How AI Bots Buy Internet Services

Andrew Van Anken  0:10  

Welcome to tokenize. The show focused on stable coins and institutional adoption of tokenized real world assets. My name is Andrew Van Aken, stable coin lead at Artemis, a crypto analytics startup. I would like to welcome you to a special episode of tokenize, where we're diving deep into all things AI. You might have heard about it. To do this, we are joined by Luca Shin, an analyst at Artemis, sir. How are you doing on this fine morning?

Unknown Speaker  0:34  

Doing well. Excited to be here.

Andrew Van Anken  0:36  

Have you heard of this AI trend, or is this

Speaker 1  0:38  

new to you? I have Yes, just hearing about a little bit all over the timeline.

Andrew Van Anken  0:44  

Excellent. Well, we're in good hands, because Lucas has done an extensive work on all things agent to commerce and payments, and I am very excited to learn more and dive into all things AI and payments. A quick disclaimer before we get into things, I need to remind you that the views or opinions of our contributors today are their own and do not necessarily reflect those of the companies they are representing. Nothing we say should be taken as tax, financial investment or legal advice. Do your own research. And lastly, before we get into things, I'm thrilled to remind you that this podcast is made possible by Visa and powered by Artemis so Lucas, I wanted to have you on the podcast because I think you've been doing some really great work recently on all things agentic finance and all things really tracking data around agentic finance and agentic commerce. And I kind of wanted to start by just asking, because I feel like this term, agentic finance, agentic payments, has just been thrown out and used all the time. When you hear that, when you hear me rambling about agentic payments, like, what does agentic payments mean to you?

Speaker 1  1:46  

Yeah, so there are definitely a lot of ways to interpret agentic payments and agents and commerce. What I'm most focused on, and what I'm most interested in, is the bucket of agentic payments, where you're using an agent, or you're interacting with your cloud code, and you're able to just easily pay for web resources through any of these means. So it's less so the agent tech shopping cart, like through your chat, GBT, being able to buy the same things that we already pay for today. It's more the new surface area, where, as a developer, you might be able to pay for an API endpoint and just integrate it into your workflow.

Andrew Van Anken  2:27  

So when my agent is browsing the best deal on a plane, this is not something that we're talking about. We're talking about like, Okay, I have a clawed bot. I have an agent that is looking to buy a API call or a piece of data from something, and that agent is going to do that for me exactly.

Speaker 1  2:45  

Yeah. I think both can be included. But I think the more exciting part is the new, the new emerging economy, which is things that we wouldn't pay for today.

Andrew Van Anken  2:55  

Yeah. And so this, this is really, like, almost, I would say, six months old or so. What pulled you into this? What was the catalyst to get you interested in that space?

Speaker 1  3:04  

Yeah, so I was brought to Artemis to kind of build out the crypto and AI vertical and be the expert there. But if you know Artemis, you know one of our specialties has been stable coins and payments, and I've always been interested in AI before Artemis, I was at, I was researching the intersection of crypto and AI at Delphi digital, covering things like decentralized training deep in and some identity protocols. But when x4 two started gaining traction back in October, something about it felt different and it felt tangible. So as I began interacting more and more with AI tools like Cloud code, and seeing how powerful AI was becoming, it just became apparent to me that agentic workflows were becoming a reality. And to me and many others, it felt obvious that using AI for coding assistance was was really just the start, and I think AI is bottlenecked only by the tools it has access to. It's already very intelligent. So if we could give our AI the ability to pay for things on the internet and pay for more tools and access it, can effectively be just a lot more useful to us. And then that's where x 402, and MPP come in. So to answer your question. Finally, it all started because I wanted to find out what people were paying for with X 402 how much of it was real, and then how much people were spending on the things that X 402 and MVP make available.

Andrew Van Anken  4:35  

That's a perfect segue. It's almost like you knew the next question. So since this is a data podcast, we're going to take a look at some data. And so what I have on the screen right now for the people listening home is our agentic payments dashboard, which I believe you built from scratch, from nothing. There was nothing, and then Lucas came, and then there was something. So we're looking at a chart of total agentic payment transactions. I. Since the beginning of these protocols, and what we see is like there's enormous spike in transactions right around Thanksgiving. So maybe as agentic payments are doing some Black Friday shopping and they're buying in all their API calls, that was a joke, and then there's like a decline after Christmas. So it definitely seems like, you know, even the bots are not buying things. It's, yeah, it's very seasonal. So when you saw this spike of transactions, I guess, like, my first question is, what's going on here? Like, what are, what are these agents actually doing?

Speaker 1  5:40  

Yeah. So especially in 2025 or the end of 2025 I mean, obviously that was the peak and activity, but it was pretty obvious to most of the people deep in the ecosystem that the activity was very meme coin driven. So a lot of the projects building X 402 related products. Or maybe they weren't even X 402 companies, but they were just trying to capitalize on this narrative. A lot of these companies were, or projects were attaching a meme coin to what they were building. And there's a good tracker in the ecosystem, X 402 scan, which, that was kind of the first place that you could track what's going on in X 402 and there's a leaderboard on that website. So a lot of a lot of new projects were trying to get to the top of that leaderboard. Gain attention.

Andrew Van Anken  6:35  

Let's take a look real time. Why don't we?

Speaker 1  6:36  

Yeah, yes. So this is kind of what fueled a lot of the huge activity and volume at the end of 2025

Andrew Van Anken  6:47  

so if we go back 30 days, we're seeing that. Let's see so Dexter cash block run stable and rich. Dot Dev, Clash of coins, probably a mean coin. So it's like, essentially, these protocols launch a coin associated with it and are trying to get to the top of a leaderboard as quickly as they can. So they essentially, like create volume to make it seem like they're actually facilitating transaction volume.

Speaker 1  7:19  

I would say that is the case sometimes, definitely not every, every name on this list is doing that, for example, like something like block run is a really cool use case. It's basically routing your inference. So if you're using open claw and spending 1000s of dollars a month on tokens, you're a lot of times overpaying because you wouldn't want to use open 446, for like a simple question. But with block run, you can submit a query, or your your open cloud has a query, and based on how, based on how complicated it is, it will route your query to the cheapest model that can accomplish or get the answer for your query. That's actually a cool use case that

Andrew Van Anken  8:03  

I've seen, wow, that I've kind of thought about that, but I've never put into the context of crypto, like Dex aggregators, where, like you're putting in an order and it's like you're getting best execution. Yeah. Okay, okay, very interesting. So from that, you decide, Hey, okay, look, let's try and figure out, like, what are people, like, actually, really using it for? Like, block run, it seems like would save people a lot of money and so that they'd be like, actually using it for. So can you walk us through? Okay, now that you see that there are a lot of mean coins being launched and whatnot, what's the next step to determine, like, how can we get to an accurate picture of what's going on in gender payments.

Speaker 1  8:41  

Yeah. So this has been kind of an iterative process, but I'm the purpose of this game versus real flagging is obviously, there's a lot of testing going on. It's a developing ecosystem, but to be able to get a good picture of what's happening, I need to at least filter out wash trading. If people are circling back funds through facilitators, just back to themselves. That really isn't telling us anything about genuine demand. So I guess my first iteration captured two scenarios. One of them was I was seeing that some of the merchants were actually routing X 402 payments through a facilitator back to their own wallet. So obvious wash trading. And then there was also another scenario where a seller would directly fund the buyer, who would buy a service or send money back to their own wallet. So these are classic wash trading patterns. But what was interesting was I wanted to build in the open, so I decided to make this methodology public early on. But when I did this, we saw an immediate drop off in the flagged games transactions, literally, like the next day. It's like someone Wow, flipped this rich and turned the gaming off. And so overall activity. Activity went down, and then also the percentage of gamed transactions went down, but then slowly, X 402, activity picked back up, but this time with a lot less gamed transactions. So obviously, a lot of these projects found a way around the flagging that we were using. So then I iterated further, made the approach more heuristic based, but this time, kind of decided it would be best not to publicly announce the exact methodology, just because that would make it a little bit too easy to find a way around. Yeah. So that kind of takes us to where we are today.

Andrew Van Anken  10:40  

Gotcha. I mean, the data looks pretty interesting. So what we're looking at is a chart of percent of game transactions, of x row two and tempo machine payments protocol. And it fluctuates day by day, but it's anywhere between 10 and 30 ish percent, I'd say, of game transactions, which is not that bad, I would say, I mean, considering that, you know, for some campaigns like AirDrop farming, you see tons of transactions that are so I think we're trying to see most like a an evolution or in of the space to like actual use cases. So what I found super interesting about this chart, so we're looking at a chart of average transaction size between the 2x 402 protocol and machine payments protocol, the tempo protocol. And I find it very interesting that the average transaction size x 402 is about $40 but the average transaction size on MPP is actually six cents. So to your point earlier about using API calls, it seems like that would be in the realm of price. One of the things I'm most curious about is like, and I know you kind of alluded to this earlier, but what are people actually spending on right now? Like, what are the most promising use cases you're tracking, and what are they buying? Because they're clearly not buying cars. For me, I'm not buying my car, but like, what is like an actual workflow look like?

Speaker 1  12:05  

Yeah, so couple questions there. The first thing I want to start off with is, like, this is filtered to the last month, and in the last month, we've seen on X 402, set of unlabeled merchants being included and picked up. And this has been a question that's come up a lot, and it's really just a handful of x 402 endpoints. Or they may not even be endpoints. They could be people using x 402 payments for non agentic use cases, if that makes sense. So there's really only a handful that are contributing very large dollar amounts in their payments. So, like, this is being skewed up a lot in the past month, gotcha,

Andrew Van Anken  12:49  

but it seems like early March, okay, yeah, like early March in February. Like, you do get that like 30 cent level transaction type, yeah, I

Speaker 1  12:57  

would say that's more aligned with what gotcha actually happened. Okay, I could filter out some of these high value transactions, but I do want to stay objective. And when I don't have the the exact labels for some of these, I don't want to just filter them out, just because, yeah,

Andrew Van Anken  13:15  

that makes sense. But to answer your other

Speaker 1  13:17  

question, what people are really using this for? It's, I think Kai Shek fields coined this term, but command line interface, I think that's the biggest use case. It's either using Cloud code from your terminal and connecting API endpoints to enrich your research process or your web search or your like lead enrichment, those are the biggest use cases right now. There's also just, if you're using open cloud, you want to make your open cloud more powerful and give it some skills, then, then you give it the ability to pay for things, and it can go out and find the best resources to do so. But I think the biggest user right now is either developers who want to test this cool emerging technology or sales and GTM people who want to say something like find me 50 leads from Series B fintechs and send an email to them based on where they went to school, What they majored in, how long they've been at a company, so it can go in and scrape LinkedIn or some other, some other platforms on the internet and kind of aggregate this information and actually send an email for you. So that's the closest thing that we have to product market fit. I would say, gotcha.

Andrew Van Anken  14:39  

Gotcha. I think I glossed over this question, but what's the difference between x 402 and machine payments protocol?

Speaker 1  14:47  

Yeah, so at the highest level, they solve the same problem. Basically, an agent requests a resource, the server says, pay me, and then the agent pays, and then the server delivers. So both of these use the 402 status code. But I'd say one of the biggest differences is that x 402 uses a facilitator model, while MPP does not. So a facilitator, for if you aren't familiar, is essentially a trusted third party that sits between the client and the server. So the facilitator receives the payment, verifies the transaction, and then they send the money, the payment to the server. Some of these facilitators include Coinbase CDP. They have their own. There's pay I Corbetts and many others. But on the MPP side, there is no facilitator, and instead, they use direct two party setup, where the client pays and then the server, the server themselves verifies the payment and then delivers the paid resource. And then, I think the other important difference between these two standards is, what kind of payment does each support? So right now, x 402, only supports one shot charges, where you basically settle funds each time you use the web service. So one settlement per use basically and MPP supports this type of payment, but also session based payments, which are basically streaming escrow channels. So as a buyer, you'd be able to deposit some amount upfront, and then settle in batches, incrementally as you consume the web service. So after you close your session, the remaining funds are returned back to you Gotcha.

Andrew Van Anken  16:31  

So do you think that we could even go lower than, you know, four or five cents a transaction, if it's like if we don't need to settle every single transaction, we could go even lower than that, 100%

Speaker 1  16:43  

Yeah, I think so. So if you go to the MPP deep dive page, yeah, let's do it, there's like a breakdown by payment type. Okay, interesting. So charges are that one off settlement type, and then sessions are the streaming I guess if you even go further down, you can see average transaction size. So what's interesting here is the session is actually higher dollar amount, but that's because one transaction is essentially one settlement. So that may be multiple batched transactions settled at one time. Does that make sense?

Andrew Van Anken  17:24  

Yeah. So you're doing all these payments, and then they all get batched together, and you're sort of like netting out the differences and then charging people

Unknown Speaker  17:35  

exactly, exactly. So I can make like 110 cent transactions, and that would be $1 gotcha.

Andrew Van Anken  17:41  

Gotcha. Okay, yeah, that makes sense. So maybe it's not like we're doing, you know, hundreds of millions of transactions a minute of an hour or whatever, but we're okay. We're kind of like batching them together, yes, okay, that makes sense. That's interesting,

Speaker 1  17:55  

because it gets around the the problem of being able to support billions of transactions, like, if you know them all, each time you make a transaction, it's, I think it's a lot more difficult than batching together. So it's just another way we can make this work.

Andrew Van Anken  18:13  

Yeah, that makes sense. I mean, I think that the, I wouldn't say it's the elephant in the room, but the volumes are still pretty small. And I look, I acknowledge that we're still very early on, and it's like mostly a developer focused game. And so what we're looking at is essentially x 402, volume by chain. There's obviously been a spike recently. I mean, earlier in March, we're seeing, you know, maybe 50 to $100,000 a day. Now we're seeing in the millions of dollars a day. It's kind of like what you're talking about this, like, other category popping up. But what do you think needs to happen for this to really take off? And like, what sort of resources do you need to see come to fruition?

Speaker 1  18:57  

Yeah, so it's two things, supply side and discovery. So on the supply side, if I'm a user, there aren't even enough useful endpoints right now that I think agents would actually want to pay for. There are some cool hints of useful use cases. But I think the supply side has to grow a ton. And I think by the end of the year we'll we'll usually have 10s of 1000s of these x 402, or MPP enabled endpoints, especially with AI, making it so easy to build these types of web products. But then when this happens and the supply side grows, we'll need a reliable way to find the right resource for any given task. It kind of defeats the purpose of this whole agentic payments thing. If there are all these useful resources but we can't find the right one, then it'll have to be manual. I don't want to have to tell my agent or my cloud bot to use a specific endpoint every time like I just want to give a high level of objective and the agent is. Able to find the right resource. So I think kind of like that recommendation system needs to improve. It's a really tough problem, but we'll kind of unlock this whole ecosystem. And then the last thing is just, we're seeing hints of what the type of customer is that is going to use this technology and payment method. But that customer base is still being born, and right now it's a slow ramp up, but I think the demand side will also increase very quickly by

Andrew Van Anken  20:27  

the end of the year. Yeah, we're looking at a chart here of a cumulative number of extra two sellers, and it's doubled since March. So March, we had about 2000 cumulative sellers. Now we're looking at it's more than double, actually 5000 so is this probably a good metric to watch in terms of, okay, like, Are people experimenting with the protocol this, like, number of things that agents can do, essentially the number of sellers?

Speaker 1  20:50  

Yeah, exactly. It's what I was just kind of alluding to, just the supply side growing. And I think one thing I want to call out here is, do you see how it says adjusted at the top here. Yeah. So we're not actually tracking only the number of sellers that we've ever seen. We're we're only filtering for the sellers that have greater than two quote, unquote, real transactions and then at least two unique buyers. So you kind of have to qualify to be included

Andrew Van Anken  21:21  

in this, in this metric, gotcha, gotcha, gotcha. But that makes sense. Like, I mean, if I can just buy and sell for myself, I probably shouldn't tell people that, right, right?

Unknown Speaker  21:30  

Another attempt at trying to clear the

Andrew Van Anken  21:32  

noise, yeah, yeah, that makes sense. Lucas, I feel like we covered a lot. Did I miss anything, or is there any anything else that people should know about the agentic commerce revolution that we're

Speaker 1  21:44  

seeing, I would just encourage people to experiment, play around with the tools that are available right now. I'm always a little worried about being left behind, and I think this is a cool area to pay

Andrew Van Anken  21:56  

attention to. I haven't made an agentic commerce transaction yet. So I'm already behind. If you're sitting at a payments company, like, let's say you're like, the VP of payments at like, shift four, or a company like that, like, how would you recommend people get started? Or how would you recommend payment teams get started with these new initiatives?

Speaker 1  22:16  

Yeah. I mean, I think it always comes down to testing things yourself. I think that just gives you a better idea of how powerful this stuff is. And remember, it's an important emerging economy, and I think it could become very large. So I wouldn't want to miss out on some of this, some of the volume that goes through this emerging economy.

Andrew Van Anken  22:35  

Yeah, it makes sense. Lucas, I think that's all the time we have for today. Where can people find more about you and more the work you're doing at Artemis, I tweet a lot. My Twitter

Speaker 1  22:45  

is or my ex is on chain. Lou one word, think

Andrew Van Anken  22:48  

that's about it, awesome. Well, I already follow you. And I mean, you've put out some great research recently, some very long, detailed dives into all things, agent, tech, commerce, market, maps, analysis, payment. So check it out. If you haven't followed Luke's already, a lot a lot of good materials out there. If you haven't already, please subscribe to tokenize on Apple Spotify or whatever podcast app you use. And finally, if you enjoy the show and want more, please leave us a review. Stay stable everyone.