Y Combinator Startup Podcast
00:34:41 · Garry Tan, Paul Copplestone
And then, you know, 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 Supabase 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 a 100 gig or 300 gigabyte, you know, database corpus Yeah. Really fast. And it's been great. Like, 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, OpenClaw can go into Supabase and, you know, use the database or the new APIs you made, and it just like managed it itself. It can like self manage. [00:35:40] S0: Were you having to prompt it to like figure out the data structures, like graphs or or, like, rag or anything like that? Or... [00:35:47] S1: I mean, what's funny is, like, I I remember going into Claude Code and just saying, like, use Exa and download every... Download and diarize every paper you can find in AI research. Today, I could use, like, Alpha Signal or something. [00:36:02] S0: Yeah. [00:36:02] S1: But it's like, go and find every piece of research about Rag [00:36:06] S0: Yeah.
00:36:06 · Paul Copplestone, Garry Tan
[00:36:06] S0: Yeah. [00:36:06] S1: 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 here's how we're gonna do it. [00:36:21] S0: 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 Oh, 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 reg. And then one of them, 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 dataset this many times and it's taking ten 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?
00:36:58 · Garry Tan
[00:36:58] S1: I mean, dynamic retrieval is pretty clearly going to happen the second you have really, really smart cogen. Yeah. So, I mean, for now... I mean, I use it for OpenClaw, so then I use my production instance, and then the traces from that, I extracted out, like, an eval dataset. And then I have a separate repo called gbrain 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 default. 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 Capi. There's a... You know, Capi has this concept of captains, which is kinda like similar to Grockbots bots. Uh-huh. 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's 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. And that's just going all the time. It's a very, very primitive version of RSI for, 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 OpenClaw 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 dataset.
00:38:42 · Paul Copplestone, Garry Tan, S3
[00:38:42] S0: You obviously know about eval's, 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? [00:38:54] S1: Or... Maybe not. I mean, honestly, I feel like we're so early that, you know, I'm just messing around. Yeah. 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 software factories and things like that, and then not being able to totally figure it out. And it's like, oh, no, I feel like now I've cobbled together something that works for me at least. [00:39:24] S0: Yeah. [00:39:25] S3: Yeah. When we had Boris Cherny here, 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. And, like, Claude just, like, figures out how to build the thing that you have in your mind.
00:39:25 · S3, Paul Copplestone, Garry Tan
[00:39:25] S3: Yeah. When we had Boris Cherny here, 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. And, like, Claude just, like, figures out how to build the thing that you have in your mind. [00:39:41] S0: I've got this domain name, openusd.dev, and it's because there's actually a file format called open usd, and I collect domain names. And then suddenly, Stripe and Visa and everyone Yeah. Announced open usd. And I was like, oh, I better put this domain to use now. But I never had time to to use it. So I was chatting to the dot and it's like developed this, like, developer site with the ecosystem and everything. And I was just chatting to it on my computer and then like, as I was winding down my day, I called it on my phone, like, as I'm getting to bed, which sounds sus, but I'm... I promise I'm [00:40:17] S1: just Checking in.
00:40:17 · Garry Tan, Paul Copplestone
[00:40:17] S1: just Checking in. [00:40:18] S0: Just coding with it. And I was, like, saying, I don't like the information architecture of how, like, the ecosystem links to the the use cases. Can you make some suggestions? And it did. And I said, okay. Yeah. Implement them. And then twenty seconds later, it said, oh, yeah. Okay. I've done it. Do you wanna 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 was... I I sort of saw a glimpse of the future. [00:40:56] S1: I mean, you can do that because you know what you want to build. [00:40:59] S0: 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, like, point it towards GitHub or anything like that. And I'm trying to code like a like a normie or someone will do next year so that I know it's basically so I can dog food Supabase and know that, oh, is it going to have a good time with Supabase in one year's time?