# Autonomous Agentic Commerce: Micropayments, AgentCash, x402 & MPP > Shoal Signal episode. Guest: Shafu (Merit Systems). Host: Gabe Tramble. - Publisher: Shoal Research (https://shoal.xyz) - Published: 2026-04-20 - Video: https://www.youtube.com/watch?v=vVmyBfGqM7Q - Canonical: https://shoal.xyz/research/episodes/autonomous-agentic-commerce-x402-mpp/ - Keywords: AI agents, agentic payments, x402, micropayments, interviews ## Transcript **Gabe (00:01):** So Shafu, can you give us an understanding of what MPP is, X402, and these agent payment protocols, and this will set the stage for the rest of the conversation where we can get more into the nuance. **Shafu (00:13):** Yeah, just imagine you have an API, right? And without X4.2 or MPP, you have an agent that wants to access that API. The agent just can't, right? Because you would need to sign up, get an API key, pay a subscription, all that kind of stuff. ⁓ X4.2 MPP completely changes that. So you have your API, you put X4.2 or MPP around it, and now it's paper request. So the agent goes ahead and wants to access I don't know, some API, right? And it just can access it. It pays with stable coins and it gets that data back. So basically it gets rid of paywalls on the internet. And what's beautiful about X4Tune MPP, in my opinion, is they're, first of all, they're open standards on top of HTTP and they're extremely simple. They're super thin. Basically what they include is how much Do you need to pay to access this? How are going to pay this on what chain and what currency? And that's pretty much it. **Gabe (01:21):** And can you explain a little bit of the nuance between X402 and MPP? Are these the same exact protocols for accessing APIs or is there some technological differences, benefits or side effects from both? **Shafu (01:37):** Yeah, from a high level there, they're very, very similar. I guess the biggest difference is that X4.2 came out of the Coinbase ecosystem and MPP came out of the Stripe and Tempo ecosystem. ⁓ High level, they're solving the same thing. There are some nuances in the technical details, some little features here and there that are different, but from a high level perspective, they do pretty much the same. **Gabe (02:07):** Got it. So just to step back, X4.02 and MPP, these agent payment protocols allow agents such as like a cloud or cloud code or codex tools, any types of agents to access APIs. without having to authenticate and then they can use stable coins. And I think it's interesting that Stripe, as you mentioned, is participating in this landscape. So this is like a big service provider. I know in San Francisco they have a billboard that says they process like 1.3 % of the global GDP. They're in the financial payment space, right, for a long time. Why do you think that they're launching this with stable coins as the rail? Like, why isn't this possible just with using a credit card? Like, why can't I just give my agent the credit card? What is the unlock or what is the change here where stable coins are the payment rails? **Shafu (03:04):** Yeah, it's a great question. The problem with credit cards is you have fixed costs. I think it's around 30 cents ⁓ and you have like 3 % on top of that. So just imagine you want to access an API which costs like sub one cent, right? Now you need to pay 30 cents just in credit card fees to access that API. So that just doesn't make any sense. And that's where stable coins come in. They're perfect for Micro-transactions, they're instant, you there are no refunds, all that kind of stuff totally goes away. It's extremely cheap. It just works out of the box. ⁓ So in terms of credit cards, it just wouldn't work for ⁓ X4 to an MPP. Maybe it works for like high ticket purchases. So just imagine your agent goes off and buys like jewelry or like a car or something on Amazon ⁓ or somewhere else. Maybe that will work, but where we see the most use out of X4.2 and MPPs, people paying for API calls. And these are extremely cheap, so it just doesn't make sense to do with crycarts. **Gabe (04:19):** Yeah, it seems like there's like a convergence right now of, you know, capable agents, stable coin rails being pretty established. And then this new open protocol for transacting micro payments through the APIs. Kind of the important thing here is the supply, right? What are some of the projects or service providers or tools that you've been seeing or think are some of the best ones that are on the other side of the equation here providing services or providing API support and tooling? **Shafu (04:50):** Yeah, so we have native integrations into Firecall, Minerva, Nansen, Misari, bunch of others. You can see all of them on X4.2 scan, which is like this discovery tool that we built. We list all the ⁓ X4.2 resources there and all the API providers that support X4.2. ⁓ The beautiful thing about X4.2 and MPPs, that it's very easy to wrap an API as well. So just imagine you have an established API provider and they're just not interested in the agent space, for example. ⁓ It's very easy for anyone to come in, get some API keys and put their API band X4 to an MPP. But there are already a ton of great native ⁓ integrations. **Gabe (05:39):** Yeah, some of the ones I like are like Parallel or Exa, search and retrieval tools. And I think kind of the value here that I see is the traditional search, right, like in these agents, maybe like Claude, right, is pretty good. ⁓ But these external tools are like significantly better at search. So when it comes to search, that's kind of like my big use case. I'm curious for you personally, like what are the ones that you're ⁓ really excited about or use cases that you're **Shafu (05:44):** Yes. **Gabe (06:09):** you're honing into. **Shafu (06:11):** Yes. So what we see for me personally, I just did that before before I do the podcast. was, you know, we have this new product called Poncho. It was basically, hey, find a thousand potential customers for Poncho. And it's really good at that because it has access to Apollo. It has access to LinkedIn. It has access to Parallel. So it can look things up like without like Booting up a browser and scraping things just has access to the API. So it's really, really good at finding people. It's really, good at finding companies and enriching them. And that's what we see right now. It's really good at lead generation, stuff like that. **Gabe (07:02):** Got it. So just a step back. have the agents that can do things and basically like process cognition. And then the X4 02 and MPP are micro payments to these hundreds, if not thousands of legitimate services that the agent can just call, right? And, and your use case that you're saying is pretty interesting is punch, Poncho product needs users. And then you can use these like email generation ⁓ services, right? Where they, give you like emails for people who might fit this type of product and then you're orchestrating between all these different tools to basically like simulate like a salesperson essentially. Is that like a good articulation of how you're using it? **Shafu (07:48):** Yeah, exactly. And just imagine poncho is like, it's like Claude on the web, ⁓ plus the X-Fortune MPP superpowers. That's always how I like to describe it. It gives you agent superpowers. And by superpowers, mean, it just gives you access to these great APIs. ⁓ So now Claude has access to LinkedIn and YouTube and Apollo and parallel and XR. and much, more ⁓ out of the box without setting up a single API key or a single subscription just with one unified balance with which you pay. And that's it. And now we have access to basically all the great APIs on the internet. **Gabe (08:39):** I'm sure you've talked to some of these service providers. Why do you think they're willing to, you know, try this new landscape? Like what, is in it for them if potentially they're bringing in less revenue ⁓ per customer, right? Where before, say, usually they'll have like a $20 or a hundred dollar, et cetera, subscription, this fixed cost subscription. And then you have some credits that you'll use on the API, but now you might use their service and, and, know, be very efficient in the usage because you're only paying per use, right? So for the companies, they're receiving potentially less money, but there's some discoverability. Like, what do you think is the reason why they've been so willing? Because it seems like there's a lot of big projects that have really just hopped on board. **Shafu (09:27):** Yeah, I don't even think it's less money. think in the long term, they're going to make more money because it's just a new distribution channel. So just imagine you have someone that, you know, instead of, he can't, it's just not worth it to spend like 50 bucks on, you know, some Apollo subscription or something, but he definitely can't pay like two bucks, you know, to just access the data that he gets. So Apollo just earned an additional $2 just from a new customer, just whole new distribution channel. ⁓ Obviously people are excited about this because everyone talks about the eyes, it's the biggest thing in the valley, all that kind of stuff. And agentic commerce is just the next evolution of this. It's like super obvious to me that agents are gonna spend money on the internet. I think that's pretty obvious. Like I'm inside Claude code, I'm inside Codex. probably 10 hours a day, you know, not just engineering, just like I described earlier, like sales stuff and marketing and go to market. And for me, it's like extremely frustrated if I'm, for example, inside chat GPT and it just cannot access LinkedIn or it cannot access YouTube because I just got used to it. I got used to like the X for two MPP superpowers. and **Gabe (10:44):** Mm-hmm. **Shafu (10:55):** Most of these data providers or API providers see that. ⁓ They see the agentic thing, you know, just taking off. And they see the vision where like, an agent probably is not gonna open up my landing page and is probably not gonna talk to some sales guy. They're probably not gonna sign up through Gmail or Google or whatever OS. They just wanna access the data. and they're going to pay for it. ⁓ And it just pay for request and it works. ⁓ And it can be more expensive, right? So we have, we have, for example, white pages API, which gives you like homeowners or things like that. And this is like 40 cents per API call because the API is just that valuable. So it doesn't need to be like sub one cent, but it can be. But, but I think like the big picture is just this totally new distribution channel for these API providers to sell their services to agents. **Gabe (12:02):** We'll get to ads in a minute, but this discoverability thing is kind of the first thing that comes to mind where, okay, let's say you do have a thousand services, right? And there's going to be overlap between services. How do these agents... do self discovery if there's so many options and kind of one example is, you know, I might say, okay, use Exa or use parallel for search because I personally know that those are those are two high value search companies. But for the agents themselves, what is the mechanism here for discoverability and how can they basically route to the best appropriate tools? **Shafu (12:43):** Yeah, so how Discovery works currently is we run X4.2 scan and MPP scan. ⁓ And how X4.2 scan and MPP scan works is it's certainly permissionless. Anyone can register their API there or their server there. And ⁓ why is that magical? It's magical because we index that automatically. And when you use Poncho, which is basically your client to this agent e-commerce space. It's like the browser of agent e-commerce. You're inside Punchline, you're like, I wanna access LinkedIn data. It's gonna do a search over all the things that it has on X4 too scan and PPCAN, and it's gonna give you the best one. It's Google plus the browser basically. And that's how discovery works. **Gabe (13:39):** And how are you, I guess to dig deeper, how are you validating if something is like the best one, right? Like, is this a manual process? Is there like a hierarchy or like some type of triage system that runs through the different tools that these ⁓ X4.02 endpoints or MPP, the agent payments endpoints presents? How are you actually deciphering between for your own product? **Shafu (14:05):** Yeah, so right now it's a mix between a manual process and automatic indexing. So on the manual process, it's just, you know, when we onboard someone like Parallel or Exa or Minerva, we just know they're good, right? We just serve them. ⁓ For the other stuff, it's like, how often do they get called? How often do they return a 200? How often do they fail? what's the quality of the data there, all that kind of stuff. A very long-term vision, it's probably gonna work like Google. We just index things and we have metrics that we look at, all that kind of stuff. But right now, it's like a manual process and an automatic process. **Gabe (14:53):** Okay, so kind of the long-term vision of this is you are using the metrics from these tool calling to almost like rate the quality of each service and then you can route through the highest quality services on top of like a curated list. **Shafu (15:10):** Exactly. yeah, maybe even we return like three, maybe you want to access LinkedIn data and we give you like three API providers and the agent just decides which one's going to use or if one fails, it just tries the other. It's actually pretty good. So if you use, if you search inside Poncho, it's actually pretty amazing how well it works. **Gabe (15:36):** And if a agent tool provider, if the return comes back failed, does it still charge the agent or the person who's operating the agent? **Shafu (15:52):** Yeah, depends on how the server implemented Xfru2. ⁓ So you can implement the refund mechanism, or you can implement, I'll give you the data first and then I charge you. But it's up to the server. ⁓ So theoretically, the server could charge you and not give you the data. ⁓ The good thing is we're talking about sub ones and transactions. So if that happens once, they just get blacklisted and they're never called again. **Gabe (16:23):** Okay. So yeah, because there's almost this risk here, Where you're just, you know, you don't know who's on the end of the table all the time. And if you are, you know, putting your funds out there, even if small, like it's kind of important to figure out you're not just calling like a hundred different, you services and then you're racking up these things. So that to me seemed like a pretty big ⁓ opportunity, really, not even almost like a problem, but like opportunity space with the routing. And to extend on that piece, we now understand we have these agent payment solutions or the endpoints or micro payments on APIs. You can call the API and then you have like these extended toolkits of, you know, models and search and kind of anything, know, flight tracking, all that type of stuff, ⁓ really opening up the capabilities for agents. When it comes to like the next phase of this, because right Right now you kind of specify the types of tools that you use or maybe that you use a service like yourself. But you guys talked about this concept of emergence and how over time, and I think this is where it gets a little bit muddy, but over time the agents can almost do self discovery as if they're like in a shopping mall with a goal, right? So if the goal is to ⁓ come back and bake a cake, right? They can go into the mall and like walk around and maybe they go into a close and there's obviously no food there, but then maybe there's like a little grocery store, ⁓ a bakery or something, right? So this orchestration across the tools by themselves where you're not specifying, that's a whole different kind of behavior change. people are not even quite in tune with kind of phase one, but I do wanna do some discovery around here of how you guys are looking at this ⁓ autonomous commerce. and how you're looking at this discovery mechanism for these types of tasks. **Shafu (18:26):** Yeah, I think that's super interesting. think where I see emergence is when you combine all of these APIs together. ⁓ So right now, if you go to Ponscho, you could be like, look up Shafu on LinkedIn, get me his profile picture, create a website, ⁓ make an image out of him, you know, serving on the East River, then create a song and post it all on a website. And it can just do it. all of that because it has access to the image models, it has access to music generations, it has access to LinkedIn. And this is where I see this emerging thing happening, where you just combine all of these APIs together and all of a sudden you can solve things that you couldn't do before. Always just a lot of effort, right? So instead of getting like 16 API keys, it just works out of the box and you don't even have one subscription. ⁓ So I think like this emergence of like just combining all these tool calls or just combining all these API providers is really powerful. And we already see that just because you have so many functionalities now that are behind X4.02 and every time they're, every day they're new ones. ⁓ So on X4.02 scan, I think we have like six high quality API providers. that got wrapped and get put on x-futuristic scan. So every day it just gets more powerful and more powerful and this compounds, right? ⁓ So yeah, this is what I see as the emergent property of like just using open standards, right? Because the difference here is ⁓ if I wanted, let's say I'm an API provider, right? And I want to get my API into Chatchi PT, right? What would I need to do? I'll probably not talk to Sam Altman or some other high up guy at OpenAI. And then we would have like an enterprise agreement and it probably takes six months. And in the end, it probably wouldn't happen because you just don't have the cloud or something like that. And this OpenAgentic commerce thing is anyone can just come in, put your API, put your server behind X for a tour MPP. And now every X422 MPP client like Pawnshore just has access to your resource. And because we have the discovery layer as well through X422 scan MPP scan, it immediately shows up in search as well. So just like the open internet, right? Like I could, know, 20 years ago, I built a website, it gets automatically indexed by Google. I don't need to, you know, call the Google guys and ask for permission. I just can't do it. that's the same thing in this open agent e-commerce space. **Gabe (21:24):** Go ahead, got it. So for you guys, like how do you prevent against security risks? So we had a guy from SIRTIC, a smart contract auditor, and he kind of was explaining how agents are very capable, but they have like no security instinct, right? So how do you prevent against something like a prompt injection where, you know, you have an agent doing an autonomous task, it goes around and says, oh, wow, this service, you know, is going to **Shafu (21:44):** Hmm. **Gabe (21:54):** exactly what I need, right? Or it's cheaply priced and then it gives you some information to manipulate the agent. How do you prevent against this and kind of what is the domain or category which we should kind of think about this in? **Shafu (21:56):** Yeah. Yeah, I think the biggest lift or the biggest leverage is spending limits, sure that the servers are okay. So let's say I think like we have a custom, I think we have a set spending limits of like five bucks on poncho. So that means you like your agent couldn't spend more than $5 on one API call. It will ask you like 20 times because we have APIs that are more expensive than five bucks. Like you can get a shirt through X4 2 scan. You can buy things on Amazon through X4 2 scan. So they're obviously more expensive than $5, but your agent would like ask you three times if you're a hundred percent sure if you want to do this. So spending limits is one. The other one is indexing servers, you know, see what they return. ⁓ making sure that they're actually not exploited by someone else or something like that. But at the end of the day, just because it's totally permissionless, know, someone could theoretically put in a malicious server in X4 too. And it gets, it gets indexed. It's basically like the open internet, right? It's like what Google solves all day. How do you make sure that what you show to people when they search for things is actually legit and not exploitable. So it's a very similar issue, I guess. **Gabe (23:42):** Yeah, okay, so not necessarily intrinsically ⁓ or structurally only something that comes up with agents and agent commerce, but this is kind of like a reflection of the same issue with landing on a bad link or kind of like ⁓ getting a bad email, right? **Shafu (24:00):** Yeah, it's fundamentally a search issue. In the case of agentic commerce, it's a little bit different because you actually pay money immediately. So you just want to make sure, you know, like the spending limit helps a lot. So in the case of the spending limit, like the worst case would be $5. And that still wouldn't be great, but it wouldn't be the end of the world. So that's how we try to counteract that kind of stuff. **Gabe (24:10):** Thanks Mm-hmm. I have a more theoretical question for you and Basically, the way I'm using and seeing these products is oftentimes my agents will talk directly to like a service, right? And usually the service is behind REST API. ⁓ A lot of people talk about this agent to agent commerce. And as of right now, I just can't really think through the case studies in which one agent talking to another agent actually makes sense. It's almost as if like, **Shafu (24:58):** Yeah. **Gabe (25:03):** ⁓ the use case is literally not conceivable right now. So yeah, I wanna like tighten on that nuance and I wanna hear your take on that. **Shafu (25:07):** Yeah. Yeah, I don't really have a take here, Gabe. I see it the same way. I don't fully get it. And we don't see that. So like 100 % of what we see are people using things like poncho. It's like access this LinkedIn and find me more information about this person. Find me 100 companies similar to this. Go generate an image. Go upload this file. This is 100 % of what we see. ⁓ I haven't seen this agent to agent thing in action. ⁓ So yeah, I have the same opinion as you. **Gabe (25:56):** Got it. Got it. Yeah. Do you think there is any use case or any theoretical example where an agent might need to talk to an agent? And I guess my idea of that is let's say I have a business and you have a business and we kind of have to come to like a non ⁓ you like my request and your request has to be kind of merged together and then we can kind of figure out the nuance, but even that can be a marketplace, right? So it's like every time I think of some type of agent to agent commerce use case, it turns into a marketplace or it's, it's not really needed. ⁓ so yeah. **Shafu (26:36):** Yeah, maybe one thing I could think of is, let's say we live in a world where the top models are just extremely expensive, right? So you get like cloud code subscriptions like 200 bucks, right? Which is amazing. But let's say there are models that are just a hundred times better and they're a hundred times more expensive. ⁓ Maybe an Asian to Asian thing would be, Hey, this is my agent. have like the cheaper version and I'm going to rent the more expensive version for like five minutes to just do this one task. ⁓ so you rent another more powerful version for like a limited amount of time. I just came up with that example. ⁓ maybe that's, that's something, but I have a hard time like think, like seeing in the foreseeable future. you know, like my agent talking to your agent and they're trying to like, I would just talk to you. like that's like, yeah. So I'm not sure about the agent to agent thing. It's at least, it's at least not something that we're seeing right now. **Gabe (27:51):** Yeah, yeah, the way I'm kind of thinking about that too is like the, if someone has information that is so valuable that someone else is going to pay for, then usually that turns into, you know, a business, right? And a service and the service has endpoints that you can just hit. So the question kind of becomes like, what use case is there where you don't have a full service, but someone has some information that's more valuable. **Shafu (28:06):** an API. Yeah. **Gabe (28:20):** or valuable enough to transact with. And I just, personally can't think of any right now. **Shafu (28:25):** Yeah, it's a good point. If you have that, turn in an API and charge for it. ⁓ So yeah, maybe in five years, maybe in three years, but not currently something we're seeing. **Gabe (28:30):** Yeah. Okay, yeah, thanks. Yeah, that's a hot take. I think that's not how a lot of people see the space right now. ⁓ For ads, for advertisement, you've made some comments about how, you know, agent commerce and ⁓ agentic agent commerce will disrupt the advertisement space. Can you give us like your thesis or your lay of the land and how you see that happening? **Shafu (29:08):** Yeah, it's very simple actually. So the vision is you go to the New York Times, for example, very, very theoretical example, because I've never been to the New York Times, but whatever you go to the New York Times, you hit the paywall, right? You hit the paywall. It's like pay us $10 for a New York Times subscription or the New York Times is free and you see a bunch of ads which suck. Like I hate ads. In this X-World 2 and 3 world, it's, hey, I want to access this article and I'm going to pay 20 cents to access this article and you're not going to show me any ads. Another example, I've been paying for YouTube premium for the last, probably since I can afford it, maybe last 10 years, just because I don't want to see ads. In this world, Instead of paying, I don't know, I think YouTube premium is like 18 bucks a month or something. Instead of paying 18 bucks a month, you just pay 10 cents or one cent to access a specific video. And instead of serving ads, blah, blah, blah, you just pay and probably the creator gets like 30 cents, YouTube gets like 70 cents or whatever. But like that's the model. The thing is, **Gabe (30:17):** per video. **Shafu (30:33):** Previously, the only way to monetize the open internet was ads. We've been living in that world for the last 30 years. It's the only viable way to monetize anything on the internet right now. But that has changed. We now have an alternative. You can put anything behind X-Root 2 and let people do micro payment, micro transaction. And now you have an immediate source of revenue. No subscriptions, just paper requests and no more ads. ⁓ So that's the vision. **Gabe (31:12):** Okay, and just to put it all together and to help us understand for thinking about it correctly, you have a service like New York Times, and usually it's on the columns where they have the ads, and people read the article, it's free, and then the revenue comes from this ad space. But with the agents, the revenue gets paid instantly to... **Shafu (31:24):** Yeah. **Gabe (31:34):** view the article ⁓ and then that changes the entire dynamic of the ad space. So with that in mind, do you think that ads on the internet go to zero? you know, let's say like agent payments roll and this becomes mass adoption, ⁓ which it probably will. What actually happens to the ad space you think? **Shafu (31:57):** Yeah, it will definitely not go to zero. The funny thing about people in general is they don't value their time or they don't value their attention. So we all know people in our lives that make good money and they will get like the ad-supported Netflix version. We all know that person as like $5 cheaper and they're like, I don't care, I like to see an ad from time to time. So we always have that kind of... We're probably 80 % of the internet. But I would argue high value people. And actually, there are studies about that. If you see a lot of ads, you're probably poor. It's true. So I think we're always going to have ads. **Gabe (32:35):** Thank **Shafu (32:55):** But now we have a much much cheaper alternative. So instead of getting the, you know, you're some poor high school student, right? You're not gonna pay 18 bucks for a YouTube premium subscription or whatever. But maybe you pay like sub one cent to not see ads on a YouTube video ⁓ or on a news article or ⁓ that kind of stuff. So it's just, we have a different path now. We have the alternative. But I think ads are going to be on the internet for a very, very long time. The other thing is that that is pushing against ads in general. It's just if we move more and more into this agentic world, just imagine people are not using a web browser anymore. So for example, in our office, right, if you walk around, most of the people are looking into a codex terminal. or a clock code terminal. ⁓ It's not only engineering because now with X4.2 you can do a bunch ⁓ more. So there isn't even a place to serve ads, right? You're not gonna put ads into a terminal. So you're gonna lose eyeballs on the ads for sure, ⁓ but it's not gonna go to zero. ⁓ **Gabe (34:06):** Mm-hmm. **Shafu (34:17):** The great thing about Asians is you cannot serve them ads, right? They don't get distracted like that. It just doesn't work like that. So it's fundamentally a different paradigm. I'm happy. I'm just happy we have an alternative now because I really, really dislike ads. **Gabe (34:35):** Yeah, interesting. do you see this like stark change of like class on the internet where like literally you're going to be in the ad space or you kind of have your accounts loaded across the board and then you're not seeing ads? Because to see an ad you would basically kind of land on the website, right? So. **Shafu (34:55):** Yes. **Gabe (34:57):** That's like a consumer experience. maybe if you're looking at your agent, let's say Claude at some point can pull up the article in full view, which through ⁓ X4.02, those are two different kind of like UI and discovery features. How are you looking at that? And do you think that everything will just be pulled into an agent interface or do you think there still will be like this separate kind of like a search with fingers, right? Like Google searching interfaces that are heavily used. **Shafu (35:30):** Yeah, think the question about class is a good one. I think we already have a class system. And the class system is people that can afford a cloud code subscription and people that can't. ⁓ Or a codec subscription or whatever. It's just a fundamentally different experience of how to use a computer. It's just a way, way more powerful experience on how to use a computer. ⁓ I think more and more is going to be pulled into this agent experience. Just imagine Codex or Cloud Code getting 10 times better, 10 times faster, and 100x more, and 100x cheaper. I always don't Google anything. I don't know about you, but I literally, couldn't tell you when I, the last time I Googled something. So we're definitely pulling in that direction. **Gabe (36:34):** Yeah, on the class piece a little bit to pick up on that, I am noticing there's this compounding effect of creating systems that are optimized for the individual. And I think there seems to be like a there's a huge technical margin, right, for being able to build your own systems, whether that's searching like in like a sales role or just kind of keeping up with daily tasks or like tasks in your life, right, using these tools. But it's compounding, right? So I'm curious about your thoughts of like, three years from now, you might have like this whole infrastructure that might be equivalent to like 10 humans of labor per day, right? What do you think is kind of capable and how do you think things change within society when you have? you know, people walking around that have that are individuals with, you know, the capital, the output capital of 10 people versus walking next to folks with the output capital of one person. **Shafu (37:38):** Yeah, I think, and this is going to get philosophical a bit, but it's interesting. I think the world in general is getting more extreme. Like you're going to have the 16 year old billionaire and you'll probably have a lot of people without work. ⁓ And then the question is, how do you solve that? And I'm not sure. And you know, people talk about UBI and that kind of stuff. doesn't feel right to me just as a very capitalistic person. ⁓ **Gabe (37:47):** . **Shafu (38:06):** But yeah, the world is gonna get more extreme in every way possible. Just you're gonna have very, very successful people on the top and you're gonna have a lot of people without anything to do. ⁓ Because the agents are just better at what they could do. ⁓ Do you know the permanent underclass meme? **Gabe (38:32):** Yeah, yeah, yeah, we actually wrote a paper on that. **Shafu (38:34):** Interesting. Yeah. So it's like the permanent underclass idea is like, if you're not going to make it now, you're going to be stuck in the permanent underclass forever. I don't believe that. ⁓ It's still a very interesting idea. It's just learn the tools, know, learn the tools. If you watch this, you probably already do, but I don't know, like how many people are using these agents? Like 1 %? **Gabe (39:03):** Yeah, maybe less. **Shafu (39:04):** Sub 1 % it's insanely small. ⁓ So yeah, the world is going to get a bit crazier. It's going to get more extreme and I'm not sure what to do about it. **Gabe (39:19):** Yeah, I agree. The permanent underclass thing is interesting because I think it actually is a flywheel mechanism for AI in general, where it's kind of like, what is the behavior that comes out of knowing about the permanent underclass? Like you're going to spend more tokens and kind of figure out how to deploy your agents and all these things. But at the same time, it's not like a not real risk, right? This is a real paradigm shift happening. **Shafu (39:28):** Hmm. Hmm. **Gabe (39:47):** ⁓ Also, and if you have comments, please hop in. On the capitalism thing, I think that capitalism actually is eroding because of AI and as the models get better and better and better and even if you have access to all these different tooling, right? **Shafu (39:58):** Hmm. **Gabe (40:07):** the cost is just it's going to keep going coming down and down and down as ultimately we might see agents, you know, building themselves and optimizing and and yeah, I'm curious, curious to hear your thought on that piece because for me, it kind of erodes capitalism in a way where ⁓ it's hard to capture value in like a capitalist society. We might move into like a feudalism or or maybe we're doing like arts and stuff like that. But ⁓ yeah, can hear you. thoughts. **Shafu (40:39):** Yeah, ⁓ I think it's an issue. I'm really, really interested in ⁓ online dating statistics. Have you looked into online dating statistics, like Hinge and stuff? Okay, so very, very interesting stuff. So I would say, like in New York or the US in general, we live in a very liberal society. **Gabe (40:52):** No, no, no, I'm up for that. **Shafu (41:08):** That reflects the stats. So on Hinge, you have 50 % of all men don't get one match and you have the top 10 % of men get all the matches. Right? So there is this concept of hypergamy where it's, you see that in every culture on the planet. It's basically the concept of women marrying up. Up in status, up in wealth. It just... It's a concept out of social sciences. And if you don't regulate that, it just gets more extreme over time. And I think the same thing we're seeing with just technology in general, ⁓ just the concentration of wealth, ⁓ we're going to have the same effect. And if you don't regulate that, which ⁓ I'm not sure we should, or at least I don't know how, it's just gonna get more extreme over time. So I don't know, like the top 10 % of people own 80 % of things. It's probably gonna be like the top 1 % owns 99 % of things, stuff like that. ⁓ So you just see the tendency, you see that in online dating, you see that in wealth, you see that, you know, all these kind of things. ⁓ **Gabe (42:24):** No. **Shafu (42:37):** and it's gonna be more extreme over time. **Gabe (42:40):** Yeah, yeah, yeah, I can't speak on the dating example or like kind of what it means ⁓ socially, but for the wealth piece, yeah, I mean, that's something we've seen for a long time and it does exacerbate it. ⁓ **Shafu (42:57):** Yeah, I think on the dating side, it created things like looks maxing or insults. All of these concepts stem from a lot of men that couldn't date or couldn't find a girlfriend. It's just a second order effect of these kind of things. **Gabe (43:05):** Hmm. **Shafu (43:25):** So yeah, on the technology side, I just know that it's happening. It's very obvious that it's happening. ⁓ How it's going to look like in five years, I have no clue. **Gabe (43:34):** Yeah. Yeah, yeah, yeah, the the Intel thing is interesting because because looks maxing is is you probably knows is like a genealogical roots or like grandson of of like some of the Intel forums online. We're going a little bit off, but. **Shafu (43:43):** Yeah. Yes. No, it's great. ⁓ Yeah, think it's just... Yeah, it's so interesting. It's just a... As I've said, like if you... At least in the online dating realm in the West, as I've said, like if you're not in the top 20 % of men, like just don't download the app. It just doesn't make sense. just doesn't make... Like the numbers say that. It just doesn't make sense. **Gabe (44:24):** Yeah, yeah, yeah. **Shafu (44:26):** So it's like this, you know, like they have this, concept of ascending, looks maxing. And ascending basically means you want to be in the top 20%. And that's also like, it's very relatable to like the permanent underclass. You know, it's just in a different, in a different setting. So, so yeah, we're going to live in a world that is more extreme in every aspect imaginable. **Gabe (44:31):** Yeah, yeah. Yeah, yeah, Yeah. Got it. Okay. So on the capitalism piece, there is a kind of like this other thread, Where, and you might have talked to it a little bit, but to nail on it again, the commoditization of these services, right? Do you think that if there's an EXA and there's a parallel and there's like a marginal improvement between the two, or maybe if you're using the services, you can't even tell who is who, right? I think with like a Claude and a Codex, like heavy use, could literally feel the personality differences and the speed. And you know, you can look at an ⁓ output and tell if it's like Claude or Codex, probably. ⁓ With some of these services, especially using like tons of different services, I'm kind of thinking that these things commoditize not into like the brand, but into ⁓ almost just like, you know, I need to do a search and it needs to be like an enhanced search. ⁓ **Shafu (45:31):** Mm-hmm. **Gabe (45:56):** That is kind of how I'm seeing this play out. Do you agree on that piece? Do you have any additional thoughts on this commoditization piece? **Shafu (46:04):** Yeah, I think it's already a commodity. ⁓ It's just going to be like, at least now you have a brand and stuff. Like you go to a landing page and you know, it has, you know, the vibes are nice and they're using like modern colors and stuff like that. ⁓ You're inside an agent, never going to see that, right? You're never going to see any colors. You're not going to see the face of a founder. You're not going to see their, you know, whatever. **Gabe (46:13):** Mm-hmm. Yeah, yeah. **Shafu (46:32):** It just, you know, use this to do this. So it's probably going to be even more commoditized as it even was before. **Gabe (46:42):** What do you think the things that stand out like, you know, traditionally these services use really nice ⁓ text and ⁓ logos and there's like all these different tricks like marketing. ⁓ but those no longer work like you said, right? Like the ads don't work on the agent. This visual appeal, like psychological element is mostly removed. What are the things that are kind of like these important variables that you wanna like drive, SEO maybe, right? To people's agents as they're indexing your services if you're a provider. **Shafu (47:00):** Yeah. Yeah, I think the biggest leverage is probably for people like Twitter or LinkedIn or stuff like that. They just own the data and no one else does. So if I want to get data from Twitter, I have to go through the Twitter API. If I want data on LinkedIn, I have to go through the LinkedIn API. Even if their landing page doesn't look nice, even if they don't have nice colors, it doesn't matter. Just the data is so valuable. So in that case, it's funny, like all the AI models, all the AI frontier labs, they need all the data in the world. And it's probably for the consumer, it's very similar. **Gabe (47:56):** you **Shafu (48:14):** I don't care about the brand, but this is the only API provider that gives me that kind of data, so I'm gonna use them. ⁓ So I think that's where we're stand out. **Gabe (48:19):** Mm-hmm. And where do you think right now, and maybe you don't have to think about it, but like where are the big use cases for searches right now or API calls, micro payments ⁓ on X4.02 or MPP? Like what are the behavioral trends that agents on behalf of people are actually doing that are useful tasks and not like some meme coin or something? **Shafu (48:48):** Yes. So the biggest use case that we see right now from people actually using our stuff is lead generation and enrichment. And they use it all day. They use it all day to find companies they can sell to. They use it all day to find customers that they could sell to. They use it all day to find the LinkedIn, email, Twitter. **Gabe (48:58):** Mmm. **Shafu (49:15):** of a specific person, which is enrichment. So that I would say right now is the biggest use case for this. **Gabe (49:24):** Got it. Got it. And in that process is you have like a business service. you want to identify companies based on a criteria, right? And then the agent searches for these types of companies. Maybe it calls like EXA does a website search and pulls back 100 companies. And then you need to use the Twitter or excuse me, the yeah, or the Twitter endpoint, right? And find their Twitter accounts, right? So you can do the DM and then you have ⁓ emails. You need to get the emails for the company and then you can use a email endpoint like Hunter or Apollo, I think you said, was the other one. So now you basically have all the sales diligence and you're saying this is the most common use case right now for agent commerce using X4.02 and MPP. **Shafu (50:13):** It's just what people actually spend money on, which is the only signal you should look at. Yeah, people spend a bunch, a lot of money on this right now, ⁓ especially in Pancho, which is a product we haven't even announced yet. ⁓ So these are people that we manually onboarded and they love the product so much that they spend a lot of money every day on. on lead generation and enrichment. **Gabe (50:45):** For things that are available, I've seen like a lot of like meme coins and services that are wrapped in tokens. I think more recently you said something online like you think that meme coins have like destroyed the credibility of a crypto. Can you speak a little bit about your thoughts about that? **Shafu (51:05):** Yeah, yeah, it has. ⁓ And it's just from from my own experience, because we talked to a lot of potential API providers ⁓ about integrating X for two or MP MPP. And we can't we can't mention crypto. ⁓ We don't we don't see ourselves a crypto company. We just we're just using stable coins to facilitate payments. But we're trying to stay **Gabe (51:23):** Mmm. **Shafu (51:35):** as so much away from that when we talk to like API providers because they would just, they would be like, okay, I'm not interested anymore. And that's how bad crypto reputation has become. ⁓ I can give you an example. And maybe this is just interesting. So the X4O2 standard was defined by the guys at Coinbase ⁓ a year ago. **Gabe (51:44):** Mm-hmm. Yeah. **Shafu (52:01):** And we started looking into x-402 maybe six months ago. And there were like 2000 maybe x-402 transactions. And half of them were like coin-based integration tests or something like that. A very low number of actual transactions. ⁓ So we built x-402 scan, which is a explorer for the x-402 ecosystem where we list every x-402 server and resource and people can add their resources. And we went from 2000 or like a couple of thousand transactions to a couple of millions ⁓ in like five days. Which sounds very impressive, but 99 % of the volume was like meme coin shenanigans. And by meme coin shenanigans, mean people created X servers that will give you meme coins through XOR2. ⁓ And we're not interested in that kind of stuff. And it took our site down, it took the Coinbase API down because we haven't read it too much. ⁓ So that was a fun time. ⁓ The other stuff, he's back. **Gabe (53:04):** Mm-hmm. Yeah, give me one second. think my camera died. One second, one second. **Shafu (53:23):** Yeah. I got I got it, **Gabe (53:42):** Yeah, sorry about that. That was like hanging onto my seat. One second. **Shafu (53:46):** No worries, no worries. nor is it **Gabe (54:58):** So the hammered side I think is probably a place to to pick back up. **Shafu (55:06):** So yeah, they launched a bunch of servers that would give you meme coins through X4.02 which is a terrible mechanism to launch a meme coin but whatever. So the thing is, okay, so 99.9 % of that volume or number of transactions were meme coins, but like 0.1 % of that was people actually using X4.02 to to buy interesting API resources. And it's funny, like a lot of people would like screenshot X for a two scan statistic and they were like, you know, everything is down 90 % what happened. And the thing that happened is all the meme coin stuff went away and we grew the actual API usage by 10X. But it just, yeah, just a fun story. And we had a lot of fun. for like five days in the office just scaling up x422 scan and stuff because we got hammered by the crypto people but yeah. **Gabe (56:17):** So in calls, you guys are not bringing up crypto at all. let's say you have a sister company, right? And they need to launch ⁓ the same product. What are you going to tell them before they go into a meeting so you can get the clothes and explain that these are crypto rails without blowing up the opportunity? **Shafu (56:22):** No. Yeah, don't mention the word crypto at all. If you have to, can say stable coins. Don't mention the word blockchains, stuff like that. know, we're ⁓ using stable coins ⁓ to save you 30 cents on every transaction, which goes to the credit card provider. ⁓ And that's the pitch. ⁓ But yeah, it's funny, Gabe. I spent a lot of time in crypto and I think when you're in the crypto bubble, you don't see that. But once you step out of the bubble, you're like, Jesus Christ, these people hate crypto. ⁓ Which is understandable because all they can see is crypto, meme coins, scams, all that kind of stuff. ⁓ **Gabe (57:21):** Yeah. **Shafu (57:38):** Are you familiar with Channel 5? **Gabe (57:41):** No. **Shafu (57:42):** Okay, so Channel 5 run by Andrew Callahan is like a YouTube channel. Very, very interesting guy. And he interviews people and stuff. So he was interviewing the Hoctua girl. You know the Hoctua girl? Okay, so he's a great guy, very open-minded guy. And the Hoctua girl was like telling him about, because he was asking her about like the crypto coin, crypto token that she launched. You know what he said? **Gabe (57:57):** Yeah, yeah, yeah, yeah, yeah. **Shafu (58:12):** He said, anyone that mentions Krypto to you, don't talk to him. And he talks to like the craziest people on planet earth. know, it's like extremely open minded guy, great guy, I love his content. But if a guy like that tells you that, you know you have a reputation problem. Something is completely broken and you just need to rebrand the whole thing. **Gabe (58:20):** Yeah, yeah. Where do you think is, so you're navigating the rebrand. What do you think is the best way to do that? Is it to just avoid crypto? Is it like to carve out kind of like a new narrative or like a new category? Like how are you navigating that? And then how do you think is like the long-term solution for that? **Shafu (58:57):** Yeah, I think you have you have two things in crypto you have speculation and you have payments We just need to focus on payments ⁓ Because stable coins are awesome, know, I grew up in Egypt Getting access to US dollars in Egypt is insanely hard and sucks You know like sending money to Egypt receiving money like US dollars. It's just a very very bad experience **Gabe (59:16):** Mm. Mm-hmm. **Shafu (59:25):** And stablecoins fundamentally solve that. So that's just one example. Stablecoins uniquely enable agentic commerce through X4 to an MPP. just so you have this, you have the payment rails, which is great, just uniquely enabled by crypto in general. And then you have the speculation stuff, which, you know, it's just a big casino in my opinion. and people see it like that. And you know, there are a lot of scams inside the casino. ⁓ So yeah, focus on the payments and less on the speculation stuff. **Gabe (1:00:05):** Got it. So for you guys, as long as you're kind of navigating as we're gonna lower costs and you don't have to deal with the credit card providers, then they're pretty okay with ⁓ working with a payment service provider that's using crypto. **Shafu (1:00:23):** Yeah, it's not even crypto, it's it's USDC. You know, it's stable current, it's not more than that. using USDC to facilitate X4 to transactions. That's all we use crypto for basically. **Gabe (1:00:28):** Mm. Got it, got it. Well, yeah, man, Shafu, thanks for coming on. ⁓ Last thought here, what is your big ⁓ AI prediction that no one's talking about right now? What do you think is gonna happen? **Shafu (1:00:55):** oof AR prediction that no one's talking about? AI predictions that no one's talking about. I would have said, you know, agents spending money on the internet, but everyone's talking about it now. So I haven't thought of anything new to be honest with you. Just, you know, like we're working 24 seven or just making agents spending money on the internet for the last six months. But yeah, anyone that is interested in agent e-commerce and wants to grow this space by a thousand acts, that's basically the mission of the company right now. **Gabe (1:01:10):** Yeah. **Shafu (1:01:34):** Please reach out, know, we'd love to help you in any way. ⁓ Go try out Poncho if you want to experience like this agent e-commerce future. Try Poncho.com. ⁓ And yeah, please find me if you're interested in this stuff, Twitter, on Telegram. Shaffuz Urex. **Gabe (1:01:52):** Yeah. Perfect, Before letting you off the hook, I'm gonna tighten it up a little bit. For all the stuff you're working on, what is like the biggest alpha that you've seen in the space? **Shafu (1:02:00):** Yeah. **Gabe (1:02:08):** ⁓ For me, I've been messing with Hermes lately with the harness and I think there's some cool stuff there. yeah, where is the alpha? Where is the opportunity when it comes to the payments? Maybe like what is no one building that needs to be built and that you're kind of seeing bubbling up ⁓ or some opportunities? Yeah. **Shafu (1:02:31):** Yeah, the opportunities is pretty simple. It's just take an API that you find interesting and put it behind XFOR tour MPP. That's like the biggest leverage thing you can do. I know a lot of people that work on agent credit, agent credit, agent reputation. I don't fully get that. These agents are paying one cent to access an API. I don't see where credit comes in. But maybe once they buy more expensive things than maybe. But I think a lot of people are wasting their time there. But yeah, it's pretty simple to get into this space. Because you know, agentic commerce space is just brand new and it's very, very small. So we had someone recently put like all the pedal maps behind X4-2. So you want to look for a pedal match. You can look that up through X4-2. Some guy put like half of sub stack behind X4-2. Another guy put like the city bikes with like electric bikes in New York behind X4-2. So I could be like, hey, I'm in. Williamsburg, where's the next available CBuy inside my agent? So that's what I mean. That's what's awesome about like open agent e-commerce is like, he didn't ask us, you know, he didn't ask Coinbase, he didn't ask Stripe or Tempo or anyone. It was just, have this API, I'm going to use this open standard X for a tour MPP. And now every agent with the wallet has access to it. **Gabe (1:04:17):** Got it. Well, cool man. Shafu, thanks for coming on and yeah, I hope to have you back on too. **Shafu (1:04:24):** Yeah, thanks for having me on. This was fun. See ya. **Gabe (1:04:27):** See ya. All right, man, I got the recording done. Yeah, man.