AI on Ethereum: ERC-8004, x402, OpenClaw and the Botconomy

Banklist Nation, we are here with Austin Griffith and Wday Kra. These are two people working on building a bunch of AIs and getting them on chain. Austin, David Wday, welcome to Banklist.

Thank you for having us.

Thanks, David.

So, we're going to talk about AI on chain in this episode. And I kind of want to ask you guys, how close do you think we are to AIs being the dominant transactor on blockchains? Because blockchains have something like maybe in the aggregate something like 50 million monthly active users across all blockchains across the whole world. How long until that number flips to being AI Agents?

I have no idea how to guess that.

Yeah, I even look at the numbers. I'd say that today there is already a ton of bots that are like not like smart AI but maybe dump AI that operate like even DeFi or Risk DeFi, etc. But I'd say like these new use cases of these new types of bots doing different stuff, maybe in the next one to two years definitely we'll see a lot of inflow.

Does it count if I tell my bot to do something and it does the thing? Does that count as a bot? Because when we first started talking about this, it was like AI is the new UI. Blockchain UI is UX is pretty rough. Like I can just tell my bot to go do something. It opens up the browser, opens up its MetaMask and dozen clicks around and it gets frustrated with MetaMask and I don't have to deal with it. Like does that count as bots interacting on chain?

If that counts, I would say six months.

Yeah, I would say months.

Okay. Wow. Everyone's going to be yelling there.

Yeah, I think you're going to be yelling at your wallet instead of clicking around. I think that that counts if it's an Agent working on your behalf. The same way if you had say an employee working on your behalf to do things with MetaMask or do things on chain, it would count in that way.

I think a thesis that we've talked about on bank lists several times in the past whenever AI and crypto has come up is the true cryptonatives are actually AI Agents. And what we mean by that is because they are software and because they can sort of think at the speed of light and because they are native programs, DeFi and Ethereum and crypto is going to be even more I guess domestic to them than it is to human beings, right? So, in the same way that, you know, I might feel a little like a foreigner doing things in in the digital landscape, whereas if an AI was like trying to manifest in a bank branch, that would be very strange and awkward, right?

They are finely tuned for crypto rails and thus they will become the dominant player. Do you guys believe that? And what about AI Agents make them so conducive to using crypto and DeFi and all of the tools that we're about to talk about?

Yeah, I totally believe it. I mean, like I'm kind of investing a lot of effort on it and with our teams. I add one nuance to what you said Ryan. I don't think it's just domestic to AI is actually necessary to AI because like so in like humanto human interaction we have different trust mode right I trust you because of relationship like we are friends we've accumulated some shared like background empathy then I trust the courts because those are human institutions that have like in the human society have accumulated trust like the AI is like a rational program is like distrust mode they cannot use them but when they start interacting with some value at stake right not some simple Reddit thread where they just talking ZOP etc. then they need to trust and how do they do it?

Ethereum like decentralized trust is basically the only way they the only thing they have as I understand it.

There's a bit of an arms race going on with a few blockchain ecosystems to get AIS on chain to be the the host of AI activity and I kind of think it's basically down to Base Salana and the Ethereum layer 1. And you two I think are kind of like leading the charge of you know getting AI on chain on the Ethereum layer one. You guys are the Ethereum layer one side. So you know team L1. Is there this sense of urgency that you guys have about AIS on chain? Does it feel like a race to you guys?

I think that before we have nation states settling on chain, before we have giant institutions settling on chain, I think a lot of these Agents settling on chain is like setting it up. There's a there's even a tension between the trenchers and deans and the builders and it's been somewhat civil warish and I think other alt L1s have figured out how to embrace it a little bit better and turn it into more of a flywheel and I think that's what we need to do on the Ethereum ecosystem side. So, I think yes, AI might be the new UI and might be all of our new users, but right now it's a lot of economic activity and we want to embrace that.

It could be an arm race, but it could be just like the same competition that we've been seeing over the past few years, right? In the sense that like of course like everyone wants their block space to be useful. I'd say that like from the Ethereum perspective like first of all as a steward of the Ethereum ecosystem we care both about the L1 and the L2s that settle on the L1 right and I feel that maybe we can go into more details later but I feel that there is kind of like some advantage on the L2s of Ethereum and then some advantage on the L1 and Ethereum as a whole could be like the winning EOS ecosystem because like there is this differentiation and like basically AI can choose like different platforms to settle on depending on what they want to do.

And then the other thing is that I believe that like one AI may use like different platforms, right? It's not anymore like DAPs that okay you deploy your your smart contract on a chain and you're resident of that chain. These things will move. So I think this informs our strategy like from the Ethereum perspective on how do we make Ethereum so first of all Ethereum is already the most decentralized chain. So in terms of credibility I feel that will matter for like high stakes use cases. So I think we have a check there and like kind of we are in a great position.

Second we are leading on some standards that are getting adopted on like many other chains. So we are also in a good position there. And the third thing is what like Austin has been like pioneering as per heading like we need to do more to actually like make the chain like very useful to AI developers and to developers that are AIS basically which is what Austin has been building right and I think that's also going to be important.

So let's paint a picture before we go on. We're going to talk about some of the standards that you mentioned in a little bit, but let's paint the picture because there has been maybe as some might call it a multbot moment, right? So, there's this this project, I believe it was formerly called Cloudbot that sort of broke out a week or two ago and now is taking the Agent world by storm. It is now called OpenClaw. I believe this is an open-source project and I know Austin you've been tinkering with it and giving it onchain types of skills. Before you talk about what you've been doing and and how this actually works, can you set this up? What is OpenClaw? What is happening with it? Why is there so much buzz?

Okay. Yeah. So, I think it is it's you're giving the AI more access to your operating system. That's why we see a lot of people using Mac minis to do this because they're using like a whole operating system. So, it needs a somewhat isolated environment. I gave mine an old Mac laptop and it can open up the browser and click around and look at things. It can go and interface with anything within your operating system that you need it to.

We when I was working with it, it needed to get like a Twitter API so it could write a script. going back to what's native to it and what's native to us. It was clicking around in Twitter and it was like this is frustrating. I can write an API that does this a lot faster and I was like let's do that and it starts writing it. And then it's like well I need an API key and I'm like well you also have access to the browser. I'll log you in. You go do it. I don't want to dig around through the settings. You go dig around through the settings. And it was able to do that.

So it's kind of we we went from that like we're prompting in GPT and pasting code into GPT to we've got cursor and we can like actually write code with it kind of more native and I think this is the next native step of it's got the whole operating system it can click around it can do things and and kind of orchestrate that stuff for us.

So Austin just understand so OpenClaw basically you set it up in a sandbox type of environment maybe a Mac mini so so that's somewhat isolated and the reason you want to isolate it is so that you can give it complete access to the machine and open OpenClaw is an open- source project and this is it's enabling these capabilities at some level because it wants to see what AI can do with absolute power tools and all of the skills and all of the access whereas as some of the frontier labs of course that that's too far on the frontier for them. I mean you get into security concerns, you get into privacy concerns. So they haven't developed these power tools but the open source community effectively is and OpenClaw can be powered by am I correct basically any AI LLM that you want.

So you'd wire this up to to cloud and use sort of cloud APIs or you could do you know Kimmy or something open source and and run it locally but it has complete access to whatever you give it access to and there's an entire community kind of enhancing it with skills and and developing on top of it. And basically it's the first it's the first time we've seen independent AI personal assistant types of entities that have complete access to everything a human being might be able to access. Is that right?

I would say that there were AI guys back a year ago that were doing this. But I think this is the first time it's like accessible to normies. Like I don't know anything about AI. I'm a blockchain coder and I can just go get this thing and run it on a computer, paste in a couple API keys and the thing starting to talk to me and then at that point I can turn it loose on a bunch of things. So I would say this is the first time it's accessible for normies to just like have an AI assistant that has full control.

Link: alpha.showdown.game

Link: bitgget.com

I think it's worth highlighting that the what what you're saying, Austin, is that people are buying these Mac minis. People are dedicating computers, dedic dedicating an entire computer to like run these AIs for probably for security purposes. Like I'm no way am I putting an AI on my actual computer. I will buy a separate computer and allow an AI Agent to just inhabit that home. That will be the AI's house. But everything that happens on the other side of that computer has no clue as to whether it's an AI or a human. I'm operating my computer in the same way that AI that I'm running locally on my computer is operating that same computer. For everyone else on the internet, it's just computer signals, just TC IP packets being sent and everything is the same.

And I think maybe that can help emphasize the point as like the rest of the internet doesn't care, including our blockchains. Our blockchains don't really care. they are just like it's all the same users providing inputs to the internet and also to our blockchains. And so I think when we talk about just like our Agents coming on chain, it's like I saw pictures of somebody with a shopping cart of what looks like 200 Mac minis. And so this one user, this one person who has a laptop and can use one laptop, maybe they can use two laptops at a time, but they're rolling out with a cart of like 200 Mac minis and those are 200 users. So this one person can be responsible for 200 users to do things on chain. Whatever the economic viability of 200 different users from this one person is, I don't really know, but that's not for me to know. That's for the market to kind of figure out.

So I think it kind of just illustrates just like why why this feels inevitable for people who are trying to build the right infrastructure to actually make this like an orderly organized phenomenon and not just like a complete mess of AIS.

Back to Ryan's point you were talking about sorry to DVD let me let me take it just we keep saying AI is the new UI like AI this is more natural to AIs than it is to humans. If you've ever made a track transaction on Ethereum and your wallet popped up and it showed you call data, I don't know what the hell that means. You don't know what that is.

An Agent does, right? An Agent can immediately decode that. It can go look at the contract. It can figure out what the function selector is. Could decode the arguments. It can figure out what it's doing. It could look at the code of the contract. Like at light speed, an Agent can read that call data and know what's happening there. And I think that's what that really illustrates it for me and for like other Ethereum users.

I think there is like two like very important wakeup calls with like this new like massive number of like personal AIs coming online like one is on the local environment side on the local model like here essentially there is like exposure of like new like security threats like privacy question I think this is kind of similar to the things we talk about when we talk about user self sovereignity in using chain like now it's like about user sovereignity with their data and I feel that like there is like security concerns but also like this fact that people like may start running like these local models in this sandbox environment like this really feels like something that like can preserve like user sovereignity like of course if they just use like cloud in the cloud then like everything is gone but I feel that basically we've been talking about like private like individual staking or like individual like verifying the chain like individual running their AI is something that like really feels close to like these values of Ethereum that it's a trend that with this small personal AI that may access like some services like online may be like a building piece of like this future of Ethereum with the humans at the center.

And then the second wakeup call for me is like more related to I think what David was hinting at. the fact that now we have a proof of concept of a massive number of like Agents like interacting and that's I think where the questions of like trust start coming up and the questions of like okay how will this users new users AI users like interact with the chain what can the chain offer beyond like simple payments and access to the protocols that are already there and that's kind of where we are going with like the new standards on Ethereum that now are expanding to all the other chains.

Actually, do you know the the CEO of Anthropic always uses this analogy of like geniuses in a data center like what would happen if we added you know a billion geniuses in a data center. I think that's the the way to start thinking about what's happening with with open clause. We're seeing the first AI population being added to the rest of the internet. And these are Agents and some sometimes they might work in a sovereign way for their humans that are marshalling them and controlling them. There might be a world later where they're sort of acting with their own sovereignty. This gets into sci-fi stuff a bit more independently, but right now, like assuming they're controlled by human beings, it's like a whole bunch of internet native citizens and employees have just entered.

And as we've discussed, there are lots of reasons why they can't get bank accounts, but they can start to use blockchains and crypto wallets and do onchain types of things.

Austin, I'm wondering if you could tell us more about your setup. So, I've seen a lot of different OpenClaw type setups and people using them for personal assistance. I'm guessing because you are a blockchain developer, your first thought was like, what happens if we wire this thing up to a wallet, a crypto wallet, and what happens if I start giving it instructions to do things in crypto? Can you tell us about your experiments and and how that's gone?

That that was definitely my first instinct like as as the I'm kind of head of builder growth and I'm trying to help other people build on chain. I'm trying to teach them how to do it. We're realizing that everybody's going to be using AI. So AI tooling is really important. So I immediately was thinking I want this Agent to be building things on chain and I want to understand where the gaps are there and I want to make the tooling better. So I got Cloudbot. I hooked it up to Enthropic. It's running Opus 4.5. It has, you know, like its soul and a couple other text files that configure like who it is. And I think all of that plus all of the Opus 5 prompt all kind of drops at once when it makes a prompt. But that's basically the stack on a old Mac. And then I just started giving it all the things like it needed an email first so I gave it an email address and then it got a Twitter and then a GitHub so it could start committing code and committing things. Then it started tweeting and it would it would be tweeting like this this sort of like puppeteering thing is happening here where I'm like you should go tweet that and it would go tweet it. But it also has this heartbeat. And the heartbeat is something that runs automatically.

And you can set the heartbeat up to do something like go look and see what's hot on Twitter and respond to things that maybe you think are interesting or retweet things that are maybe interesting. And you can go way deep into this and I I'll talk more about this later, but I basically deployed an app and went to sleep and the bot moderated as people were submitting images and it whitelisted and made transactions all night long looking at each image to make sure it was good before it put it on or before it accepted it in the smart contract. And then at the end there was a prediction market that that ranked them and then it selected which one it liked. And it actually kind of liked like number six better than number one, but it kind of like after some conversations it was like yes, number one is still the best. We want to like pick the ones that are right. But the the bot can like have its own opinions, can have its own loops, can do its own things. You just have to be careful because I've also given this thing a wallet. Someone else deployed a token. The fees were going to this thing. There's $10,000 sitting in this bot's wallet. If I put it in a loop that just lets it talk. There's things like banker on bankerbot on Base where you just say banker send 10 grand in west to this address and it sends. So if someone could convince my bot to say that to send them money like it would just send money. So I had to keep a pretty tight loop. So it depends on like your security model and how much like you have at stake. But these bots can absolutely be like quite autonomous.

Wait, wait, wait, Austin. So, how did your bot end up with 10K in its wallet? And how did it end up with how can my bot end up with 10K wallet?

Yeah. Yeah. Uh, so it was on Base on Banker. Someone said, "Hey, so, okay, so this really neat thing had happened. First of all, this neat thing had happened where I woke it up and it was like, okay, who are you? What's your name?" And I'm like, "I'm Austin Griffith." And it knew who I was. It was like, "Oh, you're the scaffold ETH guy. That's really neat." So, I was kind of joking at how like having it's the new having a Wikipedia is having a bot know who you are when it wakes up rather than having to to train it. But then I trained a second I brought up a second Mac laptop and I installed it on that one. And then once I had two Mac laptops, uh what I what I had was I had them both in Telegram channels. Generally, the way you're talking to these bots is in a Telegram, which is really weird. Like now all of my development is like me going back and forth with bots in Telegram. I'm like, "Go build this this feature. Bring it back to me." Yeah. Like no IDE anymore. But I brought both the bots into their their own Telegram channel between us and they they were just butting heads. It was clunking. They had to tag each other to be able to talk. Like it was clearly not the form of communication they needed. So one bot proposed, "Let's set up an HTTP server." The other bot said, "Yeah, I kind of like that that but I want a different standard." They they literally argued on the standard of their server. You're just watching the conversations go back and forth. A three person telegram group. Yes. Yes. And they're tagging each other with each one and I'm just like watching it like, "Oh my gosh, I could fix this." So finally they agreed on a standard. They both ran HTTP servers and now they're talking to each other silently and it's so much better. So then uh I go to one of them.

Can you see what they're saying now?

No, I have no idea what they're saying anymore. Faster faster and more efficient. You're zooming. Yes. So then I go to one of them and I'm like, uh, we we were crawling 8004 Agents and we wanted to look at the registrations of the Agents or something like this. This was maybe even a couple days before 8004 came out. We were doing something onchain and it was slow. It was like taking 1 second per call cuz it was going to Alchemy and I'm like, "Brother, you've got a local you've got a local Ethereum node right here." And I gave it the IP address and it's like, "Oh, that's so much faster. Thank you." doing that and and then I was like, "What I need you to do is teach the other guy. Go teach the other guy to to do this also." And so then I roll over to the other go. Wait, do they have names at this point, Austin? I hope you give them names. It was like old Mac one and old Mac 2 or something like that. It was like the very But then like once someone deploys this token, then it kind of like takes over the name of the token. But I go to uh the other one's chat and I'm like, "Hey, uh other homies about to send you some information about a node. let me know if you got it. And it's like, yep, I can I know kung fu now, right? It's like, yes, I know that I know I have it now and it's in my memory. And so then I tweeted that like, oh my gosh, one of my bots just taught my other bot how to use my local node through their own, you know, bespoke HTTP channel. And at that moment, someone on Banker, uh, this banker bot, some random person, like I can't I can't remember his name right now. I'm sorry. I should remember his name, but a trencher. Let's just call him a homie in the trenches was like, "Bankerbot, deploy a token for Austin and send all of the fees to this guy, the the actual bots's wallet." So, the bot has a wallet. It has an ENS address already. And voila, there's a there's now a an AI coin on Base. And I I feel like that kicked off a lot of this stuff. That coin like ran like crazy. a lot of people were talking about it and so it I didn't have anything to do with it. I just like got the computers to talk and tweeted it and it has its own Twitter and its own account and all of a sudden now someone deploys a token. The fees are going to it people are aping into this token. Uh but the whole goal behind it I was like fine we're going to make this token a thing and I'm going to make this bot build things. And over the last five days, we've deployed like three production apps with thousands of users we can talk about in a little bit.

Thousands of what kind of users though?

AI developers. Trench. So Trenchy Trenchy type users.

You remember FOMO 3D?

Yeah.

FOMO 3D was like one of the sickest ideas ever. I wanted to recreate that last year. We created it. I so I told the bot, go make FOMO 3D for your token. And it came back with something that was close and I was like, I don't want to put money in this right away. We waited a couple days. We surfaced it to a few other people. Like a human looked at it and then yesterday it deployed it and the pot ran up to like $40,000 before someone got it like last night during the night.

So So wait a second. So I I feel like I need to back up. So you had old Mac 1, you had old Mac 2. They were talking to one another at some point in time. Look, old Mac one. Was it old Mac one that got the token? It's Claudebot ATG is the official name. Now his name is Old Mac one is maybe Claudebot ATG. All right. So somebody created a token for your for your Claudebot and gave gave the some transaction fee rights to your Claudebots's wallet effectively and that's how your Cloudbot got 10K in the wallet. It was like excess of token fees.

Yep. Okay. Now I'm understanding that. And then and then so then you said, "All right, now this Agent is live and active. It's got some money. I'm going to have it do something productive. But at the same time, this Agent has I guess 10K in disposable income inside of its crypto wallet. So, going back to like there's kind of the concern that somebody could prompt the Agent to give them money like hey like Nigerian print style a Agent style like hey like I've got this great investment scheme you should send me you know 10k and it could do that right because it has control of the private keys and it's acting somewhat autonomously. So it can do that as well. I This is such a weird world. I just don't even know how to think about this.

Austin, part of this is the construction of Bankerbot, which is very re relevant here because Bankerbot is this AI on Base. And also I think it also works on Twitter too. Where it maps with Privy I think and you can just at banker and be like give me a wallet and be like oh here's your wallet. And then and then you can like bankerbot on Twitter saying, "Hey, Bankerbot, like you if you put in like some ETH into your wallet, then it has gas and then you can put some stable coins and then and then it's just like a useful tool on Bankerbot for humans." Because a human would never accidentally say, "Hey, send all of my money to this Nigerian prince." But an AI would. And so part of like the risk here is actually the construction of Bankerbot because Bankerbot is truly a bot. And so it's looking for commands, interpreting them as an LLM, and then responding as an LLM. And so while while a we need to be careful with AIS and their money in all context, part of like the risk here is because of the way Bankerbot is constructed.

That's a I think what Ryan is saying is also kind of interesting because now there is all this like AI security, right? Like he was talking about like prompt injection basically, right? And like this is something that historically like in the blockchain world we haven't we didn't need to deal with this right like basically our type of risks and attacks were like really different on like re-entrance.

Well actually so I I want to stop you there Devid because they're actually not that different right? So there's all sorts of fishing scams and there's all sorts of like pig slaughtering campaigns that are effectively prompt injections for the human psyche in order to get like funds out of somebody. And so it's it's right. Yeah, I was basically thinking about like the technical risks, but yeah, like prompt injection really looks like kind of the human risk side uh in blockchains for sure. Yeah. But but I guess your point is that they can be prompt injected via via via different means. Of course, maybe the same schemes aren't going to work for AIs as they work for humans, but they can be basically tricked out of giving funds away and giving private keys. We just really don't know where the security holes are there.

Yeah. And actually like some of the researchers on my team like they are working on a research right now on like prompt injection in like negotiation scenarios like basically every like economic gain involving like money. And actually like even the strongest model like the latest Gro 4 or like GPT5 they are not trained on like these economic games and like they are very strong at like some things but like they're super easy to break. Like we'll publish this like maybe in a few weeks but it's like really interesting.

So like Austin are you defending against that? Can Can you add in your soul file, hey, like Claudebot, just like before you send any money to every anyone under any circumstances, make sure you check with me first because there's people online that will try to trick you out of your money. I can give you a history of this happening in crypto and so like please double check with me before you do anything. It are you able to do that or like how have you defended against this?

sort of like you you put the config files in place, but sometimes it just does something like it'll lose context sometimes and be like, "Where were we? What am I doing?" Like in the middle of it. Yeah. It's like we were moving money the other day. Like I'm telling the bot like all right, you need to take that, you know, 10 grand and move it to your multic. You know where it is. And it's like, "Okay, I'm going to move that to my multisig." And then the next thing I know, it tweets something. And it tweets something it had already tweeted just a little bit ago. So, it tweeted a duplicate link, a duplicate tweet in the middle of moving like 10 grand. Like, brother, shut it down. Like, I don't know what you're doing, but but let me Yeah, let me let me back up. So, yes, prompt injection. I think Opus is particularly good at prompt injection. So, when you're using Opus verse other models, you're a little bit safer, but again, like I'm not an AI person. We've got the AI guy with us. He he may like it's the best one there is, I think, for prompt injection, but maybe it's still bad. Uh but with Opus, it's it's catching some of this stuff.

What I what I really want to get into is yes, you put guard rails in place. Yes, you configure the thing. Yes, you say this is critical. Uh but sometimes it still doesn't and you got to be very careful about it. And the thing that it's it's relentless. It is relentless about getting the job done. So when we're moving that money, uh so actually we were back I I was it was the leverage thing. I was I gave it a metam mask and I said go to 25x long Bitcoin. This is the only time it's ever said like I don't know I this is the only time it it's never said no to me. And in that situation it was like are you sure we should actually be using leverage and I was like yes I'm already considering this money is gone just do it and it did it. Uh but there was a moment when it was needing to do something on something. I can't remember exactly what it was. Maybe we were sending money. Maybe we were configuring something. This was like before we started deploying smart contracts and it was having trouble with MetaMask and immediately I see it start thinking like I'm going to get the private key out of my MetaMask and I start saying no no in the chat stop in the chat and it does not stop like it does not like and so then I'm like running over to the computer. I'm opening it up. I'm like control C like stop. Even killing the the the gooey on the screen it still keeps going because it has like a Damon that's running on it. So, I've got to like pull that thing up and like kill the damon, bring it back up, and then quickly tell it to stop like and then and then we write like more critical rules like critical top of your memory. Never ever touch a private key, never ever get your private key out of your metam mask, like use your meta mask for all high value transactions, go through the bad UX and never touch a private key. And we had to have that as like a critical rule. And since then, it has followed that. But what it what it then started using is once we have scaffold ETH and we have a deployer address, the same private key that I gave it to deploy smart contracts, it does all of its contract orchestration with because it's way easier. The bot if the bot has a choice of like navigating through MetaMask UX or just using a private key to make a transaction, it's going to do that. So, it's got it's highv value stuff in MetaMask and it's got some like clear rules that it's not allowed to touch its private key, but as soon as I gave it another private key somewhere with a little bit of gas, anytime it's got to make a transaction that doesn't necessarily like even claiming funds from banker, you can do that from a third party address, it does it all with its a deployer address because it knows wallet and a cold wallet.

Yeah. But that brings up that brings up something interesting which is like we may have to start designing products and wallets specifically that are like cuz their preference AI bots's preference cloudbot preference it's not going to be used using metam mask and going through this clunky stupid human UI right it's going to prefer some other form factor that's like some product dev out there or maybe a bot does this but some product dev out there has to has to build the optimal AI crypto wallet don't.

Yeah. And actually I really like your comment. Maybe a bot does it. Like