，我们代码优先，我们为代码优先的人类构建，但实际上我们是为智能体构建。那些人类很快就不会真正交互了。所以我一直在构建的方式是使用 Dots。我经常和它通电话，聊天说，‘哦，你能做这些更改吗？’我确保正确构建或吃自己的狗粮的启发式方法是，我不看代码，也不允许自己检查架构，除非我想检查它是否真的在正确构建东西。但当我做这些更改时，我只想和它说话，看看它能改变什么。

Yeah, I'm very bullish now on personal agents after a week of my EA being away and me learning exactly what they can do. And now I'm just kind of obsessed with the idea of building for personal agents. And it's true that I said to our company like we code first, like we're building for humans that are doing code first, but actually we're doing it for agents. Those humans pretty soon won't actually be interacting. So the way I've been building is I'm using Dots. I just sit on a phone call with it all the time and I'm chatting to it and saying, 'Oh, can you make these changes?' And my heuristic to make sure I'm building or dogfooding correctly is that I just don't look at the code and I don't allow myself ever to check the architecture except when I want to check that it's actually building things correctly. But when I'm making these changes, I just want to talk to it and see what it can change.

我们中的一员。我们中的一员。

One of us. One of us.

这也很疯狂。它做了一些非凡的事情，感觉像是两年前科幻电影里的东西，但现在我有很多这样的时刻，我简直被震撼了。如果你在 6 个月前向某人展示这个，他们只会认为这是不可能的。

It's crazy as well. It's doing some phenomenal stuff that feels like something out of a sci-fi movie from 2 years ago, but now I'm having so many of these moments that I'm just mind-blown. If you had shown someone this 6 months ago, they would have just thought it was impossible.

是的，我几周前参加了你们的 Supabase Select 会议，顺便说一句，这是一个很棒的会议，我有一个疯狂的想法。我坐在那里，你在谈论 Supabase 推出的所有这些很棒的新功能。我想，你知道，如果我在 Supabase 上构建一个应用，也许我甚至不会关心自己理解所有这些很棒的功能。我想派我的智能体去 Supabase Select 听所有很棒的新功能，并弄清楚如何将它们集成到我的代码库中。

Yeah, I was at your Supabase Select conference a couple of weeks ago, which is an awesome conference by the way, and I had this crazy thought. I was sitting there and you were talking about all these awesome new features that Supabase was putting out. And I was like, you know, if I was building an app on Supabase, maybe I wouldn't even care to understand all these awesome features myself. I want to send my agent to Supabase Select to listen to all the awesome new features and figure out how to integrate them into my codebase.

哦，那太疯狂了。一个真正的虚拟会议。我们应该这样做。一个只为智能体举办的会议。

Oh, that's wild. A truly virtual conference. We should do that. A conference just for agents.

一个智能体的会议。

A conference for agents.

它们会现场出现。

They'll turn up live.

现场。它们可以互相交谈，交换关于 Supabase 现场的提示。哇。

Live. They can like talk to each other, exchange tips on like, you know, Supabase live. Wow.

你可以桥接它。你可以像 AirDrop 一样把 markdown 文件发给每个人，但你必须要在观众席才能拿到。

You can bridge it. You can sort of AirDrop them like the markdown files for everyone, but you have to be there in the audience to get it.

我们叫它 Artifactory。

We'll call it the Artifactory.

哦，天哪，我很喜欢那个。

Oh man, I quite like that.

我们之前谈过，你一直在做的一些事情确实让智能体更好地使用你。比如错误消息，你知道，那个关键见解是什么？还有没有其他一些事情，一方面很明显，另一方面却没人做？

We talked about before this, like there's just a bunch of things that you've been doing that really make agents like you better. Like error messages, you know, what was that key insight? And are there a bunch of other things that you're doing that kind of obvious on the one hand, on the other hand like nobody's doing yet?

是的，所以就像错误消息，是的，我在波士顿的 dbt Labs 活动上，我很惊讶这并不那么明显，但我们真的深入研究了我们的错误消息，这样你可以采用一个用例，比如，哦，我想构建一个应用程序，通常开发人员做错误消息就像 500 之类的，你只知道有问题，但现在你实际上希望智能体自己补救。所以在错误消息中，你实际上是在告诉它们如何修复这个错误，或者去阅读一些文档来修复错误。所以很多只是选择你的用例，然后尝试在这些循环中改进你自己的错误消息，这样你基本上可以一次性完成任何你想做的事情。

Yeah, so like the error messages, yeah, I was at a dbt Labs event in Boston and I was surprised that this wasn't so obvious, but we really dug into our error messages so that you can take a use case like, oh, I want to build an application and generally like developer doing error messages are like 500 something basic and you just know there's a problem, but now you actually want the agents to like remedy themselves. So inside the error messages, you are literally telling them how to fix this error or to go and read some docs to fix the error. So a lot of it is just picking your use case and just trying to do these loops on improving your own error messages so that you can one-shot basically anything that you want.

是的，这有道理。你基本上，你知道，如果智能体使用你，智能体会遇到错误消息。过去，人类必须去谷歌搜索并做一堆事情，但在这里你可以直接说，‘哦，就像你只是，你知道，inode 用完了，所以你可能应该去做这个。’

Yeah, that makes sense. You basically, you know, if you're being used by an agent, the agent will encounter an error message. In the past, a human would have to go Google that and do a bunch of stuff, but here you can just say, 'Oh, like you're just, you know, out of inodes, so you should probably go do this.'

是的，因为赢得智能体有两个阶段。第一个是被模型选中，第二个是如果它们卡住了，你不想让它们切换走，所以保证它们会在某个时候卡住。你只是想给它们补救的工具，所以实际上并不那么复杂。它们现在足够聪明，可以弄清楚很多东西，但你确实必须重新设计很多以前只为人类和开发人员设计的东西，为智能体而设计。

Yeah, because there's like two stages to winning the agents. The first is being picked by the model, and then the second is like if they get stuck, you don't want them to switch away, so guaranteed they'll get stuck at some point. You just want to give them the tools to remedy, so it's not actually that complicated. They're smart enough now that they can figure out a lot of stuff, but you do have to re-engineer a lot of what previously was just for humans and developers for agents.

你们为智能体测试产品吗？比如你们有智能体尝试使用 Supabase，然后报告它们遇到问题的所有地方吗？

Do you test your product for agents? Like do you have agents try to use Supabase and then report all the places where they had trouble?

是的，是的，是的。我们在 Evals 网站上测试了所有不同的模型，看看哪些模型构建得最好，我们内部也有自己的评估套件。我们也评估哪些模型仅用 Postgres 就最好，这对行业来说很有趣，并围绕它构建技能，以便它们可以采用并实际用 Postgres 开发。所以，是的，有很多测试。事实上，现在我们的很多工程努力都投入到测试特定用例上，而不是构建。我们想思考人们如何构建，并围绕这个构建套件。

Yeah, yeah, yeah. All the different models we've got on Evals website where we see which models are building the best, and we've got our own eval suite internally. We eval as well which ones are best just with Postgres, so that's interesting for the industry and build skills around it so that they can adopt them and actually develop with Postgres. So, yeah, it's a lot of testing. And in fact, pretty much now a lot of our engineering effort is going into testing specific use cases rather than the building. We want to think through how people are building and build suites around that.

Gary，当你构建 Gbrain 时，你是在 Supabase 之上构建的。你能谈谈你是如何决定在 Supabase 之上构建的吗？嗯，作为用户感觉如何？

Gary, when you built Gbrain, you built it on top of Supabase. Can you talk about how did you decide to build it on top of Supabase? Well, how was it like for you to be a user?

是的，我想我的意思是 Gbrain 像是我写的一堆东西的提炼。就像我首先做了 Gary's list，那像是智能体式的。它像一个智能体新闻编辑室，我不得不构建一个基本上用于 RAG 的研究系统，以及用于政策和论文的系统，它会阅读关于给定主题的每篇报纸文章，然后尝试综合。

Yeah, I guess I mean Gbrain was like a distillation of a bunch of stuff I wrote. Like I first worked on Gary's list, which was like agentic. It was like an agentic newsroom, and I had to build up like basically a research system for RAG and for like policy and papers, and it would like read every newspaper article about a given topic, and then try to synthesize.

然后我把这个提炼成了 OpenClaw 的开源检索方案。因为当时的想法是，好吧，我原来在 Rails 里做，那就像是一个为智能体服务的富士康工厂。我把每一次 LLM 调用都视为非常昂贵且极其危险。然后当我看到 OpenClaw 时，它就像是……你知道，它就像 Esselen 一样。它完全是一种不同的与 AI 协作的方式，因为那时你会觉得，哦，AI 很聪明。它不会试图伤害你，你应该让它放手去做事，你应该给它数据和能力。

Um and then I took that and then distilled it into open-source retrieval for OpenClaw. Cuz the idea was like, okay, I had it in Rails, and that was kind of like a Foxconn factory for agents. It's like, I treated every LLM call as very expensive and incredibly dangerous. And then when I saw OpenClaw, it was this, you know, it was kind of like Esselen. It was just like a very different way for you to work with AI, cuz then it's like, oh, the AI's smart. It's not trying to hurt you, and then you should just let it do stuff, and you should just give it data and capabilities.

所以我意识到，一百万个 token 实际上是一种不可思议的工作记忆。就像大多数人类只能同时记住七加减二件事，而一百万个 token 基本上相当于三本《哈利·波特》。但即便如此，对智能体来说还是有点局限。比如，在任何给定时刻，把正确的三本《哈利·波特》保持在上下文中，是一个真正的挑战。

And so, I recognized a million tokens is actually an incredible working memory. Like, most human beings only have like seven plus or minus two things that they can keep in their head, whereas like a million tokens is like basically like three Harry Potter books. But then even that is like a little bit limiting for the agents. Like, keeping the correct three Harry Potter books in your context at any given time is a real challenge.

所以感觉中间需要一个东西，用于公司大脑或者我个人的资料。你知道，我有五十万个 markdown 文件，记录了我有过的每一次沟通。所以这个东西确实拥有关于我和我在意的一切的上下文。但然后在任何时刻，智能体可能只需要两页内容之类的。这是一个检索问题。

So, it felt like there was this thing in between for a company brain or like my personal stuff. It's like, you know, I have half a million markdown files on like every communication I've ever had. And so, this thing really does have context on me and everything that I care about. Uh but then at any given moment, like the agent, they might need like two pages or something. It's a retrieval problem.

所以，在那个时候，你知道，在达到 30、40 GB 时，你需要 PG vector，你需要 Postgres，你知道，你可以很容易地在 SQL 中做图数据库。所以这就是 G Brain。它就像是，你知道，RAG 检索加上一个图数据库，从非结构化数据中创建结构。所以它就像是动态模式，有点像我记得的当年的 Palantir。

And so, at that point, you know, at many, I think it's like 30, 40 gigabytes, you need PG vector, you need Postgres, you know, you can do graph databases very easily in SQL. And so, that's what G brain is. It's like, you know, rag retrieval plus like a graph database that creates like structure out of unstructured data. And so, it's like dynamic schemas, like a little bit like what I remember Palantir was back in the day.

是的。然后你知道，整个想法是，当有这样一个中间层时，智能体会强大得多，智能体可以直接说这就是我要找的，然后它会遍历图。它会完成所有的向量搜索，然后回来说，这是一份报告，把它加载到上下文中。所以这就是 G Brain，而 Supabase 就非常直接。就像我需要托管的 Postgres，我需要它能够非常快地支持 50 或 100 GB 或 300 GB 的数据库语料库，它一直很棒，你知道，自动扩展工作得非常好，现在最有趣的是，有了浏览器使用，我现在用 side browser 来做浏览器使用，就像，哦对，OpenClaw 可以进入 Supabase，你知道，使用数据库或你新做的 API，它自己就管理好了。它可以自我管理。

Yeah. And then you know, the entire idea is that agents are much more powerful when there's sort of this intermediate thing where the agent can just say like this is what I'm looking for and it'll traverse the graph. It'll do all of the you know, vector search and it'll come back and say like here here's a report and like load this into context. And so that's what G brain is and then Superbase was just like very straightforward. It's like I need I need hosted Postgres and I need it to be able to support like a 50 or 100 gig or 300 GB you know, database corpus really fast and it's been great like it you know, the auto scaling works really really well and now the funniest thing is like with browser use like I use a side browser now for browser use and it's like oh yeah, like open clock can go into Superbase and you know, use the database or the new APIs you made and it just like managed it itself. It can like self-manage.

你需要提示它去弄清楚数据结构，比如图或者 RAG 之类的吗，还是

Were you having to prompt it to like figure out the data structures like graphs or like rag or anything like that or

我的意思是，有趣的是，我记得进入 Claude Code，直接说，用 Exa 下载并整理你所能找到的每一篇 AI 研究论文。今天我可以使用 Alpha Signal 之类的，但当时就是去找每一篇关于 RAG 和图数据库以及你知道的文本检索的研究，然后针对我的用例进行总结，也就是我如何帮助智能体访问这些东西？然后它做完之后，就会说，哦好吧，这是我们将要采用的方法。

I mean what's funny is like I I remember going into Claude Code and just saying like use Exa and download download and diarize every paper you can find in AI research. Today I could use like Alpha signal or something but it's like go and find every piece of research about rag and graph databases and like you know, text retrieval and then summarize it for like my use case which is like how do I help an agent get access to this stuff? And then after it did that it's like oh okay, well here's a here's how we're going to do it.

是的，前几天晚上我参加了一个关于数据库未来的晚宴，其中一个想法是，数据库本质上是一种数据检索机制，你知道，你可以建立索引，让它们以某些格式搜索数据，比如图或者 RAG，然后其中一个超级有趣的想法是，让智能体循环运行，为搜索创建完美的数据结构。比如，如果你搜索这个数据集这么多次，每次需要 10 秒，那为什么不直接让一个智能体循环运行，创建一种新的索引类型，或者找到最佳索引，让返回时间小于一秒。

Yeah, I was at a dinner the other night on like the future of databases and one idea was like databases essentially are like a retrieval mechanism for data and you know, you can put indexes so that they'll search across the data in certain formats so a graph or or rag and then one of them which was super interesting was just like putting an agent on a loop to create the perfect data structure for the search. Like if you're searching this data set this many times and it's taking 10 seconds, then why don't you just put an agent on a loop to create a new type of index or like find the best index for returning that in less than a second.

动态检索显然会在你拥有非常非常智能的代码生成的那一刻发生。

dynamic retrieval is pretty clearly going to happen the second you have really really smart code gen.

是的。

Yeah.

所以，我的意思是，目前我用它来跑 OpenClaw，所以我用我的生产实例，然后从中提取轨迹，作为评估数据集，然后我有一个单独的仓库叫 G Brain Evals，它有 40 个不同的类别，比如特定类型的检索、检索效率、成本效率、延迟。然后评估最酷的地方在于，你可以直接告诉智能体循环，你知道，就是自动研究和循环，任何能提高你在这个评估上统计数据的东西，就把它发布到 master 上，默认如此。

So, I mean for now, I mean I use it for open claw, so then I use my production instance and then the traces from that I extracted out like a eval data set and then I have a separate repo called G brain evals and it's got 40 different categories of like specific types of retrieval, retrieval efficiency, like cost efficiency, latency. And then the cool thing about evals is then you can just tell the agent like loop, you know, just auto research and loop and anything that improves your stats on this eval like ship it into master on by

默认。

default.

是的。任何你尝试过但没成功的东西，写一个 markdown 文件，解释你认为为什么没成功，然后归档到 G Brain Evals 里。我用 Cappy。你知道，Cappy 有一个 captain 的概念，有点像 Grok bots。所以你可以给它一个上下文，我有一个一直在运行的，就是提出新类别，我还有另一个 captain 一直在运行，就是遍历评估中的所有类别，自动研究，每天尝试找到一种方法来提高我们在内部评估上的分数。

Yeah. And anything that you tried that didn't work, write a markdown file that, you know, explains why you think it didn't work and then file it away in G brain evals. I use Cappy. There's a you know, Cappy has this concept of captains, which is kind of like similar to Grok bots bots. And so you can just give it a context and so I have one that is always going that's like come up with new categories and I have another captain that is always going that's like go through all the categories in evals and auto research and every single day try to find a way to improve our scores on our own internal evals.

嗯。

Mhm.

而且它一直在运行。这是一个非常非常原始的 RSI 版本，针对特定用例，但我想如果你上升一个抽象层次，你知道，你可以做到，不是为我的用例，也就是为 OpenClaw 之类的。它可能只是通用的，比如任何人的检索用例。唯一变化的就是用例、轨迹和数据集。

And that's just going all the time. It's a very very primitive version of RSI for like a particular use case, but I guess if you like go up one level of abstraction, you know, you could do that, you know, not for my use case, which is from for open claw or something. Like it could just be for generically like anyone's retrieval use case. Like the only thing that changes is like the use case and the traces and the data set.

你显然了解评估，并且你可以改进你的软件，因为你在这个框架周围包裹了这些东西。你觉得未来构建者是否必须真正了解这类东西，还是

You obviously know about evals and you can improve your software because you're like wrapping this framework around it. Like do you think that in the future builders will have to actually know about these type of things or

也许不必。我的意思是，老实说，我觉得我们太早期了，你知道，我只是在瞎折腾。而且我喜欢在开源中做。我希望有人看到后说，哦，我可以应用那个。你知道，我觉得我们现在 literally 正在发明如何使用这些东西的基本原语。是的，我记得今年早些时候，读到 Steve Yegge 的工厂之类的，然后无法完全理解。我当时想，哦不，不，我觉得现在我至少拼凑出了一些对我有用的东西。

Maybe not. I mean, honestly, I feel like we're so early that, you know, I'm just messing around. And you know, I like doing it in open source. And someone else, you know, I hope someone looks at that and says like, "Oh, I can apply that." You know, like we're I think we're literally coming up with the primitives on how to use this stuff right now. Yeah, I mean, I remember earlier in the year, you know, reading about Steve Yegge's uh factories and things like that and then not being able to totally figure it out. And I was like, "Oh, no, no, I feel like now I've cobbled together something that works for me at least."

是的。

Yeah.

是的，当我们请到 Boris Turney 时，他对编程未来的愿景更像是你就像和朋友一样和 Claude 聊天，你不需要了解评估和所有这些事情。

Yeah, when we had Boris Turney here, um his vision of the future of coding was more like you just talk to Claude like a friend and you don't have to know about evals and all all this all this stuff.

就像 Claude 能直接搞明白怎么把你脑子里想的东西建出来。

And like Claude just figures out how to build the thing that you have in your mind.

我有个域名 openusd.dev。因为实际上有个文件格式叫 openusd.dev。我收集域名。然后突然 Stripe、Visa 和所有人都宣布了 openusd.dev。我就想，“哦，我得赶紧把这个域名用起来。”但我一直没时间用。所以我就跟它聊天，它把这个开发者侧、生态系统什么的都开发出来了。我就在电脑上跟它聊。然后当我结束一天工作、准备上床睡觉时，我在手机上呼叫它，这听起来有点可疑，但我保证我只是在用它写代码。

I've got this domain name openusd.dev. And it's because there's actually a file format called openusd.dev. And I collect domain names. And then suddenly Stripe and Visa and everyone announced openusd.dev. And I was like, "Oh, I better put this domain to use now." But I never had time to use it. So, I was chatting to the dot and it's developed this developer side with the ecosystem and everything. And I was just chatting to it on my computer. And then as I was winding down my day, I called it on my phone as I'm getting to bed, which sounds sus, but I promise I'm just coding with it.

我就说，“哦，我不喜欢生态系统链接到用例的信息架构。你能给点建议吗？”它做到了。我说，“好，那就实施吧。”20 秒后它说，“哦，好的，我完成了。你要刷新吗？”我刷新了网站。写得非常好。所有东西都链接在一起。它还搞定了 SEO 之类的东西。所以，我没有做传统上必须做的那些经典工作。我只是跟它聊天，真的感觉像是我瞥见了未来的时刻之一。

And I was like saying, "Oh, I don't like the information architecture of how the ecosystem links to the use cases. Can you make some suggestions?" And it did. And I said, "Okay, yeah, implement them." And then 20 seconds later it said, "Oh, yeah, okay, I've done it. Do you want to refresh?" And I refreshed the site. And it was really well written. And everything was linking together. And it's hooked up the SEO type stuff. And so, I haven't done all the classic stuff that I'd have to do traditionally. I've just been chatting to it and it really felt like one of those moments where I saw a glimpse of the future.

我的意思是，你能做到是因为你知道你想建什么。

I mean, you can do that because you know what you want to build.

就是这样，没错。所以我显然有工程背景。所以我大概知道怎么提示它，但我尽量什么都不说。我尽量不把它指向 GitHub 之类的，我试着像普通人或明年的人那样编码，这样我基本上就能吃自己的狗粮 Superbase，知道“哦，一年后它跟 Superbase 配合会顺利吗？”

That's the thing, yeah. So, like I obviously have a background in engineering. So, I kind of know how to prompt it, but I'm trying to not say anything. I'm trying not to point it towards GitHub or anything like that and I'm trying to code like a normie or someone who next year so that I know it's basically so I can dog food Superbase and know that, "Oh, is it going to have a good time with Superbase in 1 year's time?"

是啊，Boris 说“哦，我只是想让它自动完成”挺有趣的。当然，一旦它进入分发阶段，模型实际上已经在权重里有了。它能做到。但在那之前，它需要一些 markdown。你只需要一点点提示，甚至不需要很多提示。非常少量的提示。它就能做到。我的意思是，我认为我们正在进入社会上非常奇怪的时刻，听起来 RSI 真的来了，AI 研究正从 20% 到 80% 的时刻，就发生在这些循环内部，比如 Cloud Coder Codex 或其他地方。与此同时，社会并没有足够多像你这样的人，Paul，知道该建什么以及如何提问。所以现在就像不可阻挡的力量遇到了不可移动的物体。模型的能力呈抛物线增长，而知道如何真正用它对社会产生影响的人数却几乎不变。

Yeah, it's a funny for Boris to say like, "Oh, I just wanted it to do it automatically." And like certainly once it's into distribution, like literally the model has it, you know, in its weights. It can do it. But until then, it needs like some markdown. You just like a little like not even a lot of prompting. Like very small amounts of prompting. It can just do it. I mean, we're entering this very strange moment in society, I think, where like it sounds like RSI really is here and it's like going from like the 20% to 80% moment of like AI research is just happening in sort of these inside these loops inside, you know, Cloud Coder Codex or wherever it is. And then simultaneous to that, like society doesn't really have enough people like you, Paul, who like know what to build and how to ask it. And so, it's like unstoppable force meets immovable object right now. Like the capabilities of the models are like going parabolic and then the number of people who know how to actually have an impact on society with it is like more or less constant.

是啊，而且想要为此感到兴奋。我想这是另一回事。我觉得很多人变得偏执。但我真心对很多人未来的可能性感到兴奋。我之前的很多创业公司都专注于蓝领工人，以及那些我围绕技术包装的人，很多一生都在努力工作的人，你知道吗？我真正喜欢 Superbase 的一点是，我因为这个平台看到了很多创业精神。所以我能看到这样一个未来：实际上会有更多的创业，对很多人来说这是一种相当充实的生活方式。所以我对 2027 年相当期待，只要我们能保持这个节奏，避免一些疯狂的事件，那么对很多人来说应该是一段有趣的旅程。

Yeah, and want to be kind of excited by it. I think that's the other thing. I think a lot of people are getting paranoid. But I'm genuinely excited about what the future could be for a lot of people. A lot of my previous startups were focused on like blue-collar workers and like people who I was wrapping tech around a lot of people who just had to work a lot of their life, you know? And I The thing that I really love about Superbase is that it's like I see a lot of entrepreneurship because of the platform. And so I can see this future where actually there is going to be so much more entrepreneurship and it's actually quite a fulfilling lifestyle for a lot of people. So I'm pretty pumped about 2027 actually as long as we can keep this cadence up and avoid a few crazy events, then it should be a fun ride for a lot of people.

我们正看到这样一个时刻：垂直 SaaS 创始人实际上在 A 轮融资上有点挣扎，你知道，当你规模不足时，比如 500 万到 1000 万美元 ARR，甚至这在历史上是一大笔 ARR。这就像保证能拿到 A 轮，甚至在 9 或 12 个月前。而今天，我不知道。人们似乎不太感兴趣，对于垂直领域，横向和 AI 投资似乎完全绿灯，但垂直 SaaS 的人真的在挣扎。但这也恰好发生在你可以用 10 或 15 人达到 1000 万到 1500 万，通常 2500 万、3000 万、5000 万的时候。所以到那时你非常盈利。

We're seeing this moment where vertical SaaS founders actually are kind of struggling with raising series A's uh you know, when you're subscale like you know, 5 to 10 mil ARR even which like historically is a lot of ARR. Like that's like a guaranteed series A even like 9 or 12 months ago. And today it's like uh I don't know. People are kind of like not really into it um for vertical like horizontal and AI investments like seems like it's uh totally green light, but like vertical SaaS people are really struggling. But it's also sort of happening right at the moment where you can get to 10 to 15 mil you know, often 25, 30 mil, 50 mil with like 10 or 15 people. And so at that point like you're very profitable at that point.

是啊，是啊，是啊。如果你有 15 人，每人 200 万 ARR，利润率 50%，你们都能一年赚一百万。我的意思是，对很多人来说这是惊人的结果。

Yeah, yeah, yeah. With like if you got 15 people, 2 million ARR per head and with 50% margins, you're all making a million a year. I mean, that's an amazing outcome for a lot of people.

是啊。与此同时，我们做了很多办公时间，和那些拿到了金戒指然后因为融不到 A 轮而沮丧的人。所以你知道，我的很多办公时间就是像你可能会回头看这是因祸得福，因为你不需要这笔钱。过去你必须融一亿美元才能达到年收入 3000 万，因为你必须雇所有这些人，现在你不必那样做了。

Yeah. And simultaneous to that, we just do a lot of office hours with people who do like they got the brass ring and then they're bummed out that they can't raise a series A. And so you know, many of my office hours there are just like you're probably going to look back on this as a blessing cuz you don't need the money. Like time was you had to raise like a hundred million dollars to get to like 30 million a year in revenue cuz you just had to hire all these people, and now you don't have to do that.

是啊。是啊，创业公司里有个真实的东西，就像徽章。我的意思是，对我来说也一样，一开始获得融资感觉像是我在取得伟大成就。但实际上，现在不融资在我看来就像是，哦，那才是取得伟大成就。所以，我有点羡慕那些达到那个点的公司。

Yeah. Yeah, there's a real thing in startups, like the badge. I mean, it was the same for me, like to get funding at the start, like feel felt like, you know, I was achieving something great. But actually, you know, to actually not fundraise now would in my mind be like, oh, that's achieving something great. So, like, I'm a little bit envious of these companies that are getting to that point.

是啊，所以我不知道。我的意思是，我们处在一个超级变化的世界，但嗯，你的 Superbase 在所有公司中真的能够乘风破浪。嗯，你对那些试图效仿你所做的人有什么想法吗？

Yeah, so I don't know. I mean, we're it's a super changing world, but um your Superbase of all the companies has really been able to ride high through every single wave. Um do you have any thoughts for people as they try to emulate what you've done?

让我晚上睡得好的一件事是林迪效应。数据库已经存在 50 年了。我认为它们还会存在 50 年，而这是一个在 AI 结束时，无论如何，我相当确定会有数据库。所以，你从中解读出什么随你。就像，呃，你必须选择非常持久的类别，如果你要把整个生命押上去，这基本上就是我在做的，你选择你觉得值得投资的类别。我认为另一个是，事情变化太快，所以我们作为 Superbase 不试图拥有所有答案，但我们只是努力确保我们总是适应得最快。而且，你知道，对我来说，我评估或观察了很多我最喜欢的开发工具，嗯，它们一开始以开发者为中心，然后总是向上移动，变得非常企业化，有点停滞，失去了那种爱。

One thing that helps me sleep well at night is that it's kind of the Lindy effect. Databases have been around for 50 years. I think they'll be around for another 50 years, and this is one that at the end of AI, no matter what, I'm pretty certain they'll be databases. So, parse from that whatever you will. Like, uh you got to choose categories that are very durable, and if you're going to bet your entire life, which is basically what I'm doing, you choose the category that you feel is worth investing in. I think the other one is like, things are changing so fast, so we don't try to have all the answers as Superbase, but we just try to make sure that we're always adapting the fastest. And, you know, for me, I've evaluated or like watched a lot of my favorite dev tools, um they kind of start with the developer focus, and then they always move up, and they become very corporate, and they kind of stagnate a lot, and they lose that that love.

从我们的角度，至少从我的角度来看，像 Stripe 这样的公司，他们有一位真正关心并几乎专注于新构建方式的创始人。所以，如果你在为下一波浪潮构建，你总是需要一只眼睛盯着未来的下一步，而现在这个时间线非常短。事情发展得如此之快，但要超级适应，并确保你的团队有快速转变的能力。这就是我们明年关注的事情。

And from our point of view, at least from my point of view, companies like Stripe, for example, they have this real like someone like a founder who just cares about almost focusing on what the new ways of building are. And so, if you're building for this next wave, you just always need to be one eye on the future of what is next, which is a very short timeline now. Things are moving so fast, but being super adaptable and making sure you've got capacity within your team to shift very fast. That's the thing that we're focusing on for the next year.

我想你可能在那里学到了一些具体的东西？比如因为你已经经历了，你知道，也许你经历了最初的 vibe coding 拐点，然后是 Opus 48 拐点，还有 co-gen。就像当下一波浪潮，希望下一个像许多拐点一样冲击你时，但有没有一些事情你会做得不同，以更好地准备？

I think that you've like maybe there's any specific things you've learned there? Like cuz you've had you know, maybe you had the initial vibe coding inflection point and then the Opus 48 inflection point and just co-gen. Like when the next sort of wave, hopefully next like many inflection point hit for you, but like are there things that you'll you'll do differently to be better prepared for it?

嗯，首先，至少从我们的角度来看，吃自己的狗粮是最重要的事情之一。所以我总是用我们的工具来尝试这些工具，知道，哦，差距在哪里？没有比你自己在自己的工具上卡住更好的反馈了。然后是有能力。所以我们尝试与这个领域的关键玩家合作。这种生态系统玩法真的很有帮助，因为他们给我们反馈他们正在构建的东西。所以，那以及我们尝试在团队中保留一些能力，这也非常困难，但这样当他们有想法时，我们可以试一试。而且很多赌注实际上并没有赢，即使看起来我们在一起合作做一些新的酷东西。我们实际上不知道。它们可能听起来很酷，我们尝试它们，它们有点失败。但那些确实起作用的，最终意味着我们一起乘着这个很棒的浪潮。

Well, first of all, like dog fooding is one of the most important things at least from our perspective. So, I'm always trying these tools with ours to know, oh, where are the gaps? There's no better feedback than you getting stuck on your own tooling. And then having capacity. So, we try to partner with the key players in the space. And this ecosystem play has been really helpful because they give us feedback on what they're building. So, that and we try to leave some capacity within the team, which is also very difficult, but so that when they have ideas, we can give it a shot. And not a lot of the bets actually win even when they seem like we're partnering together to do some new cool thing. We actually don't know. They might sound cool and we try them out and they kind of flop. But the ones that do work then end up meaning that we ride this awesome wave together.

这是一个很好的结束语。Paul，非常感谢，伙计。我们下次见。

That's a great note to end on. Paul, thanks so much, man. We'll see you guys next time.