00:32:02 · Eric Newcomer, Yoni Rechtman
[00:32:02] S0: you invest around AI? Yeah. Are you in foundation models? [00:32:06] S1: No. No, we're not in any foundation Big mistake. [00:32:09] S0: What? That's a big mistake on some level, alright. Look, on
00:32:12 · Yoni Rechtman
[00:32:12] S1: some level, yeah, we probably should have done it, it's a great trade, we should have done it and sold, and you know, we could have, whatever. We didn't. Sometimes maybe we're a little bit too like honest and principled for our own good, and if we see something we think doesn't make sense, we just don't do it. But no, we're not in any foundation model. Models. So I am at this point much more interested in the like second order effects of AI. So rather than saying like, how do we like max tokens, what happens after we do? Let's just like accept as a general premise that like, yeah, it's like, this is a real thing. It's gonna keep getting better. It's gonna wind up in more parts of the economy, like not without some pain and struggle and strife, but like it's gonna wind up in more parts of the economy. What happens then? And I think to my point on like, okay, do you like improvement in deployment of AI is a really important story? Like, but the most basic version of that is like, well, I'm just going to make the capabilities better. I think that's the wrong approach. I think honestly already at this point, we're on the wrong side of some scaling laws around this. We're like, and frankly, a lot of even feels unnecessary. I think that all of the model development could stop tomorrow and we would have like ten or fifteen or twenty five years of really big important work in front of us. And we have some evidence for that, which is that like, you know, we're still in, we, like the whole economy, is still in the move to the cloud. Right. Like twenty five years later. Right. You know? So this is not gonna happen overnight, but like what are then the things that only make sense or only possible in that world? What are either the types of problems that AI creates and new businesses will need to solve, or the types of opportunities that only become available once like the capabilities have gotten
00:33:59 · Eric Newcomer, Yoni Rechtman
[00:33:59] S0: there So give me some examples. [00:34:00] S1: Sure. So I think throughout like, probably the number one feature of, like, what AI is really good at is inverting any kind of system that relies on, like, scarce production and abundant attention. So, like, things like writing, phone calls, hiring, coding, all of these are things which previously we had pretty, like, limited ability to produce. You can only make so many phone calls. You could only write so many lines of code. And we have entire social systems built on the fact that you have like effectively unlimited bandwidth to like review and process those things that have happened, those things that have been produced. That is very obviously reversed. So like what are some of the consequences of that? Well, like everything is slopped now. Like sure, yes, but specifically, like, you know, even as the amount of code getting generated is going up in large organizations, large companies, velocity is going down because now code review has become this like unbelievable bottleneck, and like all of our system of... [00:35:09] S0: Every AI hater is just gonna be like, oh, they're just, they're pumping out tons of words, but it's not valuable, it's creating problems
00:35:09 · Eric Newcomer, Yoni Rechtman
[00:35:09] S0: Every AI hater is just gonna be like, oh, they're just, they're pumping out tons of words, but it's not valuable, it's creating problems [00:35:16] S1: with When it's like free to produce code, you produce more code, but it's not free to review that code. So like, I think, I mean I think Git is like really, really structurally challenged, and GitHub is, you know, in this weird moment where they're like basically all the good talent is left, and they're in this feature freeze, and GitLab is like sort of perpetually on the verge [00:35:34] S0: Do you have a bet in this race? [00:35:36] S1: Of course. What is it? Yeah. So so we backed this business called Atomic, which rather than trying to be like this layer between Git and, you know, the LMs or something like that, which a lot of people are working on, they want to just be an entirely new, you know, replacement for Git. So it, like, exist, you know, down here, and we've seen this happen before. You used to have centralized version control systems, and then sort of the advent of like open source and specifically Linux, like broke that open, like none of those centralized version control systems could manage it. Linus Torvalds created Git, and we have this like, you know, an unbelievable amount of enterprise value created off of the back of that. Well, that's twenty years ago. It seems unlikely to me that Git is gonna really survive like code gen and LLM assisted, you know, software development for the next twenty years. One
00:35:36 · Yoni Rechtman, Eric Newcomer
[00:35:36] S1: Of course. What is it? Yeah. So so we backed this business called Atomic, which rather than trying to be like this layer between Git and, you know, the LMs or something like that, which a lot of people are working on, they want to just be an entirely new, you know, replacement for Git. So it, like, exist, you know, down here, and we've seen this happen before. You used to have centralized version control systems, and then sort of the advent of like open source and specifically Linux, like broke that open, like none of those centralized version control systems could manage it. Linus Torvalds created Git, and we have this like, you know, an unbelievable amount of enterprise value created off of the back of that. Well, that's twenty years ago. It seems unlikely to me that Git is gonna really survive like code gen and LLM assisted, you know, software development for the next twenty years. One [00:36:38] S0: approach that has taken venture capital by storm, and you guys were early on it but are also passionate, is this roll up idea. Take some old world business, give it AI, make a bunch of money. What is Slow doing in Easy. [00:36:53] S1: Three steps. What is [00:36:55] S0: Slow's approach and what's your view of sort of venture capital
00:36:55 · Eric Newcomer, Yoni Rechtman
[00:36:55] S0: Slow's approach and what's your view of sort of venture capital [00:36:59] S1: Sure, the so cost for this is what we internally refer to as a GBO or growth by buyout. We've been doing this for a long time, since well before I joined the firm. So, you know, the first kind of canonical example of this that we have is our business Metropolis. Metropolis started off as software for parking lots. They eventually wound up raising a $100,000,000, buying a parking lot operator, installing their software, making that business way more profitable. They then raised another, I think, billion and $0.5, bought the biggest public company in their space, took it private, did the same thing. They recently announced that [00:37:32] S0: they raised another, I think, 4,000,000,000. AI for parking lots, just underline that. [00:37:37] S1: AI for parking lots. [00:37:40] S0: I think they It's doing well. [00:37:41] S1: Oh, I mean, they just raised, I think, another $4,000,000,000.04, $5.06, who's to say? [00:37:47] S0: And the idea is just like computer vision and the monitoring is much easier now?