# AI and Stablecoins | DC Fintech Week 2025

- Channel: [DC Fintech Week](https://streameth.org/dc-fintech-week)
- Date: 2025-11-21
- Watch: https://streameth.org/watch/yt-mmvHFPkq008
- YouTube: https://www.youtube.com/watch?v=mmvHFPkq008

## Transcript

[Music] Fascinating topics. We've got a great group today. I'm Kai Sheffield. I'm the head of crypto at Visa. I think we're going to go around and everyone's going to introduce themselves and then we'll we'll dive in. You want to start? Al >> I thought I was >> just just introduce >> coming in hot. >> All right. So, uh hi everyone. My name is Alan from uh from Singapore from the MES. So um great to be here. Um so I I look after digital assets and uh the work that we do on AI. So glad to be on this panel and share what we do. >> Hi everybody. My name is Tylen. I'm the co-founder and CEO of Payman. We are building the trusted way AI will interact with money. Uh we are uh an early stage startup. Have been around since uh 2024. um and excited to to be here with this incredible panel. >> Hi, I'm Abhishek Vadinatan. I go by Abby. I'm chief legal officer at the Near Foundation. Um the Near Foundation supports the growth and development. It's a Swiss foundation. It supports the growth and development of the Near Protocol. Um Near Protocol is one of the most uh widely used blockchains by monthly active usage. Um and uh is a scalable and user-friendly blockchain uh kind of purpose-built for for agenic commerce. and we'll uh surely dive into that a little bit today. >> And my name is Rich Whitman. I run crypto strategy at Google Cloud. Um and our team is primarily responsible for a number of Google's crypto efforts as well as thinking about how digital assets and artificial intelligence can play together. Uh we recently launched a protocol called AP2, which is an agent commerce protocol that we'll talk a little bit about today. But that's uh my remitt. >> Yeah, thanks for having me. >> Let's dive in. Let's start there. like last month you announced this protocol. What does it actually do? What are the use cases that you're focused on? And then as a part of the announcement, you also referenced an extension for stable coins. You know, how do you see stable coins fitting into the broader work Google's doing in the space? >> Yeah, so the the I'll just give you some background. So, um the AP2 protocol is really an extension of uh another protocol we launched called agentto agent. Um, and the agentto agent standard is really a common language um, uh, instructions for anyone who's building an AI to be able to equip that AI with the ability to interact with anyone else's AI model. Um, so you can imagine if uh, a retailer is building an artificial intelligence or an LLM to talk to customers. Um, that agent may also want to talk to a customer's agent um, at some point in time. They need a way to communicate with one another. Now, um, naturally, if we're thinking about retail and user, uh, ultimately those agents are talking to one another, hopefully to engage in commerce, right? Um, and so if you're going to have a protocol for these things to talk to one another, you also need to think about things like identity and payments. And so, um, the AP2 standard is essentially an addition to that, which allows agents to make transactions um, and receive payments from one another. uh ideally for uh you know agent any agent commerce purpose. It could be for physical goods. It could be um to get access to digital subscriptions or other things like that. Um and so why do we think about stable coins? Uh I think if anyone here is in the payment space, if you're a regulator, if you uh are in the AI space, one of the big challenges is how do we get some of these uh you know AI related applications to play nicely or integrate with current payment rails and payment systems. And um there's a lot of exploration that's happening around that. But given the the legacy kind of I'll call it like technical debt that exists, it become really difficult to actually have AI agents interact with, for example, bank APIs or APIs from traditional Rails. One of the things we saw as an opportunity around crypto is sort of the same thing that I think we all are familiar with now. It's been, you know, just over 10 years is that crypto is really in and of itself an open payment standard. And so, um, if I have an agent, I think we should be able to use the traditional rails. Um, but we also want to make sure that those agents can use new forms of payment that allow for new forms of business like micro payments, um, and borderless transactions. And so, that's one of the reasons why we thought about the X42 standard um, which is a payments protocol used for stable coins and made it compatible with AP2 out of the box. Yeah, I want to come back to the use cases, but but first you Abby, for for those who don't know, like what was the founding story of Near and you what's the relationship and connection to the birth of generative AI, you know, connected to to Near? >> Yeah. So maybe to take the the latter portion first, the our founder Ilia Palasukin back in 2017 was working at at Google Brain as it was then called and was one of the co-authors on attention is all you need. So that was the paper that ushered in really the generative AI era um by by kind of creating transformers and you know the T and GPT. So um you know very very deeply connected with with really the the genesis of of of the era and the world that we live in today. Now um following his his time at Google Brain you know Ilia and his co-founder Alex Skidenov actually established Near AI. So near was actually an AI you know project first and their goal was to really teach machines how to code and you know today that that reality is here with you know apps like cursor. Um but at the time in 2017 this was was a novel frontier research area and to power that they actually were crowdsourcing uh research from around the world and to do that they needed to be able to pay these people uh and they looked at the current solutions available then which were you know PayPal for example which wasn't really suitable at the time or Ethereum um which you know had uh had been had been established but but was not really scalable or or kind of cost appropriate for for the the types of payments that they were looking at which were you know to to a variety of people around the world. And so they actually established near protocol um at the at the time. Uh and the idea behind it is you know to create a scalable user-friendly blockchain that they could use to actually pay these these academic researchers. And that project kind of took them um you know in a bit of a different direction. And today Near um is is both scalable, you know, fast uh and and supports, you know, crosschain interoperability. So you can actually use Near to transact with a number of of different blockchains. But I guess the the story to to TLDDR it is, you know, Ilia was was instrumental in in establishing the T and GPT. Um, and that journey really took him to uh uh to to actually create a a a scalable uh blockchain to be able to to crowdsource the the type of research that was necessary for teaching machines to code. >> I think that's that's one of the most interesting use cases and and we've seen this uh in terms of early adopters of stable coin payments. Companies like scale.ai, AI, you know, who have been helping many of the models be able to get access to data, data labeling, you know, if they're paying, you know, thousands of contractors and freelancers across the world, how do you spend send a dollar from San Francisco to the Philippines? You know, many existing payment rails haven't really been designed for that. And so, we've seen more and more of these examples of companies like scale using stable coins for for payouts. Uh, but Tylen like your story fascinates me. When we met, you were working in crypto. uh I think you were on paternity leave you working in crypto you know building crypto infrastructure and then you started payman to build uh payments infrastructure for AI agents can you talk more about you know some of the use cases what problems you've been looking to solve and you know kind of your background in crypto you know how you approached it when you started >> yeah so almost two years ago uh there was a big shift in where money was going in VC where it was going to AIS to basically replace people And uh I was about to have a daughter. So I was like, well, this doesn't seem like a like a good future for my daughter to like grow up in necessarily because everyone's saying AI is going to replace all humans. So I was like, well, how do you like align incentives between AI and humans so that we don't totally get replaced? And then the thought was, well, you know, if we're paying AIs to do a bunch of things, surely AI will look at us and say, well, you guys are valuable, too. how about we pay you to help us do some things? And that's when like the idea started to come around like how do you get AI to pay people to do things? And um basically uh don't ever give me 3 months of paternity leave because I probably won't come back. Um I ended up building the first version of the product and started to really probably do the first major use case ever of AI agents trying to go and do something in the real world. And that was along the lines of getting it to pay people. Since then, what we've realized is AI getting access to money, whether it's stable coins or within banks, is really built on the core tenant of trust and guardrails. Um, what we've seen is there's a lot of use cases where AI just moving money could be way better than what we're doing today. You can think of your own banking experience where maybe you need to go and pay your nanny $1,000 at the end of the week. In which case, you'll probably go you'll go to zel. You'll see you'll have to add her as a payee. You'll have to go and check if you have enough money in your checkings and savings. Like these are all things that are tasks that an AI is well suited to do, but the actual application hasn't been built yet. So, we're seeing a lot of early work there. When it comes to actually going out and using stable coins, it's basically the same thing. like you can go and wrap a lot of this existing AI infrastructure around the payments APIs, but can you make these payments APIs and AI agents work together? I'm sure we're going to have like a lot of time to talk about that on this panel, but that's that's where we're like seeing a lot of this early stuff. And then Allan from from the regulators point of view kind of what what's your reaction when you hear about you know agents transacting agents getting money you know how do you think about you know what controls what needs to be in place what guardrails you have to exist >> well first of all I think there are a few let's unpack this a bit right so there's um there's stable coins um and then there's AI each of them deeply complex deeply technical terms um we're still at a very nent I would say stage of development for each of this perhaps you would say the technology is very advanced but I think um the supporting ecosystem and the regulatory thinking I think that's something that we are work really working together I think both uh both in Singapore I think internationally as well um but just to give you a peak in terms of how we're thinking about this we've um within the MES so we've kind of have three distinct mandates we are central bank we're integrated regulator but we also have a mandate to develop the the financial center itself um in 2023 We finalized our stable coin framework. So um that's primarily for value stability. We are in the process of um making amendments uh legislative amendments to formalize it. Um so through the process I think we see uh lots of very interesting use cases um ranging from crossber payments. We see uh use cases around payouts on the corporate side as well. um interesting use cases around people setting limits, programmability, for example, conditions in which you can or cannot use certain uh reward programs at certain merchants and otherwise. I think where we see um AI agents coming to place, I think it's the next step of evolution in terms of how use cases could be unlocked. On the AI side, um we've been working on this program um which is really looking at um how AI is to be deployed in financial services. That's looking at the fairness um ethics accountability and transparency as core principles but taking that further u from a financial institution perspective how do you allow and accommodate for example the use of generative AI what's the level of accountability for example um on the AI agent side whose agent is it um if you are kind of thinking about it so the analogy that I would like to to imagine is that when we speak about blockchains a lot of people tend to think of it as kind of open highway where you have different kind of cars and roads or cars or buses kind of running around whether it's public, private, etc. But I think what we have with essentially AI or agents is really driverless cars. So you think about if the the the the the roads itself um if we haven't gotten for example the conventions around seat belts what's what's the expectation there the highway code so that if you are driving on in a country that's a right-hand drive versus a left-hand drive and figure out what's the convention there and then you add in driverless cars and then you I think these are additional complexities we need to think about not to say that we put a break in terms of thinking about it but I think um these are I I think the road is ahead of us uh really and I We need to approach this very systematically. >> Yeah, I love the analogy and we've been hearing this concept of self-driving money for years and it feels like that the innovation and advancements in AI is like finally starting to get to the point where, you know, you could see it working even though we're still very early. Uh, and then we're also at this point where you have the existing financial and payment infrastructure, you know, that is actively updating, evolving to be able to adapt to these use cases. And then you have brand new infrastructure being built and trying to figure out how these converge. So maybe Rich at Google, how do you think about Google's role of building new financial infrastructure, some of the work that you're doing within Google Cloud and what financial institutions, what companies will need if they want to actually participate in these flows. >> Yeah. So there's a there's a reason and I I didn't get to mention this the first time, but that you know the A2A and the AP2 standard are both open protocols. So we incubated them within Google but then the A2A standard has been donated to Linux foundation and then we've had contributors from over 60 companies um in the payment space uh you know merchants and uh even crypto companies even the protocols themselves kind of contributing to that core protocol. you know, our view about what what our role is is by virtue of the fact that um you know, obviously we have our own um AI ambitions around building our foundational models with Gemini. Um we have a a deep connection to users and merchants by virtue of Google Pay and commerce. Um and then we also have um a number of of kind of leading initiatives as it relates to the stable coins. Our view is we actually serve as a really unique platform to incubate what might be those highways and then open them up and allow a public private partnership and discussion about what types of cars should be there and what should their capabilities be. Um and along the way we'll offer solutions. So if someone says well I really like this open protocol but I'm not an AI company. Um well they can leverage uh an agent developer kit that we use. Um or maybe they want to use the open protocol um but they aren't really sure what what blockchain to use. That's why we want to partner with like the Ethereum Foundation or others to say look this is this these are the rails these are these are the the core protocols or or rules that you'll use and I think we serve as a unique kind of um neutral infrastructure player around that because people think of us very similarly in the cloud space right so people use cloud to run a regulated exchange or build a video game and um it's very analogous to how we've thought about and had success in that world you know trying to do that in the AI space as well or in the agenda commerce space specifically And to like give a tangible example of that like payman's AI agent that we use to move money for people at banks or things like leverages the ADA protocol because it you know when you talk to chat GPT and you put in a prompt it's not just one AI back there that's figuring out how to get it done. It's many different AI agents with different specificities on how to actually get the task or the intent done that you put in. When it comes to handling money with stable coins, let's say I want to go and send this crossber payment, you're going to probably have an AI agent that's going to ingest this prompt. And then you might have an AI agent, which we believe is going to be one that's going to handle a lot of money functions. These two agents will need to talk to each other. When they talk to each other, the money agent will tell the other AI agent, hey, maybe I need the additional like wallet address on what chain it is. in which case that AI agent also leverages the AP2 protocol and then you figure out how to make money move by just having two AI agents go and speak to one another. The other core thing is like you need to build these AI agents on top of these protocols to make any of this work. So like the roads are there but the cars aren't and like people need to be like building the Whimos and whatever other competitor there is out there to them. >> Well, and the roads may be there but are they connected? Right. And so one of the the issues is fragmentation, right? So you've got you know liquidity and in on a variety of different blockchains. This is a huge problem of interoperability which you know is often mentioned and I think you know this is something that we've been very actively thinking about at near and building around and so one of the the kind of products that uh is emerging from our ecosystem is is called intense and you know what it is we think of it as a unified liquidity layer that provides the connection between these these these different roads which you know are here to for you know unconnected. So that allows you in turn to use near to interact with you know a variety of different blockchains. So if you want to use your you know your your near account you can actually uh you know that that can then in turn trigger messages on Salana on uh you know on Ethereum uh on Bitcoin etc. And so what that then provides is again the connection between between the roads again to to use this analogy um and and in turn reduces the cost and friction associated with with this fragmentation. And so one kind of concrete example of this uh partner in our ecosystem just launched stable flow. And so if you go to stableflow.ai AI today. This is now, you know, one of the one of, if not the cheapest way to actually transact in USDT um and and and swap USDT between chains. And so, you know, you can now do this up to a million dollars for a buck 50, which is, you know, uh pretty well the lowest I've I've heard of. And so, again, we are really focused on building those those those roads. Um and and I think it's a it's an exciting opportunity but with a lot of sort of uh I I think questions to be answered around you know the regulatory um and and other implications. One of the things that I just to kind of like zoom out and and Kai, I know we've talked about this um privately is is right now everyone's kind of getting like this brain surgery view like we're all unpacking kind of the the nuts and the bolts of this, but from a from a user perspective really what this is going to look like is the way that you're you're shopping or using a like a browser today. Um I mean if you think about it, somebody used this really great analogy which is like what's the difference between the Nike and the Reebok websites? The shoes face the different way. That's it. Like there's literally no difference. And you would think like that's a perfect area for competition is like if I am Nike and I want to outsell Reebok, I'm going to make the buying experience, the browsing experience more intuitive, more personalized. And and really people haven't done it. And I think that's where the uh AI uh you know kind of the LLM capabilities come in. But as a user, you're just going now instead of typing it into search or something like that, you may be typing it into, you know, your Gemini interface or chatgpt and then getting a list of options and not just options, but recommendations alongside of that. And so, you know, there's a whole lot of plumbing that goes in that. We're kind of talking about these layers of the payments layer, the information layer, the auditing layer, um, and and then like the actual agents themselves. But for users and for folks who are in the policy space, I think the way that we should be thinking about this is sort of it's in the next evolution of something that we're all doing every day with a browser and has now become basically invisible to us. Um, and the goal is that a lot of this again becomes invisible to the end user. So they're not wondering, hey, does my AI have this form of payment or that form of payment? It just has a balance. Um, and then it uses smart routing to send an optimal form of payment for the cheapest cost to the merchant who wants to receive it. >> Yeah, I I think this is something that that we've learned at Visas. There's so much that needs to happen before the transaction. You have to have, you know, identity for the agent. You have to be able to have the merchant know that this is a good agent that was authorized to buy and, you know, it's not an agent that that's there doing something malicious. And so there are all these steps in, you know, handoffs that have to happen before you're ready for the transaction. I want to go back to you the example you were giving Thailand of you know this concept of when people think about an agent I think it's easy to think about as a singular monolithic thing it's the agent that's doing all this activity it sounds like the vision that that you see is it's actually even if you're interacting with one agent whatever task you give it they're going to need an army of agents behind it to go and figure out that are going to be specialized to accomplish that certain task and so we're seeing things like the model context protocol different standards where agents can talk to each other to help solve something, but there hasn't really been a payment layer to that. And so it seems like stable coins could have potential in this element of, you know, you give an agent a task, it has to find an expert. How can you find, you know, search through thousands of agents that specialize in that task and maybe it only costs a tenth of a cent? And so can you then pay a tenth of a cent to help complete that task? Uh, but Alan, how how do you think about this? Are we going to live in this world of modularity where an agent talks to many others? Do you think it's going to be more monolithic? Like what what's the the MAS view? >> I I think intuitively I think uh all of us like options uh like uh uh you know that we can mix and match. Um so that's one view but I think there's also the as um the aspect around economies of scale. So there's argument to say that you don't just because you can um doesn't mean that everybody's going to learn how to assemble your own car for example or fix your own furniture for good reasons. You might say I'm just going to work with an integrated um contractor or whatever and have this um assembled on my behalf for for me right for example. So I think the ability to switch out different tires of a car may be an option. Um just to stay on the analogy a bit. Um but just having the option may be good enough doesn't mean that you would necessarily exercise that option all the time. Um I think uh the other aspect I would just as we think about this is as stable coins develop the protocols develop whether it's on a on um on the AI agents perspective I think um I think there's a recognition that as this develop needs to be done on open and interoperable manner I think everyone would agree on both sides uh on the official sector and also on the private sector. Now the question is how do you achieve that right do you do that um one I think there's certainly the view of regulatory clarity the space for innovation then the question is around what the supporting infrastructure and it needs to be in place again different countries have different considerations things like identity for example is that to be provisioned by a single centralized um entity is it going to be provide provided by a few um or there could be a federation of different types of identity I think that's something that we need to think about so I would I would urge maybe thinking about the ecosystem that supports some of the core activities um and some of it is around um the role of the the public sector but I think a large part has to be the private sector and and uh I I think conversations like this are very important to bring this to our life. >> So so as these standards emerge and as the infrastructure develops uh for agents to make payments it seems like there there are like two ways you could look at it. One is how do the traditional enterprises incorporate agentic flows you know into their core business and what are the things that you could do that you couldn't do before and then there's another of how does it change what it means to build a company you know if you're starting a company in 2026 from the ground up are you going to have these single person billion dollar enterprises that are mostly run by agents uh and so Abby like we've talked about this before of in crypto there's this term DAO or decentralized autonomous organ organization, but it's never actually been autonomous. It was more just people in a discord who were voting with tokens. Like what's your view of are we going to start to see actually like self operating entities like you know that seems like kind of a weird world like what could that look like you with this infrastructure out there? >> Yeah. Well, it's a it's a it's a great question and I think we've you know to to kind of ground this a little bit, we saw recently the the integration of of ChatGpt and and Shopify. And so I think you know if if you take that as an example I I I could easily imagine a sort of autonomously operated Shopify store, right? Where the goods are procured by uh a self-executing smart contract and then you know bought or sold electronically or online. Um and that future I mean without a h a human in the loop a little further away but with a human in the loop with you know some kind of legal wrapper or regulatory structure around it I think is is is probably you know within the notsodistant future and you know I also know for a fact that many regulatory authorities and I'd be interested Allan on on on on your thoughts on this and whether it's something MAS has considered but I know from conversations with various regulators it's something they're they're actively thinking about and trying to encourage. The first step really being, you know, for example, what is the scope of authority for an agent that has the ability to to bind a corporation? What are the liability concerns associated with that? um what are the insurance requirements or fiduciary requirements if say the agent is managing assets on behalf of um either on behalf of an LLC with human members or you know actually investing on behalf of um as you you've probably heard a variety of asset management kind of you know agentic use cases. So, I mean, without opening Pandora's box, which I believe I may just have done, uh, I think these are are very very both both both interesting issues and ones that I think we will start to see um, you know, both regulatory formal regulatory engagement and and and perhaps further clarity on in the the kind of coming 3 to 5 years. But I think again the the two probably primary examples, you know, I think could you see an autonomous e-commerce store? Probably. Could you see an autonomous uh sort of asset management firm? I mean look already on a self-custodial basis near H has this capability. So um you know if you were to take that and and instead of having it sort of self-execute and and put it within a legal wrapper, you could very quickly have an agentbased you know hedge fund or asset management firm. But it but again I mean that that's going to come with some serious um liability risks for the the developers at at the moment without having a further understanding and will probably be more of a sandbox uh kind of exploratory uh exercise for for for the for the time being. >> Yeah, there there are some great stories about anthropic giving their model claude access to running a vending machine and you know what it chose to to procure and some of the things that it it did. So I think we're we're still a ways away. Uh but I want to go back to Tylen in in the early days of payment, you know, when you were starting with agents, you know, paying humans, like where did stable coins fit into that? I know you experimented with a bunch of different payment methods. Can you talk more about like what were agents paying humans for? What are some of the use cases that you saw? And then like were stable coins better? You know, did you think that existing payment infrastructure wasn't able to accomplish, you know, the payments that they wanted to make? From a regulation standpoint, it's a lot easier to open up a crypto wallet that can hold stable coins internationally, then you can go and partner with a bank. So, it didn't necessarily matter to us which one we went. It was more speed to market. And the speed to market that stable coins offers you is insane compared to going and working with a bank. Um that said, um because I came from a crypto background and knowing AI and how scary it is for people, going full AI and crypto was not like the best way to say you can trust me with your money like nothing's going to happen. um we leaned in to um we leaned into going to the regulators and that's why you know invest Visa investing in us and doing things like that really helped bring like I think a reputable place where you could trust AI with money and then to me stable coins are this like it gives people who are maybe scared of crypto a nice chance to like dip their wa like dip their feet into the water and see like what does all this mean. I'm not saying you guys are going to be buying like Pepe token like next month, but um where stable coins worked well was we could bring in anyone across the globe and have them be part of this like economy where AIs were really trying to figure out and it's not AIS, it's at the end of the day, the things running these AIs are all enterprises. Enterprises are trying to fi find value. AI is one way that provides immense value in tons of different automated ways and tons of different logical ways that we couldn't previously we could previously do we just couldn't do it at the speed that we're doing it today where all of these rules engines that you had to build before you can get that out of the box with a prompt also. So then if you can build AI agents that could do different things with money that would take your team years to build before, but now you can get it there super quickly. And on top of that, when you had money that when you had markets you couldn't access because it would take you years to also build partnerships, it would take you capital to go and do that. Now you have this these pieces of software that could be built so much faster than before. You have markets you can enter into before that you couldn't do. It would take years. And it's just like if this opportunity is so easy to see that the way that money will be interacted with all around the world and economies will be created like that's going to change in such good ways and I just get excited about that like super stoked. Rich, how do you think about it? I think there's this interesting comparison between core ledgers in traditional accounts and stable coin wallets. And if you're going to have hundreds maybe thousands of agents and you want to be able to spin up wallets, program in different rules around what they can do and what they can't do, you know, traditionally core ledgers weren't necessarily designed for that. They were designed for humans and you had one account or you had your checking and your your savings account. So do you see wallets being this new primitive that provide benefits for for agents? >> Yeah. So the the wallets and the underlying protocol, right? So uh you know, Ethereum or Near or uh any of those, right? And and the reason why that's great is because the the function of the account, which is the wallet essentially is already integrated or has an integration capability with the underlying the source of truth or the record. And the other benefit is that the source or truth of the record is is open. Um, and so if you think about like an agentto agent world, um, there doesn't have to be really this pre-existing relationship or this agreement to share this common database with one another like you might on a like a bank ledger, for example. Um, you don't have to be the customer of near to open a wallet and go and buy some stable coins that are on the near blockchain. Um, I can't really open a Chase account without like going into a Chase bank and talking to them, right? And then then I end up as a little entry on their ledger. And then everyone who wants to accept a payment from me from that Chase account is really talking to Chase and saying, "Does Rich really have $100 that he's saying he wants to send me?" So in the protocol space, like agents can check that if they have a wallet now, agents could check whether or not an agent has a form of payment that they can accept. Um they can actually verify in an open database that the transaction has taken place. And in fact, the agent that's um sort of on the hook for performance of some kind, shipping the product or whatever, can actually store uh a transaction hash again on a public blockchain or a public protocol. Um that would allow anyone to audit that. So yes, I've sent my $100 to the agent. Um it's verified that on chain. Has the agent actually performed? Did it do it show on chain that it executed the transaction that I asked it? And and I think that's where not just the wallet but the the ledger or the protocol itself uh the crypto protocol itself becomes I think a primitive that lends itself really well to these types of interactions in a way that traditional ledgers or even closed ledgers that you might see like in core banking or in traditional payments it it becomes a little more difficult right because if I use a Visa card really no one else is seeing that right no one else is able to see that the only way that you can confirm that as the merchant is you you see the payment right you get the Um, and I think that's where for agents it becomes really interesting because I don't think they'll have that level of visibility um, in traditional ledgers. >> So, so it's interesting that there there are trade-offs here. I'm curious how you think about it, Alan, of on one hand you have this transparency where every stable coin transaction on a blockchain today is public. You could see it. So, you can audit it. You can make sure you know that it happens from a regulator standpoint. You know, that gives you more visibility into it. But then as an enterprise or as a consumer that everyone could see it and so like you don't have any privacy and if you're trying to build a business and you have an agent that's transacting on your behalf and there's a dune dashboard that's up there showing the entire volume that you've ever done. >> No more quarterly reports you just are looking at the payment flows based on >> it's an entirely different game. So how do you think about that as a regulator of like privacy versus transparency kind of where blockchains are you know today? I think I think there's a lot of uh research done on privacy pres preserving technologies. Um intuitively we think we want accountability. Um but accountability and transparency how do we merge the two? I think that's very important. Um but maybe just to just to kind of build upon some of the earlier points. I I think we've been speaking about uh a lot of act actions actually done uh autonomously. Um the question is do we really need the whole workflow to be automated or perhaps um we still need at least in the current state of things a human in the loop such that it's the human would needs to come in at a point of confirmation when you actually say I'm going to sign off on the final transaction but perhaps the heavy lifting that's done uh up front could be offloaded to machine for example. So I think that's probably an easier thought process versus going fullon automated um right now. Uh I think it's perhaps also a question of getting comfortable with how things are. I I mean that's how self-driving cars anyway started. Um you want to allow for the optionality for somebody to um intervene as necessary. So I think that's probably where we see uh things actually emerge. >> Yeah. Oh, no. Go ahead, Tyler. >> To Yes, on that point 100%. Um, and this I've been trying to build autonomous agents for a year and a half, and it's very I I don't even know what an autonomous agent means at this point. I I've even questioned what an autonomous human is throughout some of these uh things. So it's been the there are key things that the logic can make better with an AI. The other part we miss with stable coins is um for example as we were trying to work with bringing AI agents to India to do more financial tasks. They don't want USDC in their economy. You know they don't want like it's very hard to get any kind of stable coins in there because all liquidity will go into stable coins. And like India has a very great payment system with UPI. So then for us it's more so the value lies within the AI and the AI agent and what it can do. And really if there's anything you take away it's how does this AI complete this intent this payment intent. Send $5,000 to my nanny. Okay. The AI has to have memory is na nanny is not I don't put nanny as my pay. There's a actual person that's the nanny. What's the cadence? Has this been a previous transaction? Is there a prompt injection that's coming into this? The really nice thing about this is you can see the chain of thought where that AI is going to try and complete this intent. And it in some ways it makes fraud like you wouldn't think it it makes fraud easier to manage because an entire chain of intent an entire intent is a chain of thought. So my whole thing is like if you're going to go and build with these wallets or do anything like making those wallets work with an AI that can handle money is very crucial. And that's like making the car work with the road. You're not going to build a you know 100 foot wide car and then place it on there and say go drive. Like this all needs to be created. And it's like these are very tangible things that are happening today. And like you're going to see it's going to be little subtle things. And to maybe beat another not the car analogy, but like websites used to Google came out with search, right? Websites all didn't have search. You couldn't even search for products on a website before. And then search started coming in. And then the way search had to be created was you had to create different data types that could go into search. You had to have better context. You had to make search better. It's the same thing for AI agents. >> Yeah. And I mean look, payments is one area. Coming back to this asset management example. Let's take the you know the the the kind of thought experiment of of uh kind of reallocation. I want to take USDC and put it into a money market fund. I mean how do I know that it's actually been allocated to the right money market fund? Um how how do I know that uh you know that that the sort of chain of reasoning and the result is actually kind of consistent with the input or intent that I I place. And I think this is another you know kind of research area at near AI right now is is sort of the formal verification. So is is you know can you have cryptographic proof that the you know result or output is actually consistent with the intent. And you know without this I think we're going to have we're going to be pretty bottlenecked in our ability to do you know again autonomous forget autonomous payments but also any kind of really autonomous action because we will not have the ability to have trust in you know in in the output or the action that the agent is is supposedly taking on our behalf and again a host of liability and other issues that arise. Um particularly if you know there's not someone to point to when the agent doesn't do what you thought it was going to do. um and you know you lost a bunch of money and you're very upset. >> Yeah. So maybe wrapping this up, I think you as we talked about before, like the transparency of stable coins, you don't really have to guess what's happening. You know, the the protocol X42 uh that I believe was launched maybe 6 months ago. Uh it's hard to even keep track of time in in this space. Six months was a long time ago. Uh there are a bunch of great public resources that that show dashboards. You can see every payment that's happened over it. And I believe it's done about $50,000 to date. Uh I think it's done a few hundred dollars in the last week. So question for for everyone here, how far is some of this stuff away? Are we talking decades? Is this two years? Is this five years? Like you know do you expect that we will see millions billions of dollars of stable coins being moved between AI agents and like how how far in the future we think? >> So I think the hum so we talk about the growth paradigm for stables. I think we're going to see that continue to go up and to the right and in what I'll call like the traditional human in the loop type transactions and whether it's for you know money management, treasury management, uh moving value across borders, that's very clear I think as it relates specifically to what we'll call like autonomous payments or agent commerce. I think it's still very early innings, right? To your point about X42 adoption is very very small. Um we're seeing a lot of micro payments, but I do think it's going to unlock new business models. My view is um there's something we we spent some time Allan alluded to it a few others did as well which is identity is the other key piece here um and that's I think needs to get solved for a whole host of reasons both to solve the downside risk but also to really unlock potential around trust and understanding like is this Rich's agent acting on behalf of Rich to make a purchase or is this a scam and that's still not been built out yet. So I think we're we're still very early innings. I say it's inning two. Um uh I think it's very very early as it relates to uh agentic payments. Um I think stable coins is late later innings for sure. >> I'll go with bottom of the first. Um but uh but I think look I would echo a lot of this and I think again putting on my legal hat for a second you know you talk about identity you know are we going to have KYA you know know your agent. Um what does that what does that ultimately look like? And I you know actually we we have some builders who are who are considering this and and actually beginning to build this out again with human in the loop such that it's it's clear that there's a chain of authorization from human to agent to conduct activities on behalf of that human. Um how about dispute resolution right I mean we've you know we've seen some of this um in terms of uh you know some of the work that circle has done for example um but what does it mean when machine has to litigate against machine how is that actually conducted is it an arbitration you know what what are the the means and mechanisms how do they actually submit filings um in in in the event of disputes um are those resolved by you know self-executing smart contracts or and are those respected by by you know enforcement um sort of mechanisms through courts or or otherwise. And so when you really think about the popoly of issues that are that need to be resolved I really think again you know maybe top of the second bottom of the first uh is is is is where we're at and again the the opportunity set here is is is quite quite massive. I really do think, you know, we talked about the billiondollar oneperson um business. You know, we're really going to see the world reshape before our eyes, but whether it happens in the next 2 to 5 years and or 5 to 10, you know, kind of remains. And we haven't even talked about robotics, right? Humanoid robotics also um together with this this space could completely reshape how, you know, to to come back to this autonomous business question, how businesses are are run. And so again, I always open cans of worms, but uh I'm going to now slightly close it. >> Yeah, on on that note, you know, thank you everyone. I think that's all the time we have. We're going to have to come back next year and we'll talk about robotics, you know, with AI, with stable coins. We'll just keep adding complexity to it. But thank you everyone. Appreciate it. >> Thank you. >> All right, everyone. Uh we've got about 90 minutes. Uh we will be back here at 2:10. Uh lunch is being served on the second and fifth floor. Uh and we will be talking about uh the future of AI governance. Nice.