00:13:44 · Sebastian Johnson, Zeynep Yavuz-Willson
[00:13:44] S0: Yeah, and that's really interesting. Then you're making this bet on the services and I guess on the end users and the applications in these companies. That's where you're thinking the value is going to be. Where do you think about the rest of the AI stack then? What's your view on, say, the model layer, open or closed source, the chips, the underlying technology, the hardware? You've made these bets sort of at the very... You talk about diffusion, and you talk about people actually using it and adopting it. That seems to be where you're taking your bet. That's where you think there's most of the value to be captured. What does that say... What are your views then on the rest of the stack and the the value that will ultimately be captured by the hardware players, Frontier Labs. [00:14:24] S1: I have some points of views, also watching very closely on how things change. [00:14:29] S0: So
00:14:29 · Sebastian Johnson, Zeynep Yavuz-Willson
[00:14:29] S0: So [00:14:30] S1: I focus on the applied AI layer, right? So I'm thinking more about how we can apply AI, how we can have the trust of the customer. I look at the engagement and how people are using the product. And I believe that relationship is the defensibility. But the rest of the stack is also important. I think open source is a big tailwind for a lot of these application layer companies. We're already seeing model companies move to the application layer, right? I mean, we all work cloud, we work on cloud. So, that is an application. That is how I engage with Anthropic. And I think that will continue to happen, but now the application layer players also have the opportunity to move down the open source layer, and there are a number of companies, I will not name them in our portfolio, outside of our portfolio, that since the December moment, where I would say the costs of AI have really significantly gone up, capabilities have gone up, but costs also have gone up, and open source, I think, is now there's more consensus around open source catching up with a time lag with a lot of the frontier models. A lot of them have a very active open source and moving to our own model strategy. So, don't think we can say yet, oh, open source, like open source is gonna make at wide AI, you know, have the moat and like things open source, it's all gonna work. All the companies that are building out the application layer have to be actively thinking about this today. If they're not thinking about it today, like, that's... I'd say that's a concern.
00:16:05 · Sebastian Johnson, Zeynep Yavuz-Willson
[00:16:05] S0: That has definitely been the most interesting situation in the last nine months, right? Nine months ago, it was like the frontier companies, the labs, they're coming for the application companies, right? Like Anthropic, they're going to come for all the application layer companies. They're going to do the legal use cases, the vibe coding, whatever. Now it's like, actually, these application companies are going to replace the providers of the models themselves with open source. [00:16:25] S1: Exactly. So what does it come down to then? It's about the customer relationship. It's about the trust. Actually, whoever has that engagement layer has, in a way, a leg up. As long as you have the same performance or better performance or same performance, but better trust, the the focus company that's in the application layer and the customer relationship should ultimately win. [00:16:45] S0: And is that true, do you think, for consumer facing applications as well, where there's less of a relationship but you've still got, I guess, like, brand and you've got... Relationship is a bit more ethereal than a one to one customer success relationship.
00:16:45 · Sebastian Johnson, Zeynep Yavuz-Willson
[00:16:45] S0: And is that true, do you think, for consumer facing applications as well, where there's less of a relationship but you've still got, I guess, like, brand and you've got... Relationship is a bit more ethereal than a one to one customer success relationship. [00:16:57] S1: Yes, and that's a good point. It's trickier. For people to switch, consumers can switch a lot easier than businesses. Right? So, I think it's something to think about. But ultimately, at the consumer level too, the brand affinity and people changing people's behaviors, and when people are adapt, they adapt. But they don't like to change that much. You know, developers are different, for example. You know, they will actually be constantly testing new tools and switching. They're kind of like the worst people to sell to, actually. You know? They are constantly trying new things and they will change. [00:17:34] S0: And they want to find the next best thing. [00:17:35] S1: And you know what consumers do? Consumers will say if you're... They'll say... They'll be okay with whatever is good enough. Yeah. But they care about cost. So they will maybe switch because of cost because for them, if funding is good enough, like, it's good enough. [00:17:47] S0: Okay. I guess it goes back to your point on trust. Yeah. It's still about trust. Yeah. But I guess consumer facing applications build trust and loyalty in a different way, I guess, to the sort of [00:17:58] S1: Yes. [00:17:58] S0: Typical b to b or...
00:17:58 · Sebastian Johnson, Zeynep Yavuz-Willson
[00:17:58] S0: Typical b to b or... [00:17:59] S1: And and also, like, if you look at the business side, the other other thing that we always talk about is, like, the data mods. Right? [00:18:04] S0: Yeah.
00:18:05 · Zeynep Yavuz-Willson
[00:18:05] S1: So, and that's another question. Is there really a data moat? I don't know if anyone can today come and say, like, I know for sure there's a data moat, but what I can tell you is that if you look at our AI services companies, whether that's a Legora or that's a Dwelly, or that's an integral in accounting, they're effectively layered where the work gets done. Legora is a platform where the law firm and the legal client is collaborating. Dwelly is a platform where the work gets done, where the tenant is having requests and the real estate agent is responding to it. Same thing with accounting. It's where the work is done on Integral, where the customer is having questions for the accountant and all of that, that is data that's being generated on the platform. So we're talking about effectively this, a reinforcement learning environment that you don't pay for because it's part of your stack. And I think a lot of the applied AI companies, if you look at it in the last two, three years, they were so focused on go to market because it was all about like, I need to build the brand. I need to be the first mover. I need to need to be the trusted brand in legal, trusted brand in real estate. I I need to build that. And now that that has kind of happened in customer support and a few other categories, the next step is, okay, now I need to go deeper in my technology stack. Because why? I have this threat that's coming from potentially the model companies. And two, I have this tailwind of resource, and three, I have all this data that's being generated on my platform. So, now that you have some of that scale and you have that data at scale, it's the right time to actually start thinking about how am I gonna feed that data into my own models and potentially even deliver a better experience for my customers than if I would... If I were to be staying on a closed horse or in a front room model.