etn.
00:58:44 · S5, Sam, Angela, Kosta, James Foster
[00:58:44] S5: Yeah. I mean, obviously, I think it's going to be tremendous. Like, if you look at the 175,000,000,000 of spend, it's mostly coding and things that are downstream of coding. So essentially seeing a $175,000,000,000 of latent economic value just in demand for engineering problems. What is the sort of corresponding demand for sales problems, for marketing problems, for financial problems, for research problems? So I think, like, coding should not be the biggest prize, and already it's in the hundreds of billions. Mhmm. So I think the sort of stakes are enormous. You know, to comment on that chart specifically, Gavin Baker had a great take on x. Yeah. [00:59:21] S1: We asked John [00:59:22] S8: directly about this. [00:59:23] S5: Did you really? Yeah. Did he say? [00:59:24] S1: What did he say? [00:59:25] S0: Can't They they ultimately disagreed, but he, like, took his point on board. I mean, like, you know, knows that. Yeah. [00:59:31] S1: But he I don't know, really. [00:59:33] S0: Yeah. Yeah. [00:59:34] S5: I think, look, Yeah. It's, of course, it's hard to know causality. Yeah. And I think it probably is both, right? There's a lot of AI washing from over hiring during COVID, but how can you deny that this technology is driving productivity increases? Just look at the pace of Frontier model releases and how the models themselves run this exponential. Like, they themselves are the canary for this technology's economic impact. [00:59:55] S7: Yeah.
00:59:55 · James Foster, Sam, S5, Kosta
[00:59:55] S7: Yeah. [00:59:55] S1: I think there is obviously a power law with, you know, the tech community kind of adopting AI and increasing productivity, and then especially in The US, and you can see that from that graph. I think, you know, Gavin Baker literally addressed that and said, I can look at that graph and you can't tell me that that isn't, you know, increased productivity from AI. If we do look at the rest of the market and, you know, the marketers and then all the other job functions, but then also all of the other sectors and then also the geographies as well. How long do you think or like what is the time scale for like wide scale adoption? Do you think? Is it five years? Is it ten years? Is it, you know, longer? Like, what's... Where where's your head at head out [01:00:38] S5: on that? I mean, I think most of the startup community, here they're working on models. A small number of true believers who I love are working consumer. But the majority of companies seem like they're working on this problem, which is enterprise transformation. Yes. So I don't think it's gonna happen in the next twelve months, nor do I think it's implied by progress in coding. Because I think you essentially have to redesign every business process. [01:00:57] S1: Yeah. [01:00:58] S0: You do.
01:00:58 · Kosta, S5, Sam
[01:00:58] S0: You do. [01:00:58] S5: Like, why why is FedEx a relatively low margin business? Why are a lot of these operationally complex businesses lower margin? A lot of it is because of the human coordination that's involved. So the stakes of actually getting this right are very high. They have a very rational incentive to do it. It's super complex. Right? We essentially take the last hundred years of management science, throw it in the garbage, and now we've got to kind of rebuild these systems while actually flying the plane and decide what the humans do. Yeah. There's a huge amount of complexity, but there's a huge incentive to do it. And I think five years is a good estimate. [01:01:27] S1: Yeah. You spoke to to Michele yesterday. Yeah. I really enjoyed the conversation. One of the things you asked him was what is a business if you were to if he was to start a business, what would he go and start? And he said, the thing that's most interesting to him is agents or sorry, AI plus and then insert business function. I want to kind of double click on that specifically. Like, is that a huge opportunity? I'm fascinated about it. I think you kind of hinted a little bit in your last answer, but kind of, yeah, just double click on that. I'm really curious, like what you think about that.
01:02:01 · S5, Sam, Kosta
[01:02:01] S5: Well, it's it's funny because you... I mean, x is amazing, but I... We also have to think for ourselves because we've kinda gone through this cycle of first they were rappers. Yes. You know? Now they're domain specific harnesses. Incredible. Yeah. You know? Yeah. Not that different, actually. Look, think when we have intelligence as a primitive, it's sort of like cloud as a primitive. It's like, okay, what are you gonna do with the primitive? Like, here's AWS. What do you do with it? You know? You've gotta, like, build the applications on top of it to actually get the economic value of that intelligence primitive. So that's what I think all of the domain specific work is going to do. Yeah. And so much of the domain knowledge is tacit. Right? Like, what's the difference between a tenured employee and a new hire? What's the difference between an attorney that's been practicing for thirty years versus a new graduate? A lot of it is this unspoken tacit knowledge, this knack for getting things done. And I think making that knowledge explicit and starting to program it into the models, it's what's gonna create differentiation in the applied domain. [01:02:52] S1: Yeah. Think it's why I'm so bullish on Microsoft. We've... I've been talking about Microsoft. I'm very bullish on Microsoft just because... [01:03:00] S0: Yeah. I don't know. I just
01:03:00 · Kosta, Sam, S5, Fabian
[01:03:00] S0: Yeah. I don't know. I just [01:03:01] S1: think the ability of them to build out the AI infrastructure side, but then also they have a throughput into the enterprise, directly into the enterprise. They're the only unique business that's able to have that. And so, I just see that as a huge moat. And then you have copilot, is obviously that kind of like throughput. And I think, yeah, what Satya Nadella is doing is incredible. I don't know what's your what's your take on that. [01:03:26] S5: Now look. I mean, he's on x, and he's had some great takes, [01:03:29] S6: but it's
01:03:29 · Fabian, S5, Sam
[01:03:29] S6: but it's [01:03:29] S5: it's it's very interesting how so many interested parties have been telling us to own intelligence, not rent it. Yeah. You know, Karp, Satya, and others. And I... I'm actually on the other side of that where I think for most companies, there almost is no alpha in knowing everything that they know. Like, if we knew everything about our arch nemesis, could we replicate them? Of course not. You know, if Domino's Pizza knew everything about Papa John's, could they replicate their business? No. They would be pointed to different directions, FedEx and UPS. We see this with banks all the time. Yeah. The entire team will leave JPMorgan and go to Morgan Stanley. Does Morgan Stanley become JPMorgan or substitute them? Of course not. So I think unless you're developing drugs or doing things that are very sort of IP index, I just don't think there's that much alpha in the intelligence of how a company works. I think it's an idiosyncratic combination of distribution plus brand plus product plus know how plus assumptions. So in that world, actually, I think that kind of Microsoft is is relatively well positioned, but I think they may be over telling the story. [01:04:26] S1: Yeah. Oh, interesting.