etn.
00:42:14 · Angela, Kosta
[00:42:14] S8: So Amrit was doing that in pricing. I've got some other folks that are, Val and Investo that are doing it in mortgage. If you look across regulated industries, there's many more systems that are doing that. So that's that's kind of bucket one. Yeah. And then bucket two is once you've even agreed to swap it out, arguably, the even harder part is implementing it. And so mapping the processes on the old system and getting the new system in. And I think we're very early in that gnarly problem, which is part technology, but also very very human. [00:42:42] S0: Interesting. Interesting. You you made some comments on X about agentic finance protocols, particularly with like consumer agents and how problems with KYC, merchants don't know whether this is a fraudulent transaction or actually human behind this. How do we get past that? How do we get because I used to try and game supreme drops back in the day and we would all use bots. And sometimes if someone's Wi Fi was slower than someone else's and you were using a bot, it would actually work in your favor because it was slightly delayed enough to be just crossover into the past. Like the the fastest human alive Yep. But it's it's too slow to be a bot. And it reminded me of that when I saw your tweet about, like, this whole KYC process. How do we how do we move past that? How do we start getting, like, less fraudulent signs around bot activity for consumer products?
00:42:42 · Kosta, Angela
[00:42:42] S0: Interesting. Interesting. You you made some comments on X about agentic finance protocols, particularly with like consumer agents and how problems with KYC, merchants don't know whether this is a fraudulent transaction or actually human behind this. How do we get past that? How do we get because I used to try and game supreme drops back in the day and we would all use bots. And sometimes if someone's Wi Fi was slower than someone else's and you were using a bot, it would actually work in your favor because it was slightly delayed enough to be just crossover into the past. Like the the fastest human alive Yep. But it's it's too slow to be a bot. And it reminded me of that when I saw your tweet about, like, this whole KYC process. How do we how do we move past that? How do we start getting, like, less fraudulent signs around bot activity for consumer products? [00:43:26] S8: Oh, 100%. Yeah. And so I just announced... Sardine, which is one of our fraud companies, just put out a a new protocol which they're calling know your human, which I think is Yes. Pretty pretty... Which is a pretty clever naming. And so so here's the problem. Anyone who's used a personal agent, one of the best use cases is for shopping. Your purchases are blocked. The, you know, official stats aren't out, but anecdotally, you know, 30, 40, 50% of the time. Why? All of this shopping is done through browsers. And then the IDs of the... Sort of IDs of those browsers to merchants look like bots, and all of the modern fraud companies block bots. [00:44:04] S0: Mhmm.
00:44:04 · Kosta, Angela, Sam, Fabian
[00:44:04] S0: Mhmm. [00:44:05] S8: And so how do you know that the bot is actually your muse agent with you behind it? Now, when you adopt you adopt Sardine, they can pass the actual user ID session from the browser along with the agent session, and you can actually know that. And so one of the reasons I think that's a pretty clever solution is the solutions before would work, but they require all the merchants to adopt them, all the agents to adopt them. There's this huge chicken and egg network problem, whereas this is probably one of the first ones I've seen where an agent can adopt it or a merchant can adopt it, and it'll really increase the number of purchases going through. [00:44:42] S1: Yeah. I'm really fascinated by the [00:44:45] S6: kind of [00:44:45] S1: regulated industry in terms of, you know, you mentioned it's now the opportunity cost of not ripping out the old legacy system and putting in the new system has kind of flipped and it's too high. If I'm a founder today and I am operating in a regulated industry or a very legacy industry, that needs to happen. How how does a founder go and start, you know, working on a problem where the rest of the industry is very legacy? It's very hard to change the course of the ship. How do they go and do that? What what are the best founders go and do when when they're faced with that problem?
00:44:45 · Sam, Angela
[00:44:45] S1: regulated industry in terms of, you know, you mentioned it's now the opportunity cost of not ripping out the old legacy system and putting in the new system has kind of flipped and it's too high. If I'm a founder today and I am operating in a regulated industry or a very legacy industry, that needs to happen. How how does a founder go and start, you know, working on a problem where the rest of the industry is very legacy? It's very hard to change the course of the ship. How do they go and do that? What what are the best founders go and do when when they're faced with that problem? [00:45:21] S8: Yeah. So the best founders combination generally have some experience in the space. Although, to be honest, that's changing. If you're an amazing technical mind and you can attract amazing technical talent, I'm now seeing that you can hire the financial services expertise. You can hire the insurance expertise. And so the makeup of the teams are very different. But what you do need to be able to do is find, generally, a CEO of a company that has realized that they are going to be behind. Like, it's not AI that's your competition. It's your competition adopting AI faster than you. Yep. Yep. And they're willing to take a bet to be the first customer of this new AI native platform. Nice. And that's very much happening. Like, example, PennyMac Mortgage Company. Right? They were the first customer of Vesta at scale. Okay. And they now originate mortgages 25 to 40% cheaper than all of their competitors. Looks pretty smart. Right? [00:46:14] S1: Yeah. Yeah. Yeah.
00:46:14 · Sam, Angela, Kosta
[00:46:14] S1: Yeah. Yeah. Yeah. [00:46:15] S8: So it's a great way to get started. [00:46:16] S1: Yeah. Yeah. See. I see. These examples. I love that. That's a great answer. [00:46:18] S0: Within this, there there are two avenues I'd love to go down. You wrote in... I think it's a sixteen z twenty twenty six. This was published in December 2025, and it was basically a swath of predictions around what will happen in the future. There's a quote you have that says, If advancement stopped and AI got no better whatsoever, it would take five years for all of the large companies to adopt it. And so, the frontier intelligence could stop today, and we would have a five year time lag for enterprise CEOs, enterprise companies, can catch up to the frontier intelligence and use that internally. Can you talk to us a little bit about this delta between the frontier and the ways that enterprise companies are actually using AI today? Because we talk about it loads on the show. Some CEOs are always getting told by the board, you need to be using AI, you need a frontier model, you need to be using an Astra or a Sol to do your internal works. But the honesty within that is it's not really true. They need to be turning, putting $1 in and getting $2 back. This was a prediction you made a year ago. We're already seeing it happen today, which is amazing. Talk to us a little bit about this prediction. How can we change that? How can enterprise CEOs adapt their AI strategy for B2B enterprises today?
00:47:20 · Angela, Kosta, Sam
[00:47:20] S8: Yeah. So, I actually think my prediction of five years might have been a little optimistic. It's probably a lot longer than that. So I think what has changed is there are amazing use cases beyond... Like, for instance, customer service with Decagon is working very well, and lots of great use cases there. Right? There's lots of great uses in coding. Where it hasn't gone as fast is just general knowledge work. [00:47:44] S0: Yeah. [00:47:45] S8: But I think there is starting to be very good use cases with, like, you know, my mortgage example. But I think it's combo of one, pick something small and just get working. And so, think now what we're seeing with a lot of the... Like, we've got another company, Factor Labs. And they land... They use... Like, computer use has gotten very good. They can do work on top of your legacy systems. And so they've started with QA. So how do you know that there's... You know, like, you have armies of QA on top of legacy systems? Like, they can go into those point and click around and send you those reports. Right? You land with that. You can expand dramatically. So, I think enterprises are saying, I've got one specific problem. I'm just gonna get going. I'm gonna prove that it works and go from there. [00:48:27] S1: Yep. [00:48:28] S8: But, you quickly go outside of Silicon Valley, to other places in The US, to other places in Europe, and nobody's thinking about AI in the same way. So, I think, from a human perspective, it is still gonna take a while to spread. But we've got enough success cases to point to