Ex-Google and a16z founder now building payments for AI agents

The universe of agent payments today is very, very small. We're in the first inning, and we're talking a 100, a thousand x growth in the next five years. The old business model for the Internet was distraction, that you could monetize off of a user getting distracted by banner ads or display ads. In the age of agents, agents will use a different search engine. They don't want HTML websites. They don't want a browser, and they certainly will not get distracted along the way.

There's this status code that exists in the browser in networking that says payment required, but there was no way at the time in 1997 to actually pay, or much less to pay a micro transaction if you wanna pay a cent or 2¢. Fast forward twenty five ish years, we now have globally distributed ledgers that can do subcent transaction fees to anywhere in the world in under a second. Why agent cash is called agent cash is because cash is the closest equivalent to a stablecoin that exists.

If you're a merchant transacting with hundreds of thousands of agents and doing millions of dollars of volume a year, stablecoins are really compelling. There's a world of open agenda commerce where every checkout is available to every agent.

Hey. It's Drew. Quick thank you to Altitude for making this episode possible. Altitude is rebuilding business banking for teams on the frontier. They offer the full stack for the modern CFO and founder, a beautifully designed platform for payouts, bill pay, corporate cards, enterprise controls, and so much more. Move money with speed, yield, and total control all from your stablecoin balance. Operate at Altitude. If your business is on the frontier, your finances should be too. You can learn more at altitude.xyz.

We have the illustrious Sam Ragsdale from Merit Systems. Excited to dive in today. Great to meet you. I know we have St. Louis in common. I don't know if I said that. I grew up there. Nice. I was almost a WashU guy. Nice. Okay. So excited to dive into your story. Really pumped to hear about what you're building in the world of agentic commerce. Oh, getting buzzy. But take us on the journey a little bit. Like,

said earlier, you said you've been a builder from the beginning. Yep. You've always been a builder. Yep. Like, tell me about that. What did that look like when you were actually young? And then how has that kind of evolved the course of your career? Yeah. So I've I've always loved building stuff since I was very, very young. I, you know, it was the obvious stuff like Legos and RC cars and all that stuff. I got into robotics like most kids at my age, and then got into three d printing in high school.

I really like building physical stuff. I built an electric longboard before the Boosted Board came out, upgraded that over the course of college. Eventually, it it went 42 miles an hour. The Boosted Board topped out at 22, so it was a... Yeah. Did you ever have any... Yes. It was extremely dangerous. It's a terrible idea. I still have it. The battery is pretty cooked. I ended up building the battery myself because at the time, the battery technology wasn't great. You couldn't import great batteries.

But, yeah, I loved building stuff. At some point along the way in college, I realized there was no Andorol at the time. There wasn't this great rebirth of American manufacturing. And so there was not great... There weren't very fun mechanical engineering jobs. It was looking like I would probably be spending my twenties in a Nevada desert working on landing gear. And so instead, I decided to get into software, which I enjoyed. You get a lot of leverage in software. Can build more stuff as an individual.

And then got my first internship at Amazon Web Services with my cofounder and CTO, Ryan. We lived together there in a tiny little shack in Seattle. He ended up going back to Amazon Web Services. I went to Google, worked on ads, which I'm sure we'll talk about in a second, but is the first incentive model for the Internet, the one that scaled it for the entirety of my adult life, and I think is being set to be disrupted right now. And then got into crypto in 2021 doing MEV.

I went over to a sixteen z to work on their venture fund. I did sort of all the technical investing stuff deep in the stack, l ones, l twos, zero knowledge virtual machine scaling technology. Maybe we'll get into that as well. And then had the knack to get back after it and start building again and raised about two years ago now. Amazing. Okay. Yeah. Double clicking on a couple of the threads there. Love that you lived in a tiny apartment

as a college kid with your cofounder. That's a great origin moment. But then talk a little bit about the... You said what scaled the original Internet, the incentive structure that scaled the original Internet. Yeah. Web two, land Yeah. Google Ads. Like, lay the foundation if you're saying this moment is ripe for disruption of that style of Internet Yeah. And Merit Systems is doing some of that disruption Yeah. Give us a history lesson on Sure.

On the old way of doing things in terms of the incentives and what is scaled and the sort of incumbent marketplace there. Yeah. So to go back to the beginning of time or at least time as I know it, I was born in 1997.

Tim Berners Lee, Marc Andreessen, a bunch of others were working on the early Internet. And fiber had just been built out, and the infrastructure was being built out, but it was still extremely expensive. The browser was just coming online, and it was a little bit unclear why the open Internet should exist, why somebody should put up a server and then have any user on Earth be able to hit that thing because servers were extremely expensive at the time. Like, we're talking $500,000 a year to run.

There was no open source software. You had to license the software. There was no data centers or cloud as we know it, you know, like on demand pricing. So you had to get real estate and put a box in it. And then the networking was extremely expensive too. And so the idea that anybody in the world and and maybe everybody in the world, 7,000,000,000 people, would be able to access all the other stuff in the world was ridiculous. That would be extremely expensive. Like, why would you pay for that?

And so in the early days of the Internet, they put an error status code, which means, like, I'm not returning a good request called 402, which said payment required. Basically, like, if you, as a client, wanna hit my server that shows my recipe book or whatever, please pay me. But, unfortunately, there was no payment technology at the time that could support micro transactions or actually macro transactions either. They were still figuring out credit cards on the Internet too.

So they moved on and, like, didn't worry about it. Google search happened a few years later, and then you could type stuff in and it would give you all of the web pages on the Internet. And then, of course, they figured out to put ads in there as well and created one of the best business models of all time, which is that if you have the entry point to all of the information and goods and services in the world, you can monetize that choke point extremely well.

And so that's a $300,000,000,000 a year choke point or something like that. It's like the ultimate... You mentioned Munger earlier. I'm thinking about, Warren Buffett's obsession with toll bridges. It's the ultimate toll bridge. Yes. They very much would not want you to refer to it that way, and they very much would not want you to refer to it as monopoly, which I've done neither of. But

All loved alphabet, and they wouldn't want you to refer to it that way. No. And they certainly wouldn't want you to refer to that way in accord.

But it's an extremely good business model. $300,000,000,000 a year, I worked on Google Ads on the advertiser platform. And it seems to keep scaling in a way that nobody would have expected. Like in 2016, I was reading the old Internet trends report, which VCs at the time were really excited about. And I think that Google's revenue... Or I think all of Internet advertising revenue had hit $60,000,000,000

that year. And the report was like, I just don't know if it's gonna scale anymore. Like, it's got a plateau at some point. Right. And, of course, it didn't. It it continued to grow, one of the largest business models ever. Yeah. It it makes me think of the first trillion dollar valuation. It's like, oh, there can't... You know, what does that even mean? Yeah. Now we've got a trillion dollar man. Yeah. There will be many four minute mile moments. Yep. But that's the... You saw up close and personal. You took us...

I love you. Took us to the evolution of, like, the early Internet. There's an X402 pin that we'll have to put in Yeah. In there. You saw this incredible business model that has scaled this version Yeah. Of the Internet and is still in existence, although, of course, is being challenged as Google native interaction is at risk Yeah. If you will, in 2026.

I would say actually Google is doing a great job adjusting. I wouldn't say necessarily Google will be in a bad spot, but maybe ads as a business model. Yeah. Yeah. Okay. So it'll be interesting to see how it plays out. I think the last piece there is that in summary, condensing a lot of complicated topics. The old business model for the Internet was distraction, that you could monetize off of a user getting distracted by banner ads or display ads or something like that.

And so if you're looking for a recipe, you might see, you know, those shoes that you saw three days ago, and that merchant that wants to sell you those shoes, that advertiser is willing to pay lots of money to monetize that interaction. And the nice thing is also it's a really low friction business model as well. There's three parties involved. You have the consumer that's looking for maybe a recipe.

You have the publisher that is hosting the recipe, and then you have the advertiser, and you have Google in the middle. And the money only needs to flow between the advertiser to Google to the publisher, and the user doesn't need to show up with anything. And so it's extremely low friction. It's scaled. It's scaled really, really well. So old business model for the Internet was distraction.

And maybe the stack, the way that I would think about the Internet stack that has existed my entire adult life is you have a browser that shows generic HTML web pages, and then you have a search engine that allows you to find all of the web pages and go through them. And then when you click through to one, there will be a display ad that if there is no relationship between... If the publisher is not monetizing directly through the user. So, like, Amazon, for example, they sell to the user.

And, basically, that entire stack doesn't make sense in the age of agents. In the age of agents, agents will probably use a different search engine. They probably don't want HTML websites. They probably don't want a browser, and they certainly will distracted along the way. And so that incentive model is what created the free and open Internet as we know it.

The 10,000,000,000,000 token dataset that humanity put all of their information online that ironically eventually created the LLM, that incentive model is now completely bricked because people want to use agents for all of their research, discovery, and increasingly commerce and execution. Same incentive model that social media companies have used. Do you see it as fundamentally the exact same model in terms of these banners and ads that's just in a feed or a little bit different there? It's hard to generalize.

I think the free and open Internet incentive model is the one that I've described. The walled garden one is actually quite different. Walled gardens being Facebook, Instagram, TikTok, Snapchat, etcetera. Basically, you're within an authenticated context, you log in to that site, and they can personalize. I also think that most of the walled gardens are focused on entertainment, people that want to scroll to find something to do, to find something to wear, etcetera.

And so the monetization channels are slightly different there. And I don't think people are going to use LLMs agents as much for leisure. So the old garden stuff. I think those business models will be safer. They're a little safer. Okay. So let's go to the one that's maybe a little less safe. Mhmm. The former mothership. Mhmm. Who you said is adapting well. Yep. And now you have the entry of LLMs. A pretty fast entry in terms of user adoption. Wildly fast. Yep. And when I think about this introduction of ChatGPT

into life and the way that it basically roared past the standard technology adoption curves that we would expect Yep. Across the world, across ages, and various demographics. My mother-in-law uses ChatGPT with the best of them. You know? We have all kinds of people who are adopting it. And with this sort of mass adoption, as you said, people are seeking out information in different ways. There's a foot in the door. There's an opening towards a more agentic existence. Yep.

Where do you see opportunity there such that you said, oh, we've gotta go chase this? And was this...

You were leaving a 16 z saying that's what we gotta chase, or it's kind of a... This new... It's been a few pivots since you've sort of gotten into playing here. Yeah. So two very different questions. Yeah. I know. I gave you a couple. Take me on the journey. Yeah. So what I would say is that the entirety of the stack doesn't make sense anymore. Like, the way that a user goes about executing information, which is you pull up your phone, you pull up a browser, you go to Google.

Google's probably purchased the default search bar in there for $20,000,000,000 as well. So you type into the default search bar, it goes to Google, and you click through to get your information and get progressively deeper into the Internet through a series of UIs. Like, I don't think that makes a ton of sense anymore. Increasingly, you'll do your execution through an agent. And humans are gonna go through that in a very human way, which is Yeah. Very visual.

You know, there are certain ways that we would go through it in sequence that maybe is not how an agent would approach it. Yeah. Exactly. Like, I think the UIs were created to maximize the amount of information that is transmitted from your device to your brain so that you can make a decision back and forth repeatedly. If you have an agent that can do a lot of that for you, it can take the original intent and then go through all of that, then the UI doesn't make a ton of sense.

The screen doesn't make The website doesn't make sense. And so the way that we've built our entire digital economy probably gets rebuilt in some form. Mhmm. I don't know exactly what it looks like. Correct me if I'm wrong here, but I'm rewinding to the early days of personal computation Yeah. And thinking about moving from command line interfaces to graphical interfaces that were more sort of human friendly. Like Steve Wozniak figuring this out in a garage. Yeah. Exactly. Now we're sort of going back potentially?

Maybe. Yeah. I mean, I think that's a great example. Right? Like, if you go from... Which we're going back to. If you go from the CLI and you have just flat text and monospace font, there's a lot more information density in a well made UI with, you know, big colors and bold and buttons and, like, you know, when you click something, you get through to a new dimension. And really, like, in the abstract, what you're getting is better transmission of the options on the page and what to do next to the brain.

So you can go back and forth, and you can get a task done much faster. So now we're back in the stage of agents and now everything's marked down, everything's text again. It doesn't have to be pretty and simple and sophisticated. Maybe there's still elements of a fundamental principle of design there that's helpful, but it's also just a different brain. You specify your intent and then that agent goes and executes it on your behalf Yeah. As long as it can. And maybe this will be the transition moment.

A lot of the things that your intents, the things that you wanna execute on the Internet actually require money even today. And so there are the things that you currently research and get information on. Maybe you go to Wikipedia. Maybe you go to a recipe website. Maybe you get the sports scores that are currently monetized through banner ads Mhmm. The ability to distract the human while you're doing it.

If an agent is doing that on your behalf, you know, it wants to find the score of the football game this weekend, it won't get distracted by the advertisement for shoes on the side of the page. And then similarly, there's lots of stuff you do on the Internet that requires payment, that requires a credit card. Maybe you wanna do something simple like a checkout or maybe you wanna book a flight. Those are things that currently are not built for agents in any capacity, and so an agent just can't do it for you.

And it probably can't even, like, search to figure out all the options for shoes that are available to you because that wasn't built for bots and actually was built to block bots. And then there's harder things that, like, a hedge fund might wanna do through an agent, like, find me satellite data or drone data on these data centers so that I can speculate on the build out that's going on. These are all things that intuitively you think you should be able to do through an agent.

You know, you imagine an agent and imagine it's AI, so it should be able to do anything. But there's some fundamental barriers to the way that we built everything over the past twenty five years that blocks bots or assumes ads will be able to monetize that no longer make a ton of sense. I love it. So I feel like we've kind of teed it up. Right? Like, it's like, oh, here's the old way. Here's the reasons why it is changing. Yep. How are you playing in this game?

For sure. So we're starting with very simple primitives. I can't say that I know exactly how it's gonna play out until we're kind of experimenting at the edge. Maybe what makes you... First, we'll say just like, why this? Yeah. For you. You could do many things. You said you're a builder. You've been with some big, amazing groups doing cool stuff at Google and a sixteen z, and we could dive into lots of rabbit holes here. We're in an awesome era. You can work on many things.

This is what you're choosing to work on Yep. And focus on and hopefully win at.

Why this problem? We've gotten to a place in the world where very large tech companies and other producers have immense technology that they can use against the consumer. And so examples of this are that through the walled gardens, have algorithms that are written by tens of thousands of brilliant engineers to maximize your screen time, to get you very excited and addicted to your phone.

And then you have another group of 10,000 engineers that's meant to serve the perfect ads so that you click through and buy that thing. And it's kind of like an asymmetric advantage. Like, these are very, very good algorithms. They're... You know, before we had the LLM, the majority of the spend was on making sure that you would click a button and buy those shoes. And similarly, it's an incentive model that shapes our civilization, the world around us.

For the first time, everybody has access to frontier LLMs, roughly. You know, there's like the do you have fable or not battle that is happening. But roughly speaking, thanks to capitalism, everybody in the world has access to the best silicon intelligence that exists. Yeah. Because that is the most profitable thing to do is to distribute the technology as widely as possible. Maybe similar to the old Internet that it was crazy to distribute freely. Yeah.

It's like the common question people ask about, isn't it odd that the richest people in the world use the same phone as you? Right? Then the answer is, like, thanks to capitalism, the most profitable thing to do with an amazing technology is distribute it to everybody so everybody has access to it. The other tailwind that is happening is thanks to open source AI and great work from great technologists all over the world and, you know, a massive bubble in spending. The price of tokens is coming down dramatically.

And so it looks like you'll have frontier LLMs that get better every year and get 10 x cheaper every year. So...... Every month. Better every month. Yeah. And and, you know, hopefully, 10 x better and 10 x cheaper every year. And if that continues to happen, then, like, we'll be in a pretty awesome spot. But I think what this means, what you'll have the ability to do for the first time in my adult life is that every user will have a user aligned AI.

So today, a lot of your goods and services are provided by companies, corporations that are not fully aligned with you. They're aligned with the person that's paying them, which is the advertiser or the publisher because that is truly their customer. But for the first time, every human, every consumer could have access to a user aligned AI, one that cares about their preferences, their goals, their budgetary needs, how they wanna show up in the world.

And you can think about it as like a cybernetic backpack for consumerism, for the commercial world that we're headed into. And in a world where just the big companies get access to this, like we had through sort of the advertising era Yeah. You end up with a bad asymmetry, and I think you end up with a less happy population. But in a world where everybody has access to a user aligned AI, I think you end up with a much more interesting world, and it helps every consumer becomes more educated, more effective.

And I think it's a much more interesting way that this plays out. Totally. And so about two years ago, you launched Merit Systems. Yep. Throwback to the roommate in Seattle. Yep. You're together as CEO. Current CTO? Yeah. So Ryan and I lived together in San Francisco. While I was at a 16 z, he was a a fund called Blockchain Capital. I'm sure many of the viewers know as well. We built lots of projects on the side together, and, you know, we love building stuff together.

And then about two years ago, decided it was time to do the thing, start a company. At the time, we were focused on open source incentives, so we wanted to rewrite how people built in open source. We effectively wanted to create what looked like a cap table for every open source repository so that you could point money at a code repository, it would just get better over time, and all of those people would get paid based on their contributions, based on their merit systems. We spent about a year on that.

Unfortunately, by the time we got it out, when we started it, Cursor had just came out. So we were using it. I think that means it's like... I guess it launched with GPT four, which was like the first agentic ish LLM two years ago, which is such a short time period for how much has happened. It's crazy. We just started using Cursor, but it was mostly, like, you know, make a code edit here and there. You would never use it agentically.

You'd use it for, like, rewrite this for loop to make it smaller or put good comments in or something like that. By the time we launched it and got it out there, I think we're on, like, g p t four five at this point, and the rules of open source had just been rewritten overnight. The ability for people to freely contribute to anything and accept code contributions was really messed up by the fact that it was really hard to tell if contributions were high quality or a swap.

And so open source kind of got railroaded overnight. The whole incentive model collapsed. So then we went back to the drawing boards, and we looked at what we were good at. And we were good at agents and commerce, and it looked like there is a subset of the market that would be particularly applicable to stablecoins, that being sort of this microtransaction category. And so we looked around at sort of the various... All of the prior art in the space where people are working on.

The Coinbase team had been working on this thing called x four zero two since the spring, about six months earlier. X four zero two, it's very well named. Good meme. Back to the, Tim Berners Lee, Marc Andreessen thing earlier, there's this status code that exists in the browser in networking that says payment required, but there was no way at the time in 1997 to actually pay or much less to pay a microtransaction if you wanna pay a cent or 2¢.

Fast forward twenty five ish years, we now have globally distributed ledgers that can do subcent transaction fees to anywhere in the world in under a second. And so all of a sudden, it made sense. Coinbase rolled this thing out called x four zero two, allows you to pay a microtransaction for some API resource. But at the time, there was about 2,000 transactions over six months that had gone across it. And so we asked ourselves, like, what is the biggest thing we could do to push this space forward?

And so we launched x four zero two scan, which was a registry of all of the x four zero two resources, all the stuff that agents could buy. And at the time, there was about a 100 things on it. And then very quickly, it blew up at 300,000 users in month one, transaction volumes hitting tens of millions of dollars, lots of people registering new resources, lots of new developer demand.

The caveat that I will make to everybody that has context on x $40.02, and I admit every time very clearly, a lot of the early boom there was caused by speculation. People were launching meme coins onto x $40.02 scan and x $40.02 more broadly, which is not a particularly good way to launch a meme coin. But it did bring in an immense amount of eyeballs. People were really, really excited about it. Yeah. So we doubled down. We saw Sometimes innovation starts as chaotic play. Yes.

Or chaotic speculation.

Yeah. And then it it grows up a little bit over time. Yeah. The way that my old boss, Chris Dixon, used to refer to this is sort of the innovation price cycle, which is that the first thing that happens usually is there's an uptick in interest, and then there's an uptick in price. Yeah. And that brings in lots of developers, which happens slower over some period of time. It's sort of a lagging indicator.

Often, the price collapses at that point, but the developers stay around and build the real things that lead to people getting very excited about it. So brought in lots of developers. We leaned in. We made a lot of improvements to x four zero two scan. We started the early version of agent cache at that point. So x four zero two scan is sort of supply side. You can think about it kinda like the Google search of this world. Yeah. As an agent, you can find which goods and services, that you'd like to pay for.

As a publisher or merchant, you can publish goods and services to agents. And do you interact with it directly, or do you go through an LLM? There is a UI, and then it has a search that's meant for an agent for your Claude or your ChadGPT. So we had kind of the supply side. We built that out. And I think as of today, there's around 50,000 different resources on there. You should think about it as an app store in the early days of the Internet where it's sort of like there's a power...

There's a short tail of the high quality resources, then there's a lot of additional stuff. But people are kind of, like, playing around with it, toying around with it. There's a lot of developers coming in. Shortly after that, we launched AgentCache One. AgentCache is a wallet that you give to your agent, and it can pay for anything on x four zero two or MPP. And so data, goods, services, custom merch. You can buy a phone number on there. You can send SMS. You can buy an email inbox. You can rent GPUs.

You can host a website. The way we think about it is one balance access to every API on the Internet. And that's a little bit aspirational at this point, but we're slowly working towards that. Shortly thereafter, we launched MPP scan. MPP is the equivalent of x four zero two by the Tempo and Stripe guys. And agent cache now supports MPP as well. So two protocols, and it supports USDC on base, Solana, and Tempo across Love x four zero two and MPP. Yeah.

And we've sort of doubled down and been focused on that since trying to grow the market share there, bring new developers in to build interesting things. Yeah.

So flashing back just a couple of months, I'm at a very stable conference in San Francisco. And agentic, insert word, has started to get thrown around a lot. Agentic commerce, agentic finance, agentic economies. Yep. But there's also this question of kind of what's real, what's not. Sounds like some of this is is deciphering that a little bit. Yeah. Take me through maybe it's Agent Cash. What's real right now?

You said some of it's aspirational, but what's real in terms of I could go out and use it, or maybe it's for the individual consumer. Is it more institutional? Yeah. How can we interact with this technology now and maybe where are you going to? The thing that Agent Cash excels at today

is for people that are... Well, you can be a sophisticated agent user and you want to buy hosting. You want your agent to do more. Right? You want it to be able to pay for stuff. You want it to be able to buy stuff in the real world, or maybe you wanna buy data to do sales and go to market workflows. In general, one balance access to every API on the Internet.

The way that I think about it is two years ago, there were about 50,000,000 people around the globe that knew what an API is and have ever gotten an API key and and spent money over using an API. All of a sudden, we're at, like, a billion. Right? Two years later, there's a billion people that are able to put into an agent. I would like to find satellite imagery of my house and then hit enter. And, of course, there's an API for that. There's 10 different APIs for satellite data.

There's APIs for everything, but there was no ability for my fiance, for example, to, get access to those APIs. And now you can type into Claude, like, you know, I would like to do this thing. Claude can't unfortunately go and pay for that thing. And so it wanna give it a way to do machine payments. You wanna give it a wallet. Now it can interact with that service directly. It can pay them 10¢, get access to that satellite photo, and you can move on with the rest of your query.

And what's true right now in terms of giving Claude or giving GPT your wallet Yeah. A wallet, a way to transact? Yep. Some of that's happening. Yeah. So today, via AgentCash, you can pay it for anything on x four zero two and MPP. That is a subset of all of the things that you could possibly pay for for sure. Yeah. About 50,000 different resources on there. You can do most of the Frontier image and video generation by an email inbox, phone number, etcetera. There's a bunch of weird stuff there.

Ironically, what you can't do very well is get through normal checkouts. So if you wanna buy a Starbucks coffee for pickup, you can't do that on x four zero two or MPP today. And it's very hard for your agent to do that. There's been some efforts It's a classic example. Right? Can my agent buy me a coffee? Yeah. Funny enough, it's like, I don't know if I really want an agent to buy me a coffee. Yeah. So I'd rather have it buy me something else. But you're saying it's difficult. You're stuck there. Yeah.

You require the merchants to come online themselves. There's sort of this weird thing that happened that, all of the checkout boxes on the web are actually extremely sensitive because of the way that credit cards work. Yeah. The way that credit cards work is that the merchants take all of the risk. If you as a user come and you buy your coffee, you swipe your coffee, it probably doesn't settle to the merchant for about thirty days. So that merchant's basically extending credit to you for thirty days.

You then, as the consumer, can go do a chargeback. You claim that the, services provided by the merchants were bad or fraudulent, or you claim that you actually didn't do the payment. Right? It was some other guy. And so as a result of that, the merchant takes all of the risk. There's this big concern in the checkout that you need to make sure that with as high confidence as possible that the actual person is doing the checkout.

And so there's all of this bot protection put into the checkout forms very subtly that make it really, really hard for a ChatGPT with computer use or a Claude with computer use to get through that actual checkout. And so ironically, the things that are the hardest to buy through... Agentically, through AgingCash or X four zero two or MPP are the normal things that humans are used to buying all the time, all of the ecommerce stuff, all the commodities, the goods and services. Yeah.

I think that that will shift over the next year. Stripe particularly is doing some really great work here. I'm making more of these checkouts accessible

to agents. Yeah. Okay. So we talked about the old business model and old incentive structure and all of that and how an agent will interact with the Internet and with information broadly very differently. I'm curious. You said you saw one of the best business models ever at play. Yep. Sometimes people talk about this world of digital assets having some really compelling business models. You know, laughably good, laughably high profit. White meaning Circle and Tether. Yeah. Yeah. These these are who I'm thinking of.

Yeah.

What's the business model here? How does this make money or is the point like, hey, we're getting in the game, we're winning the distribution now and then we'll monetize later? I think that if you're in the transaction flow of the way that agents pay, the universe of agent payments today is very, very small. We're in the first inning. But if you extrapolate the ChadGBT adoption chart onto that, you know, we're talking a 100, a thousand x growth in the next five years. Yeah.

So I think there's a great business model there. But I also think, like, it creates... To your earlier comment about the previous business model for the Internet, I think it creates potentially new avenues for people to monetize, potentially new businesses that couldn't exist before. Yeah. And so, you know, over the last twenty years, news companies have had, like, a very adversarial relationship with walled gardens, for example, with Facebook, with Twitter, with Instagram, etcetera, because people put up headlines.

People will scroll past that, and they'll get the legitimacy of the article. They'll never click through it. They'll never go through the paywall. And those companies are monetizing much and much much worse on their advertising. But, potentially, you could sell that content, that original op ed or whatever in the morning to an agent that pays a cent for 50 different ones, assembles a newspaper that's custom to their user.

I think there's also other things like in a world where everybody has a, you know, user aligned agent superintelligence on their behalf that's 10 x smarter than it is today. You could, for example, take a picture of a molding in your house that's that's in really bad shape. You could send it off to some manufacturing agent. They could negotiate a perfect price, and maybe you send it off to 50 of them or 3,000 of them. And maybe some of those are actually in China. It negotiates that in the background.

You end up paying $5 for it. You have the perfect new repair for your home. And so it's kind of hard to think about, but I think that when you give agents these things that can do super intelligent work on your behalf, access to your bank account, and access to act on your behalf, there's a whole new economy that comes out of it.

Mhmm. The old Internet went away in 1997 from pay per use. Well, it never even got there. It never even got there. So they put the code in, but they they couldn't do it. They had It was... No payment for sale for we're we're gonna build the driveway because we might have a car one day. Yes. Exactly. And then, obviously, the technology caught up. Yep. Seems like you're alluding to we might be going back towards this Yeah. Micro payment, pay for information,

pay for the little transactions and movements. Yeah. Is that sort of what you're seeing? Yeah. We'll see how it plays out. I would say that the big caveat here is that people have gotten used to this incentive model, you know, over twenty five years. And it's a very good... You know, it's a 300,000,000,000 a year business for Google. And so we'll see how it plays out. The other thing that has happened over that time is that people have gotten very used to the concept that digital goods are free.

Because advertising is such a profitable business model, it's not cheap to run Facebook. You know? The servers that run Facebook, that run Instagram, etcetera. But because advertising is so profitable, they've they've, of course, made the unit economics their work. And so I think it'll take some time. It probably won't at first be news articles or Wikipedia that will move to machine payments because those machines will probably just go to something else.

I think it'll probably be the more niche goods and services that agents need to get stuff done. So enrichment data, take this person. Let me find their contact details. Maybe KYC data that banks use behind the scenes. Maybe satellite data for hedge funds, drone data for hedge funds. So I think it'll start kind of more niche at the edge, and then more and more of the Internet will be monetized that way. You've been on the investor side. Now you're on the founder side.

When you think about the future of kind of market intelligence Yeah. Even as a meta moment or doing now, there's some kind alpha about hearing from someone who's in the trenches, in the arena, building an agentic finance that say

an investment firm might be interested in. Do you see there as sort of like a future of market intelligence where agents scraping the transcripts of all the market podcasts of the to get research, finding information in new niches?

Yeah. I think that's very possible that you would be able to research much broader information sets and pay for different information sets. I can't tell you whether or not that would make you a better investor and give you a better portfolio, but it certainly might. Think there's another opportunity when you put an agent in the loop for, like, better polling, which is, like, if you wanna poll a population of 50,000 people. Before, that was, like, extremely expensive to do. You probably need a team.

You have to reach out to all those people. You have to make sure they're in a really good spot. But theoretically, with an agent and particularly stablecoins where sending money is as easy as sending a file, you could theoretically get an agent to spin up some sort of survey and pull a massive population of people and pay them for their time and an input. And stablecoins,

it seems like are pretty attractive vehicle in the world of a gen to commerce. Yeah. So a few things. One, subcent

transaction fees. If you are transacting for a good or service that costs 5¢, obviously, 30¢ fixed fees of credit cards does not work. So subcent transaction fees are great. No marginal fees whatsoever, so you can send any amount of money with no cost. International by default, it's like moving a file. It's fully digital. It goes cross border, and no interchange is quite nice as well. But I think the most critical piece for agents is instant finality or roughly instant settlement.

Why agent cash is called agent cash is because cash is the closest equivalent to a stablecoin that exists. When I have a dollar, I hand you a dollar, now you have the dollar, and you're certain that you have the dollar. That doesn't exist with any of the other digital forms of payment that we have, whether that's ACH or wire or international wire, swift credit card. There's, you know, t plus two at best and then up to t plus 30, t plus 60. Yeah.

And so if you're transacting with an agent, you can do this sophisticated know your agent policy thing. You can do the sort of the browser fingerprinting that a checkout does to try to figure out if this is a fraudulent charge. You can do some sort of IP checking to, you know, try to estimate if this is, a real person or a scammer in Ukraine or a bot or a bot acting on behalf of somebody, or you could just use stablecoins. If I have a stablecoin and I hand it to you, now you have a stablecoin. We're good to go.

And so if you're a merchant transacting with hundreds of thousands of agents and doing millions of dollars of volume a year, stablecoins are really compelling. You don't take any risk. The person hands you the stablecoin upfront or the fraction of a stablecoin upfront, and then you hand back your digital goods and services, and the transaction's good to go. In terms of adoption, we mentioned how fast adoption was in general of LLMs

and how sort of the agentic world... We're accelerating towards a more agentic world as those get more advanced and they become really, like, embedded and integrated into lots of technologies that we're using. Yeah. That said, even as you were giving kind of the laundry list of all the ways you could have agents into your financial life, and this is both on an individual and institutional level, I can see the risk alarms going off in many people's heads or many institutions' heads being worried about kind of...

I mean, to reference Buffett again, number one rule he said is don't lose money. Right? So it's kinda like, oh, well, this seems like it could save us time and money, move us more quickly, be more efficient, maybe find a new opportunity, etcetera. There's lots of good wins, but it also feels like maybe this could go more wrong. Of course, there's human error. But with this kind of competing forces towards adoption, do you see this adoption curve?

You think it's gonna happen fast, or you think it's like, oh, this is gonna be like self driving cars. It's gonna happen slower than... Like, the technology is gonna be there fast. Yeah. But the user adoption is gonna happen slowly. Technology

is there. Yeah. Adoption is not. So the question is when that will happen. And it's once... There's a very, very compelling user experience. And I think for stablecoins particularly, the main blocker there is the ramps, the ability to access stablecoins. Probably most viewers know this, but getting stablecoins is very hard. So as a consumer, you have to go through KYC. That limit has recently been up to progressive KYC above about $4,500 depending on your on ramp.

And so you can get stablecoins in advance and then you KYC afterwards, but there's still sort of this identity flow that happens. For businesses, in order to buy stablecoins, have go through KYB, know your business. That is significantly more arduous, and most businesses would be unwilling to do that in order to use stablecoins even if you got 3% cheaper fees, thanks to no interchange. And so that's a adoption hurdle.

The thing that I think is happening, though, behind the scenes is these neobanks are building it directly in. And so new banks like Erbor, the Palmer Lucky Yeah. And a few others. Bank has stablecoins natively. I think Mercury is doing it natively as well and many other people. My personal consumer bank does not have it, but you could imagine a world where every bank has stablecoins on day one.

There's some incentive issues with credit cards there, but maybe they don't have a credit card program or they don't make too much money off of it. And that means that everybody has stablecoins from day one, and then the adoption is is much easier. You seem like,

you know, an agile founder. You're still somewhat early on the founder journey, but you're such a builder. You're gonna be building a long time. You've had one significant pivot at least. I'm sure there is maybe some micro pivots in there. But now, as you said, you're in the flow. Like, the flow of money, you've stepped into the river a little bit. Yep. And so there's... Maybe this is a tributary of agentic flows of money, and it's a smaller stream right now. Yep. But as you said, it's gonna grow.

Being visionary for a second, where are you going? Maybe what's the vision you're gonna raise on Yeah. Next round, next iteration?

Yeah. I mentioned this earlier with my motivations, but I think that this is key. I believe that every consumer, every business will have a agent or a fleet of agents that are entirely aligned with them that do a lot of their bidding on their behalf. This is sort of how we step into the next age as humanity is that you have these superintelligence agents that do all of this tedious work that you don't want to do in the background. It's very important that these intelligences are user aligned.

And so I'll give you an example of this, which is Amazon Rufus is the LLM that is built into Amazon, recently renamed Amazon Alexa, and it helps users with shopping recommendations. The issue with Amazon providing that intelligence to you is that it's aligned with Amazon, and it's aligned with their $80,000,000,000 a year advertising business. And so... Show me the incentives. Exactly. You see a world where it's provided by Amazon as one where it tries to convince you.

The the same type of LLM that we've gotten used to being incredibly empathetic, incredibly user aligned, well tuned by the people at Anthropic and OpenAI and others to be not scary, to be your friend, to be caring. All of those things go out the window when the incentive is switched from be the best agent for a user to be the best agent for an advertiser or a merchant. And so I think the... It's really import...

We're at a really important juncture in the road where AgenTek Commerce can be user aligned or it can be advertiser aligned. And I think that the world where it's user aligned is a much more interesting world for us And so that's what we're pushing towards. We wanna make sure that all of the commerce in the world, all of the checkouts are open to all agents. We wanna make sure, particularly, one of the ways that this plays out is that there's bilateral agreements between merchants and AI providers, agent providers.

And so you can imagine ChatGPT having a specific integration with Amazon and Gemini having a specific integration with Shopify. And those two bilateral deals exist, and you must do a BD agreement in order to get one. And that's the only way that you can buy stuff going forward is to use the Gemini agent or use the ChatGPT agent or whatever the next one that comes along. Or there's a world of open agent to commerce where every checkout is available to every agent.

Whether you want to use the ChatGPT or the Gemini agent, great. Or you wanna use the one that you built in your free time with your open source. I think that that world is a world where the agents could possibly be user aligned and is much more interesting. I like it. A user aligned world of agentic commerce. That's where we're going. Final thoughts. Anything we didn't hit that we should have hit? No. I think agentic commerce coming.

If anybody out there is building goods or services that they want to sell to agents, please reach out. We'd love to help you. We spent a lot of time with merchants making their endpoints compatible. I love it. I think big things are coming from Merit Systems. I hope so. Yeah. I still like the name, by the way. I know it was originally It's still vaguely relevant. It still feels vaguely relevant. Yeah. It still does. It's now sort of the umbrella brand. We think about it as as AnySphere is to Cursor.

I think it still fits in a little bit. So we're keeping the merch around. We're keeping the name and the emails. Well, it's funny. Sidetracked. But thinking about contracts,

you ever signed a b two b contract? Feels like there's almost always some kind of parent company Yep. That's like, that. And then it's DBA, whatever the brand. Some point. Yeah. There's a pivot. There's always... Even the big legit ones, there's usually a pivot. So, honestly, super interesting. You're on the frontier of this, and I think are already playing in a way that's very exciting and relevant. It's fun to imagine what's gonna happen in year three, four, five, and so on.

I feel like this could be when we look back on and be like, So

that was the beginning. Very early innings. We're in the we're in the flat part of chart. Well,

we're excited to hear more as, the months and years come. So thank you so much for coming.