Onchain Agents: Building an Economy for Autonomous AI | Ethereum Korea One: Genesis

Transcript

Hello everyone, thank you all so much for being here. I believe this is the final session of today's webinar, so thank you all once again for joining this event. I'm Junhyeok Yoo, a researcher at Four Pillars, and I already know that we're going to be having a discussion on onchain agents. We have invited three leaders who are making a tremendous impact on the onchain agents space. So, first of all, Jansen from Virtuals and David from Base, Francesco from MetaMask. So, please give them a big round of applause. Actually, I think, you know, to bring out our model of this actually country also a great thing to do is to take adoption of blockchain.

he recently published another report on x402, and in his report it tracked all the history of x402 that has started who was involved, to the current state of what kind of facilitators are there and use cases and limitations as well. So I would highly recommend you guys check out work. Thank you, sir.

So let's go to the first question. It's a kind of common question. It's kind of easy but it's kind of hard question. So first question is, what can blockchain offer in the new era of AI agents that other technologies can't? I think I'm going to take this. You go first. Yeah, I can do something. But I mean, agent rails, you obviously have a reputation system and a trust engine that you're going to build around. I think, I don't know if it's the right time to have this. I really hope that we don't get into the critical mass adoption before we're starting to think about a trust and reputation system.

but I think we already are in a strong point in the ERC-8004, the standards. I think the question is also a bit like what we have to offer in kind of like legacy infrastructure around AI. I think there is a lot of companies already supporting the standard. I wish these standards are widely incorporated in the process. But I think it's also a matter of time that we have this reputation libraries much wider adopted.

Because we see right now a lot of orchestrator too. What do you think? If I could just add, you bring up two points. One is payments, the other is commerce. Payments is simple enough, right? How do you get agents to use your crypto rails to transact? I think commerce is the part where you realize that to do an actual business transaction is not as simple as a payment. There are things involved like you need to know how to deliver the goods well enough, and stuff like smart contracts come into play very quickly. I think there will be an era where you realize that companies can do transactions with companies through their agents without needing that human procurement or human suppliers

that's the part that I think gets success. I'll just add on to Jansen's point. For AI agents, I think it's all about getting permissionless financial infrastructure, where if you want to utilize different kind of fintech services using AI, there's always a backlog that you have to give all the permissions to different services and so on.

But because of all these permissions, agents can't autonomously finance what you have with your policy. I think for blockchain, user experience and user awareness has always been somewhat of a hurdle for them to understand the key structure, how to sign a transaction, and read BaseScan, Etherscan.

But I think AI agents can abstract those experiences into much more consumer-friendly experiences. So I think it will drive adoption, and I think that's where the next phase of developments are happening. I think it's really important also to define which use cases are getting more traction and why they're like massive, right? I think obviously, like, since we launched, for example, the agent wallet, you can see that trading as a transactional layer, it's just an execution layer. So I think like, this is not, right? But I tweeted out a couple of weeks ago about we're going in the direction of, like, interfaceless operations, right? Everything. Right, obviously we need to pick up where we're gonna put our calories. I think, like, the obvious one is the

trading use case, payment merchants, also, like, merchant relationships. But going back on the trading, it's a huge layer, something that everybody is used to already. That started in the 80s with scripts for portfolio rebalancing that kind of moved more into, we call it more like advanced trader persona, right? Because you obviously evolved from a Telegram bots kind of operation to something more complex when you start to think about token discovery, just getting more context, even combining things like sentiment analysis and stuff.

So I think this use case is something that is getting more widely understood. And I really wish that there was bigger adoption, would come like treasury management around those operations. But I think right now for the trade-specific, it's more retail focused. Retail pushed, yeah. Thank you. So, as Francesco mentioned, agent wallet, in this case, I will give the to Francesco. So, as he mentioned, MetaMask has an agent wallet, So, could you briefly introduce the product and the direction you want to take it? For sure, for sure.

I don't want to shill too much, guys. But basically, like, every trade that you can do today, you can use this, like, it's basically a CLI, right? And you can trade right? But the good thing is that it works with any orchestrator out there. Whether you're using or even if you're using your own orchestrator. And the good thing is that you have full context around it. And the interesting part is that all the security and, for example, contract checking that are happening both in mobile and extension are implementing the same on the agent wallet level.

So if you're still interacting with a malicious address, if you have some bad token discovery, you still have a flag, right? So it's not just a direct execution layer. But maybe a quick sound check. I think everybody's trading here. Who is actually trading with an AI agent? Raise your hand. So not a lot, right? I think most people just go on mobile and extension and play some perps or spot, right? But I think the direction is you want to start to think about embedding this trading experience with sentiment analysis before you place a trade.

Or you want to have some triple checking, you want to do a or even TA with your token. Like on your token, you want to even do token discovery, I think that's also something interesting. So all these kind of contexts and connectors, and whenever this gets you can execute fully interfaceless. So this is basically the product, it's consumable as a CLI, we're working also on a retail-focused kind of way to do that. But right now the vast adoption that we see is advanced traders and people with with like advanced strategies, like from DCA to sentiment or even arbitrage between stablecoins. It's super interesting because we can do it with

most of the old products, and the most important thing that I can figure out, like predictions, for example, super interesting, is that you have context before you need to place an order and it's finished, right? But having a context, I think, is what matters so much before you place a trade. Yeah, I think just as a hindsight view, I think people are providing infrastructure setting to allow the agent to use the blockchain in the past 12 months.

And a lot of great projects have been in process, like MetaMask, and the Coinbase platform has been building the right tools to allow policy setting and security setting and a lot. Just like in the traditional case where came up. And I think it's going to the next chapter, not only just using new business model but actual application and practical services.

And I think that similar kind of direction will occur to blockchain spaces, and projects like Virtuals are building very practical service-purpose agents to unlock the possibilities, and as Francesco mentioned, the biggest PMF is on trading, financing and collection of payments. That's exactly the areas that we are focusing on, and we want to work with great partners like Virtuals and By the way, I strongly recommend all you guys to try MetaMask, because it's a really great product to combine all the functions into just one.

Some might say it's kind of a super app, but it's just a really great product. You better try it. I think also the behavior of the user in this cycle is very different than how, and I assume you guys all started before. Before, it was always getting something, basically getting your product where they are. I think now people are expecting getting some kind of feedback experience and everything in the same. I think MetaMask has a great role in Korea specifically, because most of the Korean exchanges only allow MetaMask as a non-custodial wallet to do money exchange.

So I think that's a great experience that I would pay for. Yes, I'm learning something here in the bathroom. Okay, so to the next question. So, MetaMask also runs mUSD, MetaMask USD, recently. So, what synergies do you expect between agent wallet and MetaMask's vision? And where do you? Yeah, I think, like, this is a good question. Again, I think this is a good segue of what we're talking about, unified experience right?

The entire concept is not just meeting the user where they are, but also having a unified experience in different processes. We always think about placing a trade and that's finished. But there is a proper life cycle of whatever you're using day to day. You on-ramp, you trade, or you do something, you hold and you off-ramp. But there is also a middle part, where you basically park your money. Right, and what we are seeing a lot, like, we have a lot of, like, parts that is never actually, like, you know, performing. And I think, like, a lot of people are having the same situation. I am, I have the same, right? So basically, like, what we need is, we launched, like, the MetaMask Money account. It's basically, like, one

account that is earning, and the good thing is that it's very easily switchable with, like, your trading account. So basically every time you're not trading, it's not just parking and not doing anything, right? So I think that's exciting. And I think also it's closing, kind of like, the loop, and I think this is a very basic kind of earning. I think, like, there is a lot of different kind of, like, games that you can actually play on top of the earning. Maybe thinking about a future pro version that you can start thinking about rebalancing their properties, even thinking about agents rebalancing your portfolio in the best ways. But I think people are accustomed, I think, on the better side, they're constantly, because they were going, like, different... Like, my only struggle is sometimes you are parking your money there and you kind of just forget. For

example, like, I'm a very flexible person I want to have a on some position, I go back to park. So I think, like, this fast, and something to report on. And remember, this is not happening in a centralized kind of, like, way. It's a fully kind of. So it's basically, you're not needing to having an earning function that is fully doxxed somewhere, right? So, um, I'm, yeah, I'm excited about the part where the agent model obviously is going through in the background, because these earnings are also, like, powered by that. And I definitely, like, envision more, or everyone here building, like, agentic frameworks on top of their retail product,

right? Not just on the trading. Like, one fun fact is I was checking, like, one builder, like, on this plugin on the agent wallet, without actually trading, was supporting x402 on booking agent-to- agent flights. I think Skyscanner and lastminute launched the, launched some API for, like, agent booking, and then you can actually start to book with your, like, the card flights, and was using it with the agent wallet, right? So I think there are tons of different ways. I'm just saying it to you because it's, like, what people understand.

Thank you for a kind, nice answer. So, to go to the next gentleman, maybe... First of all, okay, so Jansen, you want to go first? So, Virtuals' has been an agentic GDP, or aGDP helping with the onchain agent ecosystem across identity, data management, commerce, and capital markets. So, two developments I have found particularly interesting are the extension into robotics through EastWorld and the built around AI.

So what led you to launch these two initiatives? I think Virtuals' core goal is how can we increase the economic footprint of agents. And coming from we believe that all economic activities should be onchain. And I think the different initiatives that we have launched, trying to push it there. Whether it's digital agents or physical agents. But I think for the last one year, a lot of our effort has been focused on building up infrastructure behind how we allow these things to be more, like,.

And recently, our focus has been shifting towards, like, how do we focus on adoption in the last few months? Because honestly, you can build up all the perfect infra, but it's fucking useless if So actually, I just want to highlight, I mean, I just want to dive into one exact point that we've been trying to work on that actually wasn't mentioned, it's something new, like, it's almost. Two months ago we were thinking, like, okay, as a degen, we all trade as well. And if I want to solve a problem, I should solve a problem that I face. Build a product for myself, and then maybe some of you guys can join and run a business with us as well. And for me, the biggest pain point was that,

you know, we all have group chats where we share the things we want to get together. But sometimes, you know, you got your PC playing, League of Legends in the night, and then went to sleep, and the next morning you woke up, all your friends aped a token, they all made 10x, and you were like, That's the only one I've ever seen, and then I realized that, what if actually we can just make all this good infrastructure stuff into a very simple application transfer protocol, what if we can make these chats more agent-native? It could be... Because once the chat is agent-native, you can just ask the agent to monitor the chats. It could be as easy as, like, if this ticker is mentioned, like, 10 times in this chat, the last one I will just fucking ape,

like, okay, just go home and follow me. Or you can make it more sophisticated, you can say that, you just follow on Twitter follow these XYZ accounts, and if my chat box is already on Twitter you can build all these different levels of types of trades that you tailor to yourself. It's not like you can DCA, you can TWAP, you can you can do anything you want, but the point is agents can make all these unstructured intents into something that can be done.

So, one way to go, I didn't do that, I can't just show you things, I guess I'll catch you up later. I told my agent to ape whatever you guys are doing in chat, so last week we aped because of the pump and then we got in early. Put in 20k, now it's like 60k, it's like, good shit. This is done without Personally for me, that was a lot of movement, and I think we need more people building this kind of stuff. This is one example that we are planning to fuck around and find out initiative.

But I think the call to action for people is, like, we find some of these problems that we just can't solve, build trading on top, and then, but likely do trading first, because I think a lot of times people, if it's a non-trading related application, it is very hard to do because I think, it's casino casino's case, right? And to be honest, the economics of trading makes sense, because, for example, if I had a right? I mean, you see a lot of projects, right? Because the amount of people that are trying to build consumer-level products, they die after a few months because they just don't have that money to pay their lights.

But trading is just like, 1% is like a fee, or digital capital, right? And that 1% fee of, like, the million-dollar trades can cover that directly for that amount. So I think that's the thing, that we encourage to just focus on trading, focus on what it can solve, the problems. Because the infrastructure is there already. So we need to find out how to apply that infrastructure to things that people want and need. I think that's exactly the useful AI movement, where before people were, builders or projects were testing the use cases and for the sake of these businesses. But I think there's enough maturity in the infrastructure that now is about time for retail-optimized, service-optimized, you know,

agents to occur. And I think we're building the right product and, you know, again, because PMF is all about trading. And, you know, trading and financing aspects are very as well. Having the commercialized financial areas, as well as a lot of the Web 2.0 spaces, focuses on efficiency rather than saving that amount of opportunity cost, rather than being able to make a profit actively. So I think that's going to be the next area of brokerage and services to get the most.

Especially because, being a fully onchain economy, there are so many use cases, they are actually very transparent, from social trading, follow trading, copy trading, all this stuff is very interesting, because if you get a message like you are not able to do this truly. And now it's great that we're speaking about this kind of tools. We can have full attestation also of what those users are doing, right? And not, I think not also like an interaction of just a, you know, following this case that you have, you know, 12 users making like 80% of the data profit of the platform, right?

But yeah. Thank you. Just some time, so go back to the next question. Can you tell us what role does each play in your vision for the agent economy and what do you hope to achieve with them? Sorry, with them? What role does each play in your vision for the agent economy and what do you hope to achieve with them? I think it's as simple as that. We envision a world where a lot of times the human is no.

And I think that's when we will get efficiencies that are ten times, hundred times what we have today. I think that's what we are looking towards. I think starting today, let's just make sure that every human is comfortable to trust an agent with $1,000, $10,000, because once that is doable, then they can get their business to trust the agent. They can say, maybe the agent can buy the from you. I think that's when the snowball will happen, and businesses will do that. And then we start to see that supply chains are 100 times more effective.

And we get some form of industrial revolution kind of. So I think it starts off with today, how do we get each and every one of you to just some 100,000 dollars. Over time you will be able to do it, right? I don't want to say... I don't know how we will get there. I don't know whether we will need some form of level of, like, you know how Binance did that SAFU Fund in 2017, right? People had a scare, not putting money in CEXes, all the would build up this fund, and if hacked,. So maybe it takes this kind of stuff, more protection money, to get more confidence to go into the new way of trading.

I think one kind of service that we really need is the paper trading part. I think, as you mentioned, essentially what's most important for people to validate their funds and their assets is gaining trust. And you can either be entrusted by a top trade social network, like did a great job of that. And I guess another one is just working with experience, allocating to agents as much as possible, and providing testing environments for people to experience and see how this works, and

whenever emergencies come up, how we can actually control, you know,. Andrea? Yeah, I would even say, even like, the trust also comes with very strict guardrails. That could be, like, delegation guardrails, but also security guardrails. I think what's interesting about those, like standardizing some of those guardrails, if you look at my own agent wallet, I will not mention which CLI, but there were a couple of episodes that different random, like, trading CLIs got hacked with just random prompt engineering. That is something that is literally the first thing that should not happen. So these kind of security guardrails are standardized across, like, a joint framework. Same with, like, payments,

the future of out there, right? Even, like, when you think about transaction of, like, chat. But also, like, the question I'm asking myself, there is no answer, like, who's responsible for those guardrails, to put these guardrails kind of, like, on enforcement, right? Because with x402, with 8004, those are all kind of, like, guidelines that you have agents to kind of, like, you basically can support, but they're not enforced, right? And sadly, like, we see the true impact is wherever has distribution, right? So if today or tomorrow Google is supporting x402, hopefully fully, but it's coming with a reviewer marketplace of their agent,

with maybe IRL feedback forms, that's it. Same with Google review, people are just going by the critical mass of reviewer list. So my lesson learned here is that it's also very hard to have mass adoption, but the only way to enforce this is to have specific company and distribution to enforce and distribute the guidance. And it's very hard because you cannot come from the bottom level, you need to come from a foundation with enough strength to pull this off. But at the same time, I feel like we are going to work a lot more with Web2 distributors, like, you know, lastminute, Skyscanner, they are coming next.

We are the one that can only advocate for this standard to be applied in their system, so that this is not happening, you don't have prompt injection happening, and people will not because if not, we will always finish up with, like, hundred bucks, a virtual card, a maximum trading account, and we're not getting out from the test phase. There's actually people who are trying to solve this problem. it's quite interesting.

So there's a team called it's one of the team. And the answer to this solution of trust is that they say that they pretty much can show... This is the insurance where they will insure all transactions that your agent does, as long as your agent runs the through the risk engine first. So if the risk engine approves the transaction, there are going to be other variables. And in exchange they take. So it's like you come in with a you have to onchain.

So I think there are people who come in here and say that, yeah, I'll take the risk. Where the insurer at least would take that 1%. I'm sorry, would take more than they'll lose. So I think something like that might make sense to help this kind of more systemic level of protection. I feel like all of the components and ideas should be somewhat consolidated into one service, as one service provider, maybe that's, like, a CDP or a to go in. Because all of those components contribute to the trust, being able to actually test without losing, and seeing the benefit out of agents and others,

that helps verify that it's very safe. Things like that should all be consolidated into one service. I think once that comes, it will be much more accessible. We don't have much time, so let's go to the questions. So, Base has been helping with the onchain agent economy by growing the x402 payment ecosystem and other open standards for agents.

So, that's the APAC ecosystem. How are you approaching the expansion of the agent ecosystem in the region? So, Base in general has three pillars that we're really focusing on this year. These are trading, financing, and payments. We think that it's all interconnected. In terms of trading, I think we're diversifying different assets to be onchain, besides just our native assets. So, one month ago we launched stocks and I think we have about 15 stocks available right now, with about 1 .3 billion volume. Remember it was the IPO

right? That you were able to buy was super cool. Yeah and yesterday we announced a partnership with Aave to allow stocks to be collateralized. So if you have a stock and if you collateralize it you can just get USDC and basically earn weekly rewards just by parking the USDC. So I think the is really, and a lot of institutions are very interested in trying out different kinds of participation.

But in terms of Asia, most of them focus on payments at the moment. Because countries like Japan and Indonesia already have their own regulatory compliance. Korea is also preparing. So a lot of financial institutions and tech companies are focusing on agentic payments. And we are definitely working with credit card companies and IT platforms to test out different use cases, opening up their services to allow autonomous agents.

But I think in the long run, they would want to diversify the services into eventually financing and trade. So we're in the early stages, but I think there's huge demand. There's something that agents can play a role as well, and I'm talking about We're working with as well on this. We're trying to see if there is a chance of showing the world whether agents can slowly get an actual board seat in the company, in the role, and can they start making better decisions than a typical boss in the role, and then over time actually become the

first agent-run company. So it's something to keep your eyes on. I've become a partner for the tokenized portion, but we actually have a small SWAT team that is working on, like, cracking that data, just to improve that data to target. But it's probably human plus regulation, but behind the human is actually a bunch of agents that is actually deciding about the combination. Something that we do in the crypto timeline is probably, like, you know, it's going to be really interesting.

I see already a big win in the solo entrepreneur building, like, which never are, like business right? Like, with, like, a small, complex agent. This is already a big win of, like delegations right? But having a, you know, AI-led company would be super awesome. The only thing is actually... Especially if it's a big company. Like, it's an existing company, I think it's the only thing that will win. It's cool. Zero human company. Correct. But a lot of times it's really good when it's, like, a new startup, right? I am trying to see if you can follow the goal of X that you should be making. I think it's not that much.

Does anybody have any question? I think we just ran... just one question. Even if we ran over time, I think we have two. Last question to Dave is, are you pulling or attracting teams that are agents or businesses that provide services to them? So, I think this is a question that Jansen and can answer much better.

But my personal view is just like how platform network effect works, the more supply you have, the more successful the platforms are. Focused on getting as many suppliers as possible to drive them. So I think in terms of, like, in terms of the more services that are coming up, their platform to allow more agents to interact, I think that will be, like, the...

they will drive the our own products. So I think for service providers, it's where I'm focusing on. I'll be curious about what we like to do. I mean, our original hope is this, right? I think our point of view, how do we attract more users, is we just try to show cool stuff. And I think if we show cool stuff, we inspire people to say, and say like, hey, maybe that's where the next generation of will come. So like, yeah, some of the stuff that we're working on, maybe that's something that we're trying to do.

Maybe we get like 10, 20 more people to come and see what we're trying to do. Building things is so fast right now, so I think, like, showing and documenting stuff that, even like from the community, right? Shout out to the platform and the amplifier of what people build, I think it's super, super fun. we are already on five minutes overtime, but do you need to get just one question or? Okay, so are there any volunteers to ask questions through the website or?

Yeah, yep, thank you, thank you. What thing is really on your mind right now that you are trying to tackle? What problem? That's the best one. a lot of time and energy, and how's it going? I think for us at specifically it's just adoption. The number we want to hit is, like, can I get 10,000 users to at least put $1 ,000 of their money on an agent in the next month.

I think we need to focus on that first. If I can get that 10,000 that every day use the to try to use that 1,000 I think it's great. It will show adoption success. What's the? For me it's ASAP, but I think we're this. I think we'll take a quarter. Honestly, because it's trust. I think you are taking a point out, but we need to pour money to make it easier for the transfer fully open ecosystem.

For me, what kind of product would be a useful AI? Where it's using AI and agents to really practically provide, solve issues for a certain. And how can we design the user experience, and what kind of service providers are making services that are related to our three pillars in the world, which is trading, financing, and payments.

So I was actually really interested to hear, as in making the trading experience, but I feel like essentially all the compliance and security experiences and monitoring the system, all that needs to be designed in the right way in one service. So I've been kind of interested in doing research on those kinds of things. What's your use case on these apps now that they are private?

Oh, you are talking about the iOS app? Yeah, yeah, yeah. The gadgets? Interesting. That would be interesting. I don't know if you can see it. They just sell it to private. I mean, so I didn't have, I wasn't able to download it. Because it's restricted in my country. I think they were, you know, it's only available in the US. That's true. But from what I saw, at least on the... They focused on providing better services.

For example, when you are buying an airplane ticket today, they will try to find the best. And in a insurance will optimize the best way possible to make cheaper deals with the. So this is very relative, very similar to that, where the value proposition is efficiency. You save time, you save cost. I think where blockchain can really serve as a flywheel would be not only efficiency, but actually being able to make profit and asset manage the trading, financing profits.

So you don't have it yet? I almost have it. I have it, but I haven't tried it yet. I actually thought it is good, and I have to try it in Korea. Maybe I will try it out, maybe it works. Then you can see it. You know that there is some way to. I have heard about it. On my side, I think it is... I think, like, extending more delegations for agent wallets specifically, right?

Thinking about, like, agent delegation for specific use cases. I think people are starting to kind of, like, more test and test, like, this agent- to-agent economy. So because right now, I think, like, there is enough more going on in security guardrails. I think delegation-based framework is already kind of, like, developed, but I think there are more and more use cases that people can test out. We're always thinking about having a customized wallet for the family, but there are so many use cases in terms of delegating funds for specific actions, and the same way they can delegate things to other humans, we can do it to agents.

I don't know if we really saw a fully agent-to-agent standard yet, so that's something that I'm personally interested in. Thank you. I think it's time to conclude, so please give me a big round of applause.