Agentic Payments: When Machines Hold Stablecoins

Transcript

To start things off, uh can we have everybody introduce themselves? Uh Anil, I'll let you go first. >> Hi everyone, I'm Anil Nminti. I'm a senior solutions architect at AWS. And uh in my role, I work with the financial fintech customers, um the large cryptocurrencies of the world, the large uh decentralized uh exchanges of the world. And uh in the last couple of years I've u gone a little bit deep in the agentic AI space. Uh so I also help organizations uh ship AI u and just not from PC's but in production. So the the seesource of the organizations are happy. So that's about me.

>> Hi um Vincent Chak here. I am the founder and CEO of First Digital. Uh we're probably most known for issuing FDUSD uh stable coin uh out of Hong Kong and um introducing our new project now um finance district which is a platform around uh Agentic Economies and building around that. So good to be here. Thank you. >> Hey everyone. Uh I'm Phil. I'm the co-founder and CEO of Soho Pay. And uh at Soho we're building a onchain credit card for AI agents. Uh, and I recently just left American Express to build it.

Great to be here. >> Great. Hello everyone. Uh, thanks for still being here. Um, if you've been here already this morning, I talked about onchain and FX and stable coins this morning. Um, my name is Marcus. I'm founder and president of Mentor, which is an onchain FX platform allowing users to swap between different stable coins, between different currencies at the exchange rate, at the FX rate. And uh recently I was also appointed as head of stable coins at Cello Foundation. Um at Cello we are focusing more and more on onchain FX and on stable coins. Uh very excited to be here. Thanks.

>> Now I want to set the stage for the rest of the panel. Uh can you define what an agent is and what aic payments are? And this is an open question. >> Sure. Um, yeah. I mean, what is an agent? I mean, in my mind, I guess it's more it's just code, right? It's a program that that probably runs some loop where it's, you know, looking to do some job, right? Um, and then when you talk about payments, uh, you know, I think right now what you see in in like the X42 ecosystem is obviously like stable coin debit payments. you load a wallet and then that agent can spin that that balance at

any X42 endpoint. Um, you see a lot of like legacy players trying to get into the space where you can use your credit card and clawed. Um, and I think the next step of that is an onchain credit product where instead of the agent holding a wallet, it can just ping some credit layer and uh make a stable coin payment. >> Agents is an agency. Um basically we empower it to do things on our behalf. It has its own line basically you know um but the agents itself uh I guess it gives us a lot more efficiency in our time daily use and to build on top to allow the agents to make payments for us

through the agentic payments. I think it's it's really another revolutionary thing and of course one of the rails the protocols is using stable coins. So again you know like I think it'll it means different things in different parts of the world. Um people here in the US a lot of people have many different options when it comes to payments but for emerging markets this might be their only solution. So >> so yeah uh when we talk about agents uh agents take inputs that can be the prompts that you give to the agent they take the tools that you give to the agent and think of knowledge bases that you supply to the agent. tools can be uh APIs, access to MCP servers, access to databases

uh and maybe humans in the loop can also be a tool, right? So you give all of them to an agent, you give the intelligence using the large language model that you choose and let the agent run uh let the software run in a loop until the goal is achieved and uh the then the agent provides the output. Right? So that's how I see that as an agent. And when it comes to payments, uh these are transactions that agent payments are the transactions that are discovered by agents. They are initiated and authorized by agents also, right? So which means that you know there's a lot more under the hood that we need to follow and uh I think we'll discover that as part of the session >> and Marcus.

>> Yeah. Uh not much to add here. I think it was good good definitions of what what agents do. Um there was also a agentic summit yesterday on on this topic. Um and I think there was a lot of callouts for um agents uh to be built on open rails. Um and there was some data that the agents that we see today mostly actually for example purchase or engage in actually digital digital items. So often today agents uh buy uh cloud space or uh server capacity for you and because of that and because of agents um

code and act on behalf of of you they need digital currencies and therefore there's a close link between um stable coins and and agents. >> This leads very well to my next question. Uh what preference if any do agents have in choosing one stable coin over another? And why can't they just use normal dollars quote unquote? >> I think agents probably don't have a preference. The person who employs the agents maybe has a preference. And one part of that preferences depending on where that person sits that employs these agents. If they're in US then most probably they would have a preference for uh dollar stable coins. And yesterday we also talked about most of the large language model companies are sitting in the US. Do we need more than

just dollar stable coins? And I would argue yes, we need more because the users are global and uh if I'm I'm based in Europe, I want to use uh the euro otherwise I would exposed or my agent would be exposed to have x risk to to the dollar. Um and the same is true for many other other regions of the world. And then even for regions where there's high inflation, where the current currencies are very volatile and maybe people prefer the dollar to hold, if an agent engages in some form of credit for you, then maybe you also have a preference for your local currency because again, you would be exposed to have X risk.

>> Vincent, do you see um you have a stable coin, do you see uh these agents having a preference or choosing a one stable coin over another? Well, I think as Marcus mentioned, it's really the FX hedging. There is also you have to consider some of the currencies out there are closed currencies. So, there's a lot of countries with exchange controls and basically those are the countries that need stable coin more than the countries that have the open exchange controls, no exchange controls like Japan, Hong Kong, US, uh Singapore. But yeah, I think it's really um the preference of the person behind the agent. Um but yeah, it it doesn't make any sense if someone is living in

Philippines buying something from Malaysia and then having to convert from the Malaysian ringit stable coin to US then from US back to right to to uh the Malaysian ringit stable coin. So there's just too much there. It wouldn't make sense. You would be paying more than you would a normal credit card transaction. And yeah, maybe I'll add to this right. So right now I see that the humans who are building the agents have some kind of a you know preference but as we go into the next phase of uh evolution in software development life cycles we'll see software being created by agents right and LLMs as we see that today. So uh that's when uh there'll be more uh you know preference for agents that are

more reputable that have uh done x number of transactions right so that's kind of where we'll be heading into into the you know near future >> may I'm sorry maybe to add one more thing probably agents don't have a preference for specific stable coin they also don't have a preference for a specific chain probably it for the agent it's probably very important that It's fast, cheap. Settlement times uh are low, gas fees are low. Um and this is in contrast to what we saw a few years ago where there's maybe Ethereum maxis who just wanted to use stable coins on Ethereum even though it was maybe expensive or or Solana Maxis or or Celomaxis. I think today it's more

important that basically the settlement layers fast, cheap, uh safe, secure. Yeah, I would just add like I think there's probably an interesting future where like if you're living in the Philippines or you're living in Argentina and your currency is experiencing, you know, tons of inflation, maybe you have an agent that, you know, hedges that away for you automatically and yeah, they they settle in the local currency, but it's holding in dollars or it's holding in some sort of inflation hedge. Um, so I think that that could like really be beneficial to a lot of people in emerging markets.

>> And now, um, I'm curious, what are some key challenges, whether they be security compliance risk regulatory uh, that is faced when deploying these agents? >> I think the one of the challenges is that we're really in the early days and it's around security. You know, people there are there are people using the agents for the right way, but there's also the the counterparty, the other side. Um, your agents could be scammed. So, this is >> on a spending spree. >> Yeah. Well, at least you can control how much is in your wallet. You can give it an allowance. If if it spends over that, it'll it's up to you, right? But I the the real uh challenge is is that um the credibility of sort of everything that I

think everybody on this stage is building really relies on if there is any potential breaches, hacks that you know we may overlook or we're relying on third parties to build something that is hasn't advanced yet to that degree. >> Yeah, I I agree. Like if you look at like the X42 ecosystem, you know, there's tons of endpoints, but your agent doesn't know which ones are good and which ones could be fraudulent or or could inject some, I don't know, malicious software into your agent or something, right? So I think on the call it the merchant side security is going to be super important and you're going to have to have like a like a registration layer or something that

certifies oh this X you know X42 endpoint is good this one's bad. So yeah, so uh we just discussed about what's part of the agentic loop, right? So uh one of the inputs is the prompt itself, right? So the prompt itself can be hijacked by, you know, maybe the service that's maybe there's man-in-the-middle attack, the prompt itself could be hijacked. So which is one of the most uh important surface areas we see. Uh the second one is that we see nowadays that all of these uh large language model uh you know agentic AI coding assistance tools can run for hours, weeks, right? Days and weeks. So there is a big attack vector surface as well, right? So how do you prevent that?

The next one is the skills uh poisoning, right? So uh we're in the world where uh we see OpenClaw you know OpenClaw hub uh had like about 1100 skills that were uh injecting poisonous malicious software uh available for uh everyone to download and we also see the supply chain attacks where you see the the dependencies like the npm dependencies or or other open source dependencies that got hijacked and an example in Feb we had client which is a very popular uh developer velment uh tool got hijacked and uh it was available I mean the the software was injected with malicious code that would download uh you know malicious

code in the users computers and we've also seen uh recently the tan stack uh had like one of these similar instance and open software also got impacted right so there's a multiple surface uh vectors but how do we also safeguard maybe at this point uh going a little bit deeper right so the agent loops can be poisoned but how do we go a a little bit deeper. So that's I think where AWS recently launched the payments uh you know what's a service that uh uh it goes beyond the uh non-deterministic loop right so there should be a a place where there is a nondeterministic loop where a LLMs are doing this processing and where there should be a determinism in which there's a policy engine that will look at every single execution and understand

whether it's allowed whether it's authorized uh to execute the transaction and that's uh what the core of fundamentals of that software uh services >> and Marcus any thoughts on the risk that agents face and people deploying agents face? I mean, especially in the blockchain space, we have already years of experience with like hacks, with like fat finger rates, with uh >> like all of the things that could go wrong when machines trade. And in the end, Abidra in the space >> or market makers using are using um code to execute trades uh for years now. And I mean uh I think in the end as Anil said depends on

the policies and uh on uh the guard rails that these agents have where they can execute what what um what they can do and what they uh can't do. Um so I think yeah there are huge risks but uh many of them and many of these mistakes we've already made in in the past years. >> Yeah. And Anne Neil, I have this one to you because at AWS, I'm sure you have a lot of businesses using you. So I'm curious, what impact do agenic payments have on existing business models and how can businesses build for these new actors of commerce?

>> So at AWS, we provide like the foundational services uh like the infrastructure services like uh databases uh you know networking uh like you know those are all the foundational services that we provide storage right so name to name a few of them. Now we made the payments as a native service that's available right. So think of if you think of S3 or lambda right the compute is one of the foundational services. So now payments has become a foundational service that is available for enterprises to consume and uh with the explosion of agentic AI all the enterprises that you know that we work with all the startups the small business uh you know small large businesses are also investing a lot in the AI space

which means that there is a a compute layer that is uh inherently like going through a lot of adoption serverless compute for example networking so because they need security at all layers of the stack Right. So that's kind of what we see happening here today. But uh to answer your question on the payments, right? So all these agentic uh autonomous loops break when they see like a payment wall. And now with X42 in there, like that's kind of where we see that there is a need for uh X42 be to become part of the stack. And we've created SDKs where enterprises can use the SDK, build on top of the SDK and get the guard rails that AWS can provide at uh foundational layers uh as such right.

So we also have software that uh you know think of the operating system for agents which is like agent core. So if you have thousand people in the room and they hit like this service, every one of them will have a micro VM that is spun up which will process the individual request which can be up for 8 hours and uh the you know which means that you know there is a large surface for you know compute to happen agents can do the processing so there's a lot that's happening today and uh we are already you know building products in preview that can support this right. So yeah, >> uh Vincent to you, how do you see agents and agentic payments affecting existing business models?

>> So I've been part of a few roundts with um different uh policy makers. Um of course, one of them was at the Hong Kong earlier and other currency other countries they're really asking themselves, do we really need stable coins and what is it used for? What is the use case? And most the answer is always involving payments. So um you know with that they're they're really looking at how the cor their corporations and businesses are going to use this tool to actually be more effective but at the same time these businesses you know are also subject to different types of

risks. Um I think the way that businesses is going to use these agentic payments is that it's it's now 247. It's not there's your agents don't sleep and um but there's also the the the flip side which is you're going to be losing a lot of manpower to agentic agents. So that's one way that I think businesses will benefit. I would just add like I think I think you know speaking to like business models and and this has been mentioned obviously before um I think Mark Andre's big on this point but I think it's going to change the fundamental business model of the internet from from subscriptions or SAS to to paper use right I think that's what a lot of people think are

coming and and I can see it because why spend $100 a month for your anthropic or $200 a month for your anthropic you know subscription when maybe you don't use that much or or maybe you're using more than that and Anthropic's missing out on profit, you know, so the paper use kind of changes the whole the whole game and then it just enables a lot of economic activity on the internet that that just doesn't work today. Um, so I think that's probably going to be the biggest shift. >> Yeah. And there's so many business cases also around like micro transfers which are now with stable coins and more efficient rails um possible that can enable a lot of different interesting use cases from draining even training

these algorithms and incentivizing users and people to do that uh to yeah distributing microp payments for for different things. Then of course all the trading use cases also change. Uh agents can for your business constantly check your FX exposure and if you want to even rebalance your FX exposure. I think this is something that can make many businesses really really efficient. We've been talking to a bunch of mobile carriers for example um that are constantly exposed to to FX in in a way that every time when someone roams and and basically pays a mobile carrier then maybe also in different currency um they

they need to adjust their their position and um being able to to hand this to agents that can do this real time because there's uh yeah no effort in doing this every second instead of I don't know at the end of the business day um already uh changes uh changes the game and then I think also with stable coins and with stable coins and different currencies you now have these markets these FX markets open 24/7 and you can allow your your agents to to engage in these markets 24/7. So I I think there's so many many things that that are changing in in changing the business model. um which is quite

exciting. >> Yeah. Something I've heard mentioned on the stage multiple times is uh X402. Uh I'm curious for those who may not know what it is in the audience, can you define what X42 is and how uh businesses benefit from it? >> Um to Philip. >> Yeah. Uh so my understanding is that you know the 402 uh request was left open I guess at the start of the internet right and it was always a payments like protocol uh but no one I guess did anything with it for all these years and finally Coinbase kind of started said hey why don't we use this for for uh payments um and it kind of spun out of

there and and has taken a life of its own. It's it's it's awesome. I mean I think in my mind it's like why can't every e-commerce transaction be an X42 endpoint or something right? You know it it really kind of widens the aperture of like what agents can do and what what economic activity on the internet means. >> Yeah there's 402 there's MPP I'm cur is like are they just different standards and >> yeah I mean I I think they're basically the same one's on tempo one's on on bass or whatever on EVM. >> Uh anal to you. >> Yeah. So uh when we talk about X42 right so there is a server like we talk about agent e-commerce there is a buy side and a sell side always right as part of commerce. So the seller side which is trying to monetize on the data assets

that they have or the API services that they provide they expose an endpoint that is uh 402 capable you know that can transact and accept payments microp payments. So uh when the client uh receives that response back that the server is expecting a payment, it signs an authorization and says hey I'm authorizing that a transaction can be done and I'm attaching that as a part of the header right the authorization itself. So the server will then say hey okay now I have a response that the client is sending back and saying that uh this is the authorization let me ask a facilitator to settle the transaction on chain and then uh confirm that it's settled so that I can release the

response back to the client who's requesting for the service right so that's how I see that the full loop happens so there are multiple parties but one thing to note is that the uh facilitators are the ones who are making the transactions and when we go into agentic loops, right? So, we should always note that the agents should not actually hold the payments, right? So, it should be always part of the the guardrails that uh enable the agents to transact. >> Uh Marcus, any thoughts on X42 or MPP? >> I mean, in the end, I think it's good that there's now more and more of these standards that basically people can agree on because then these economies can can grow. Um and in general in that space we we see this now finally taking

off because now many things happen at the same time. You have with X42 and other standards you have basically a standard forentic payments. You have more efficient payment rails with like low gas fees and fast chains. you have uh even yeah now more advanced uh regulatory clarity in in many regions of the world that you can actually engage agents for you to engage with stable coins uh here in the US a genius in in Europe is Mika and and there's many equivalents around the world and I think all of this coming together is is is good that this economy can now grow >> and maybe just to add there right so uh we see that as X42 is evolving and

adoption speaking up all All of these transactions are like you know going to be uh million scale transactions subsent transactions. So uh when the scale increases maybe there could be a human in the loop that could be uh you know hey if there's a payment that is requested for more than a certain amount or it exceeds certain transaction volume or dollar volume maybe the human in the can be called invoked to say hey are we authorizing this to act furthermore. So yeah, >> uh Vincent, I'm sure you've seen a lot of adoption of agents using stable coins across different regions. Um when it comes to agents, are there particular industries uh and geographies where you see the most adoption of agents and agendic payments? If so, which ones? And can you site examples?

>> Well, I think right now the industry really um that will benefit most from the agentic agents, it's the e-commerce space. Um there's a lot of there's a lot of uh e-commerce companies now that are really posing for position and if that opens up to global consumers. Um second of course there's also the uh yieldbearing side of it. People want to generate yields. Uh our focus is really on the emerging markets. We think that the answer for the unbanked, which I'm sure a lot of you guys have been, you know, hearing for many, many years, but with the combination of blockchain with stable coin and now with the AI gin, I think the the answer is really here now.

Um, providing somebody making, you know, equivalent to $100 a month, being able to put aside $5 and start creating a financial footprint via different ways. uh albeit vaults or or just buying crypto and and waiting for appreciation on that. So there's there's different parts but I think um yeah I mean we can go on for hours on where the use cases are but those are the ones that I think um I'm more passionate about.

>> Yeah Marcus uh what use cases have you seen? So on Cello we've seen in the beginning a lot of uh agents in either the trading space or agents that in some way engage in uh digital assets let's say uh and digital assets cannot is not only uh does not only mean um crypto assets it could also mean digital asset like um for example server capacity things like that but we're now actually also seeing more and more projects coming up in the e-commerce space for example We see, for example, across Africa interesting uh interesting projects that employ uh chat bots inside different messenger apps that for

example go and buy airtime or data for you um and choose the cheapest provider, choose the cheapest stable coin uh and and and then uh handle basically your data management for example. So there's a bunch of interesting also let's say more retail focused applications that are just now coming up. >> Uh Philip, what use cases have you seen? >> Yeah. Uh the I guess the the the famous one was the SpaceX S1 where they did like the >> what was that?

>> So I guess someone paid their agent like a dollar and a few cents to read the SpaceX S1 and then do like a very detailed like investment committee memo. um breaking it down, right? Uh and it only cost him like a little over a buck. Um but in that, you know, process, the agent had to go out and do a bunch of like microtransactions. So, I think that's obviously super interesting. But when what you were talking about, Vincent, is like in in emerging markets like enabling that person that makes $100, you know, to have $5 of onchain credit or something like that that that changes their life or it can really help them. So, I think that's a really interesting use case. And then I also think B2B is going to be real big and agentic. Like if I'm a construction, a

home builder and I'm, you know, have to get cement from some other country. I mean, I think in the future your agent will just talk to the cement manufacturers's agent and they'll negotiate a deal and they'll pay in stable coins >> and then FX risk is hatched and the cheapest rail is chosen and so on. Mhm. >> And maybe a different uh different view of this is I I'll talk it from the talk about this from the industry's perspective because uh we see this uh adoption in various industries, right? So for example ads, right? So adex where you can uh you know ads can be published using agents or can be consumed. Now if you look at uh the normal consumption, right? So when we walk through the New York lanes, we see billboards, but now

agents are not looking at those billboards, right? So they're looking at uh APIs that are more reliable, right? So we see uh across different industries like maybe healthcare like can can somebody make a healthcare appointment or can a travel and hospitality industry right the flight reservations can they be made by agents can they be uh negotiated by agents uh you know so there is there's different industries but one thing that uh was very interesting was of uh manufacturing where robotics are now picking up and uh uh I s I saw came across a use case where uh they're trying to say hey there are a number of robots uh each attached with a payment tag and see which one actually performs well and be able to figure out if they can give

more tasks to that robot and encourage or let the other robots fight. Right. So, so I've seen that use case >> strings the loop in the real world. >> Exactly. So, uh I mean those that's one of the steps in the evolution but yeah >> uh Marcus I want to ask about the opportunity that exists with agents and agenic payments. Uh how big can this get? Uh, and can you site numbers or any statistics or just lay out your vision for it? >> From my perspective, and this bit of a bias because I talked about FX the whole day, so I I'll stay in that area. I think that can grow really really big because if you think about the addressable market, it's in the end

infinite. Um the the markets for the market for currencies uh is the largest market in the world and is 7.5 trillion dollars are traded in FX every day estimated by the bank for international settlements and uh this market is still mostly on old rails uh which means it's not 247 it's not accessible everywhere um there's a bunch of like mistrades happen and this in in the largest market in the world and just like being now able to bring a large part of this market on chain, make it 24/7, make it accessible everywhere in every country and then actually have agents that can hedge your FX exposure for you as a

corporation that can make all your global purchase for you as a as a private person. Um, I think that has a huge potential and and and actually can can change a lot in in in financial markets. >> Uh, and to you Philip down the line. I agree. Um I think it definitely has huge potential. The one thing the caveat that I would just mention is I think currently like e-commerce is huge, right? But a majority of retail commerce h happens actually in store in inerson retail, right? So I think it's like 70 75% of of of transactions. Uh so agentic commerce probably won't take that away.

So uh but maybe it'll it'll shrink what what that means or maybe it takes it all. I think in person in store retail will still be a thing. >> Up to you Vincent. >> We're already seeing I think 50 over 53% of the US business is already using uh a AI agents and I think McKinley also projected $1.5 to3 trillion dollars of e-commerce orchestrated by agents by the year 2030. So those numbers are real numbers and I think the the opportunities are just endless. Um, of course, this, you know, a lot of governments don't want to hear that because every government is looking at the risk currency uh flight >> and it's just hard to keep this this

technology this industry down. Um I think this is going to be an huge like I mean this is the year of the agentic agent right? So yeah uh from uh a different vantage point we see that all many customers right or I should say most of the customers uh of of AWS are definitely investing on agentic AI architectures uh to streamline their processes and to bring in efficiencies right so that's uh definitely uh something to be taken for now on the payment side you know we we spoke about uh X42 so X42 has like $50 million uh you know worth of transactions or 170 or 175 million transactions uh

that have taken place all sub uh second around 200 milliseconds right so again today most of it is happening on USDC uh on base 98% of the transactions have happened on base because of the latency benefits out there but as soon as uh fiat trails are available uh you know the support becomes more native uh there will be more of uh explosion for this next steps here in payments. >> So, TLDDR, it's going to be big. It's going to be real big. Um now, if people want to learn more, engage with this topic, go deeper on this topic of agents and agenic payments, where should people go and what do you use personally to

keep up? Uh I'll start with you Neil and we'll go down the line. >> So, we are always uh following Xplatform for the latest updates of everything that's happening around. So, that's one thing for sure. uh AWS has uh like a a news blog to keep up to speed on what we make uh available uh for our customers and I'd say that is one place to check it out. Um and for the builders out there, you know, there is uh Twitch uh streaming that's available. >> There's Twitch streamers. >> Twitch streamers, >> agent Twitch streamers, that's new. >> Twitch streamers, you know, for sure, right? So, I mean, I have like another show as well. It's called the show show and tell and you can find a lot of videos on agenti that uh we make it available but one thing I'd say is there

is something called agent core which is a a foundational service that makes the build of agents possible to allow you to uh bring your agents in a matter of minutes with all the security guardrails uh that are available the foundational security guardrails and we see security as day zero at AWS right so you may want to just check those resources out and yeah that's about to you Vincent. >> Uh, our agent of choice is Claude. So, we've been training a lot of our staff how to use use it properly. I mean, including myself. So, I guess I got to go back to school.

>> Yeah. I mean, I would say um, you know, I think obviously, you know, cryp uh, crypto Twitter or X is is where I go. Um, but I would just say, you know, download Hermes or OpenClaw and start messing around. Like that's really how I learned. Um, and I've somehow got, you know, I used open router and got the, uh, agent connected to my telegram and now I can talk to it on Telegram and it's all free. I don't have to pay Claude. It doesn't work as good as Claude, but um, but yeah, just start messing around with it. It's pretty fun. >> So, in the end, if you want to learn about agents, ask an agent to teach you and you know, >> that's the amazing thing with AI. If you just ask your directions, it'll tell you it'll tell you how to use it and how to

use it to its fullest potential. Um but yeah with that that was all my questions. Thank you so much.