Hello everyone. Glad to be here. My name is Yuga Kohler. As mentioned, I am the Head of Engineering for the Coinbase Developer Platform. I'm Eric.
Our personal meeting, Yuga personal meeting, Yuga , is getting more people than usual, but yeah, Eric, uh, was my former boss. I just took up this position. So we joke that this is So we joke that this is essentially one of our personal meetings that we had. Um, we're here to talk about agency fees. Um, you know, it's a buzzword.
You've probably all heard a lot about this. Uh, we think about this a lot at Coinbase. And we have here one of the original leaders of agent payments, the founder of the X42 protocol. Um, raise your hands, how many of you have heard of or know what X42 is? Okay, I would say it's about 60%, something like that.
That's pretty good. How many of you have made X42 transactions? Okay, pretty good. And how many of you had an agent perform that transaction instead of pressing a button on an interface? Couple.
Okay, that's pretty good. I think maybe my main takeaway for those who heard about X42 for those who heard about X42 , you know, is that the sales funnel didn't convert very well convert very well . Uh, try the transaction on X42. Give your agent a wallet, five dollars and, uh, experience the agency business for yourself. I think a lot of agency commerce right now is theoretical, and, uh, people have their say on it , but they, uh, they don't realize that this is happening hundreds, tens or hundreds of thousands of times a day, and, uh, it's actually extremely useful right now.
So there are a lot of gray wallets. There are many ways to give Claw, CodeEx or something else access to funds, and being able to use x42 base mcp is pretty good. There are others. Uh, but the conclusion: make an agency transaction agency transaction . Yes, I mean, that's the main thing for us.
Uh main thing for us. Uh , Anthropic Fable came out today . Clearly these models are accelerating, and while agency payments are in their infancy, I think the total payments per year are around $15 million. XRO2 just reached its 100 just reached its 100 millionth transaction. We also expect about 50 million transactions per year.
But obviously the premise of agent payments is that AI agents , when they come to life, will have to be able to make payments, ahem, without that capability. They may not have access to the data. They may not be as useful to humans. The X42 is the first of its kind, ahem, for this use case. Maybe Eric, could you just really quickly, um, describe how X42 allows this?
Yeah, so the original sin of the internet is that there was no standard way to announce a payment, regardless of the rails underneath. For example, if the website you are interacting with is integrated with Stripe, Adion, or Checkout.com, etc. All APIs are different, all payment methods are different. So X42 provides a clearly printed scheme and a standard way of declaring how you want to receive payment.
It doesn't actually require blockchains. Simply put, blockchains are probably the best payment system in the world at the moment. So, if you standardize the format, you can create really simple programmable interfaces for agents. Agents are inherently probabilistic mechanisms. Everyone has probably seen their agent do different things during different attempts to fulfill the same request.
This is because every time an agent does something, it chooses a probability distribution of what actions it should take. And that means reliability can be tricky. So if you're trying to fill out 14 fields on a credit card form, that's 14 opportunities to make a mistake that will cause your transaction to fail. And so ideally you want software tools that your agent can use and that will work deterministically every time deterministically every time . And so your agent just needs to say, "Hey, I want to pay for this," instead of pay for this," instead of , "Okay, now I need to know your first name, your last name, your billing address, your shipping address, your credit card number, your CCV, expiration date, and, you know, all that," right?
Hmm, and as I said, X42 is somewhat agnostic to the underlying payment infrastructure. And that 's why we saw X42 on Salana, on the base, on Ethereum. There are probably about 15 different blockchains actively running on X42. And we also saw X42 on card rails. And you know, there's no reason why you can't make a payment through X42 if you really want to make the effort.
Yes. And this question often arises: why stablecoins for agency payments? Why not cards? Why are there no existing payment methods? Hmm, that's a reasonable question.
Eric and I have discussed this before. There are probably, um, a few convincing answers to this question. First, it depends on the form of the transaction. Um, charging fractions of a cent, especially in the case of microtransactions, gaining access to data API calls, etc., naturally gravitates towards blockchain-based stablecoin payments as opposed to cards.
Secondly, this is the overall programmability of the program. Instant calculations obviously also play a role. So, there are a number of reasons why stablecoins are a better fit. The CDP, Coinbase Developer Platform, for those who are not familiar with it, is essentially Coinbase's version of AWS. We provide a reliable infrastructure at scale for companies to build crypto integrations into their products, using the infrastructure we have been building for a decade or more.
Um, from an agency payment perspective, obviously we have a very strong core, um, product development team working on that. But there's another aspect, which is that , um, as more and more payments come online through stablecoins, there's this key infrastructure that's going to be important, and you've probably heard variations of that throughout the day. But it all starts with calculations; you need some level where funds can settle, whether it's on Base, or Ethereum, or some other ETH2, or some other network. You need stablecoins, ahem, that can transact for an amount. Uh transact for an amount.
Uh , you need storage. You need the ability to securely store these stablecoins. You need, um, a common infrastructure volume that, um, allows APIs, um, MCP allows APIs, um, MCP servers, etc. And then you need some kind of agent, sorry, enterprise level of control over this whole stack so that a company like Cloudflare or AWS can comfortably serve, um, using these APIs, and interestingly, you can expose that same infrastructure stack as an API for an external user experience or as an agent entry point. In other words entry point.
In other words , it is at the edge of possibilities that it is determined whether a particular crypto infrastructure package will be useful for a particular agent, person, or business. And that's how we designed our entire tech stack and API. If a company wants to enable payments, for example, for merchants, for example, to accept payments from merchants on checkout.com, we recently announced a partnership with Shopify – partnership with Shopify – another example. Usually their end users are people end users are people .
They pay, they allow payments to be made to merchants. But if they want to accept payments from agents, our APIs allow it with one click , one easy move. So, one counterintuitive way to think about agent payments is that it's just a configuration on top of an existing infrastructure stack. It's just another entry point into the payment stack that we offer. Yes.
And I think we're in this really interesting era of cryptocurrency, you know, maturity, like, there are people at this conference wearing shirts, ties, and jackets. Maybe it's not me, but you know, some people are dressed up. Um, and what you want, and I think what we should really collectively want, is for the transition to agents to be similar to the transition to the open web, where the Internet was built on open standards like HTML, HTTP, and so on. And it was a kind of foundation for people to compete on, offering the best products and services. And so you have places like CDP, my former employer, current employer Yugus, that are really building on open standards open standards , like X42, and the value proposition for the business is that it saves you from having to spend on your own servers.
If we go back to the web analogy, for example, if you wanted to deploy a website in the early stages of the internet, you would have to go and set up a server and get, you know, a high-speed connection to, for example, connect to the internet, and you had to spend it, and there are a bunch of things you had to do before you could offer value to your customers. And the same thing is happening in cryptocurrency now, right? Everyone wants to use cryptocurrency as a payment method because it is the best payment method that exists right now. But businesses need, they need responsible activity. They need a history, for example, I have a user and I have to do KYC for that user and I have a compliance obligation.
I have reporting obligations. I have to do certain things when I touch money. So you can see that we have a lot of companies competing now, and in addition to these open standards open standards , one of which is X42; but overall, cryptocurrency as an ecosystem—it's, you know, this open spirit, uh. These companies can now compete on a level playing field to provide these services on a large scale to businesses that want to take advantage of the excellent infrastructure. One of the great examples of what Eric is talking about is that we live in a very interesting time, as he said, where this confluence of open and closed solo developers and guerrilla cryptocurrency ventures and traditional companies is merging, and we all feel good about this space.
That's why this is such an interesting, dynamic time. But a good example of agency payments that this alliance demonstrates is the recent launch of AWS Agent Core, which is a core offering in Amazon's AWS base package. It's essentially a deployment and development tool on AWS that allows developers to run agents, and within that agent core is a key payments component that allows agents to pay, and X42 is a first-class payment method. And we see these developers building things like marketing agents that collect data or creative visuals from models using X42-enabled endpoints. They are research agents who compile reports on financial data or other use cases.
So, it's happening, and it's real. And it's exciting to think that something as institutional and fundamental as AWS is already at the forefront of exploring how agency payments can be used. And they can do that because X42 do that because X42 is such an open standard, and CDP in particular provides it, but other companies also provide the infrastructure and hardware needed to use it without knowing too much about the underlying cryptocurrency. Yes. And whenever you try to explore new horizons, there's a certain ambiguity and uncertainty, right?
And I think that's probably the number one question I get asked by large enterprises when I talk to them about X42: what's the compliance history? What do we need to do? Everyone is familiar with, for example: “Oh, we have a PSP, and we know the scheme, and we do X, Y, we do one, two, three, and then we say, ‘ then we say, ‘ Done,’ and our reporting obligation is this is this .” And so on, and I think people think that cryptocurrency hasn't come that far yet , but it's actually come incredibly far . A little-known fact is that the X42 has a component called a " called a " mediator".
An intermediary is an optional, i.e. optional part of the specification part of the specification that allows you to delegate some operations on the blockchain to a third party, and this significantly lowers the barrier to entry if you are a business. This means you don't need to become an expert in interacting with blockchains or managing hot wallets. You can delegate all of this to someone else who has already developed this competency, and there is a middleman in CDP .
This is very good. This is one of the biggest. There are a bunch of such intermediaries competing, and one of the key advantages of a CDP intermediary is that, being authorized by KYT and OFAC, it checks all transactions that go through it for Coinbase’s level of security and compliance. And this helps you to some extent , I'm not a lawyer. I don't want to give people legal advice, but it's roughly the same compliance obligation that you have as a business, which is to not act with the intent of understanding your transactions and not interact with sanctioned entities.
So there are things entities. So there are things you can do, like removing barriers to entry and making it safe by default, which dramatically lowers the barriers to experimentation for enterprises. So I think big business tends to be conservative not without reason, not without reason. And we're starting to see companies realize that it's not as high a hurdle as I thought. And maybe it's worth it, worth trying and seeing how these things work.
Yes. Nevertheless, we are very happy. If anyone from GIB is in the room, we're glad you're using X42 with CDP using X42 with CDP . Um, but it's a lot like that magic moment, like, I remember when OpenClaw came out , and then, uh, MoltHub , and then, uh, MoltHub , uh, or rather, Moltbook, and there was that magic moment where you could just point your agent to that MD file that was hosted online, and he just set everything up on his own. Um, a similar moment, I think, was with XRO2, that as institutions get on the chain, uh, as this technology spreads throughout the corporate stack , people are buying in more and more.
Um, we're coming to the end of our conversation, so we want to reserve a few minutes for some heated thoughts about the introduction of payments. So I will express mine. Um, my point is that if 2025 was the year of Vibe coding, I think 2026, maybe early 2027, we'll see, but really 2026, considering all the upcoming IPOs, will be the year of Vibe trading. There are echoes of 2020 here , but the difference in 2026 is that all this trading will be done agent-based, and there will be this virtuous cycle of people wanting to get in on the action, you know, preipo stocks and all these things that are being made possible by prediction markets, decentralized exchanges, Coinbase, etc. They will want to engage agents connected to various data sources to be able to make decisions for them.
Agents should be able to pay for these data sources themselves. So, without human intervention, and lo and behold, X42 will be the best payment method for this. So, these agents will need a crypto wallet, and then they will trade from these same crypto wallets to same crypto wallets to trade and make profits with the help of their people. Um, the reason I think this will happen is because actually, you know, we have a product similar to this at Coinbase, and it's extremely fun to use. Eh, you don't need to know too much.
Uh, if you have a thesis, you can delegate a lot of authority to your agent. Um, and this seems like a very organic use of agent payments, where, um, access data is a critical fuel for intelligent decisions that an agent can make. So, uh agent can make. So, uh , if vibe trading becomes a phrase, you heard it here first. I like it.
I like vibe trading as next. I think next. I think we only have, um, two minutes left, so I'm going to give a very heated opinion and then, uh, and then I'll quickly exit stage left. Uh, I think my biggest thought right now about AI in general is that it's shocking how commonplace AI is. We've created artificial intelligence, and the best we can come up with right now is, uh, productivity maximalism, which is, so to speak, extremely mundane extremely mundane , you know?
And it makes me think that we're living in an era of the internet, where "we share scientific papers," where, you know, the internet started as a way for scientists to literally share scientific papers with each other. And what we're all feeling now is that the internet has become much more interesting and, um, more powerful. Not necessarily, you know, the fundamentals of the channels are the same, but it's the capabilities that make it interesting, it's the Internet that's become more interesting not only with the increase in bandwidth between computers, but also with the capabilities of, for example, JPEG rendering. Oh, you can create interactions using JavaScript. And, um using JavaScript.
And, um , I want to go back to that and go back to why this is agent commerce, because I think payments are one of the things that makes artificial intelligence less mundane and a lot more interesting. It's just more interesting if your agent can access hidden information that isn't publicly available and you know what to do. You can think of it this way: if your agent way: if your agent is just tokens from a faucet, it is the same as everyone else because it has the same search capability. He has the same opportunities. He is trained in the same corps.
For example, if Yuga and I use, you know, Chat GPD 5.5, we get the same things. But when you give your agent access to secrets, to proprietary things, to, proprietary things, to, you know, the interesting parts of the Internet that are n't hidden, because they're real, because they're valuable. Is your agent becoming as interesting as possible? And so I really want to make agents more interesting.
We will leave you with this . Thank you very much.
Automatic transcript — names and jargon may be misspelled.