Ethereum's OpenClaw Moment: Automated, Safer, and Easier to Use | Jacob - Coin Fello

[music] Hey, Ethan Denver. Uh, so super excited to talk to you all today. I don't know if you guys got to see the opening ceremonies, but we had a lot of discussion about AI agents, about user experience, about use cases that are now possible that were not possible. Um, so I want to talk to you a little bit today about uh what I'm calling Ethereum's open claw moment. This isn't a talk only.

I mean it'll touch a little bit on openclaw itself but it's going to be a larger discussion about the implications of what openclaw means for us as builders in the Ethereum ecosystem. Uh I I believe that we are passing a threshold where onchain activity can be automated. It can be safer. It can be easier to use and accessible for mainstream audiences. Uh before I I get into that, I'm just going to give you a quick background on who I am.

Uh what the context of what I want to present to you today is. Um my my background is I've been working on uh creating great user experiences for Ethereum. Uh since since late 2019 2020, uh when I joined the Metamas team, uh I was part of uh building the growth model and the monetization model and all those things in the wallet space. And I had felt that there there were some major challenges uh especially in the 2021 2022 cycle with how we were onboarding people into Ethereum and uh so to so to talk about that and um you know near and dear to me I'm an Ethereum radical and uh when I see the challenges that we have um I see that the previous model of how we interact with smart smart contracts has basically been saturated. The people that we are able to onboard in the previous model, they've pretty much been onboarded and uh you know uh we heard uh great great thoughts from Tomas uh this morning about how we 2025 we were basically told there's there's two paths.

There's uh pivoting to institutions and there's uh building the uh c casino tech. And the the reality is the previous user experience is dictating those problems. So um if we rewind the clock to early Ethereum 10 years ago uh DevCon um Devcon 1, we had the really exciting uh use cases that were presented um at Devcon 1. We we famously had you know Socket opened a lock using an Ethereum transaction. We were supposed to get all of these use cases that you see here on the lefth hand side and we got uh I don't think hardly any of them.

Um we got primitive versions of them and uh what we ended up getting was you know uh the the casino tech. Um and and a lot of for me as as an Ethereum, as a as an early Bitcoiner, um this was very far from what I set out to to join. And some of the earliest agentic use cases, I'm going to make fun of Pump Fun. I I love you guys, but uh building an AI agent to fake volume like we can use AI agents in much more interesting ways. And uh just to to underscore this problem, the the 2025 like everything has to pivot to institutions.

This is what we were seeing. The number of wallets and the number of active wallets were going down, but prices were going up. And that looks a lot like a previous adoption curve that if we are if we have Ethereum's cippher punk values, we should be super alarmed because the Linux adoption curve um which is what I'm uh pointing out here is Linux one. It it runs all the infrastructure of our world but overwhelmingly just in the enterprise space. We seated the user space to proprietary surveillance tech and uh there is a different adoption curve that Agentic AI opens up which is the adoption curve of the internet itself.

So um why is this possible now? Why are all those use cases that I was pointing to in that previous slide possible now? It's because we have this trifecta of new innovations that are all possible. So um when we look at this, we now have the ability with 7702 to upgrade these EOA wallets into smart wallets. And with 7715 and 7710, we can actually get delegations and permissions from users existing wallets so that they're not approving every individual transaction, but they're not exposing their keys to a a centralized uh trusted third party or not so trusted third party.

um instead we're we're uh with these EIPs we're getting permissions to do things automatically. We also now through EIP 8004 we have open standards to register AI agents to be consumed by other agents and to uh to pay for those AI agents through microtransactions like X42. I would encourage you all to go and check out some of the great X42 talks that are happening this week. And then we we also have with with open claw we have personal agents. These are not just research models anymore.

We have agents that are carrying out execution on behalf of the user and that are able to to do things while we sleep. So when we bring together those three pieces, we have a moment where we can now build self-s sovereign uh AI products that abstract away the complexity of the crypto user experience and make this accessible to everyone. We can restore the cipher punk values of Ethereum while doing so in a way that also brings back great user experiences and doesn't compromise on them. So uh we built Coinfellow. This is um this is our product.

This is uh the very first uh AI agent that is capable of generalized onchain actions. It can interact with any smart contract and it can also automate any smart contract. So this means that it's pretty radical. We don't need DAP frontends. Like for the longest time, some of the main FUD against Ethereum was DAP frontends.

They have bad user experience. They're a major censorship point and they just generally are not how most people use the internet. chat GPT is it it has a mass adoption moment and this is how the majority of humanity is going to interact with the internet. So if we um you know if we were to put in a prompt in this I'm going to give you a really simple one. So let's swap some USDC for ETH.

Um in this example it it can't just be a chat interface. So what we we've built is a combination of chat models that are familiar as well as rich crypto UI that people can interact with and they can complete onchain actions um in a way where the complexity is abstracted away by the user's agent. Um we also then get the opportunity to do automation. we can build uh automations through prompts that allow our agent to act on our behalf and protect our interest and allow us to navigate uh the Ethereum ecosystem. So here's a here's a uh three examples of that.

Use the liquidity in my wallet to ensure my loan positions are not liquidated. We're all being liquidated because the market is on fire. Um this is uh pretty familiar I assume to a lot of the people here. Uh so an AI agent you can give it a delegation and permissions to carry out actions on your behalf up to the thresholds that you define without sub without tons and tons of transactions or being glued to your phone and being terrified that the market is about to dump. Um you can also give this second prompt.

Uh you you could batch a thousand transactions together to sell all of the meme coins in your wallet and move yourself out of memecoins and sell them all for ETH. the agent can create a a proposal for how to do that for you and carry out all of the complexity for you in a in a simple prompt. Um the third uh example here is an ongoing action move the profits or my volatile assets into stables. This is pretty self-explanatory but this is something that we would have never been able to do uh without aentic AI. So underlying these innovations I want to point out a couple of major innovations that have happened in the ecosystem.

Um it hasn't gotten as much publicity as it should in my opinion. The MetaMask smart account kit allows us to do these these forms of permissioning from users existing liquidity. And so this is what we have coinfellow we built on top of this stack. We want users to be able to bring their existing wallet not be forced to onboard into some proprietary wallet for their agent and only putting you know you know putting a small amount of funds and hoping that it doesn't hallucinate and lose all your funds. You should be able to control the permissions of your agent um iteratively.

And we need models that run in a self-s sovereign environment that uh is is controlled by the user even if it's it's hosted. It's fast. Uh but it needs to be deterministic. It needs to prove that the model that's being run is the model that it says it is. Um and you know I would refer you to Vitalik's talk this morning.

Um he talked a lot about this um during opening ceremonies. This is a rough example of the kinds of user experiences that we're building. So you can onboard either using your existing wallet or you can deploy a new smart wallet using an SSO credential. Um then once your wallet is connected, the agent can actually look at your asset history. It can look at uh your deposits, what you've done on chain.

And for us, we have biased the the system prompts and behaviors of the models that we run around deepening users into the DeFi rabbit hole. We want uh we want to surface new opportunities, things that they haven't that they're maybe aren't aware of and help them to go go and explore the Ethereum ecosystem and all of the vibrancy that builders are are building here at um at ETH Denver. We also need um human verification of any action that the agent is going to carry out before the agent does that. And so after the model uh produces a plan, it's going to have the human uh review that that's not enough though uh to really do this safely and to protect users from hallucinations and those sort of things, we need a multi-model uh architecture. So this means that a single model is proposing, another model is reviewing um every subsequent action that the user is doing.

Um, it's going to need context and privacy. Um, which is not frankly how open AAI or Claude or or Grock are built today. Um, happy to talk to you about how we're going to accomplish that in in future things. I would also point to some of the great work that the Venice team is doing um around privacy. Um, and then we need specialized evals for cryptocontext.

And we also need the ability for models to both be pruned and to have additional data sets added on top of them through Laura and Dora and those sort of things to uh to improve the performance of onchain onchain actions. Vitalic um uh last year at the end of last year talked about the risk of the AI hegimon and this is sort of like a dystopia with a single AI that controls our lives. Um it's it's kind of like what's in Terminator. And uh so the trend that we're actually seeing though with with OpenClaw is that we're actually seeing tons of agents that operate autonomously and interdependently uh with each other. And so I'm very excited about this.

Um I want to again point people like to uh that delegations and permissions are a better pattern where you can trust agents to do specific things. You can delegate rights from one agent to another agent and create the agentic economy uh using this architecture. And also with 8004, we should be registering our agents and their standards and how to communicate with them between each other. This should not just be something where uh we build centralized APIs like they use in web 2. Uh no, we need agents to be able to permissionlessly interact with each other um on on uh behalf of the what users have defined.

So for us in in Coinfellow, we believe really deeply in the 8004 flywheel. We are registered as uh or or will soon uh be registered as an 8004 agent and then other uh we can consume other agents as well. So this is um developers can consume us inside of your products um and we also consume your products. There are a lot of specialized agents. We've focused on generalism for our agent, but many people are also building other agents that solve specific use cases uh that are well beyond uh the the specificity that our agent does.

Uh so uh this said um please go and try uh Coinfellow. We have been in we're in a private alpha today. So you can actually sign up to uh to try this to give us feedback. Today the product is still using um it's still using individual EOA transactions. We'll be transitioning to the to the delegations and permissions that I was talking about earlier in this uh conversation very shortly.

But today um we've opened up this special preview at Ethnver. You will only be able to sign up for this at Ethnver. It will go closed again and then we're going to do a public launch uh at ECC. Um and uh I think that is it for the conversation I think uh or for the for the presentation. I think we have about four minutes to take um a couple questions if anyone wants to uh to ask a question

right there.

Uh I sorry I couldn't quite hear the question. I the question was about credentials.

Yeah. So um for for us like there there isn't a universal binary uh trust model uh as user. So when a user onboards into coinfellow they can make a certain amount of of uh they they can make a very small allowance um and grant permissions over it. And one of the things that's great about 7710 and 7715 is you don't need subsequent transactions to increase uh the the permissions or to to grant additional permissions or to to increase the allowance to your agent. So over time as you iteratively trust this more and in specific contexts you can do that uh without uh without forcing people to have uh hundreds of wallets and fragmented liquidity uh which is kind of the the default model that we see most people building in today.

Uh we have another question right here.

Yes. Yeah. Yeah. So the question was how do we think about insurance risk to users um in those contexts. So one of the great things that um so one is we need AI agents that specialize in that.

We have built a generalized agent and our generalized agent can today use a lot of the off-the-shelf DeFi uh insurance protocols and one of its behaviors because we're we're making recommendations based on things that you haven't done. So, it will actually iteratively start introducing you to concepts like insurance over time, um it can also delegate and call other agents that specialize in insurance uh to uh to to get you the best possible experience. And that's kind of our approach with everything. So that's that's our approach with token swaps. That's our approach with discovering DeFi protocols, with discovering insurance primitives.

Um everything that our agent uh is focused on is about uh being the user's entry point into the entire decentralized web. Um I as I I mentioned earlier like I'm uh from the the the MetaMask team and our our vision and strategy was always or I was from the Metamas team I should say. Um, our vision and strategy was always maximal extensibility, give users access to the entire decentralized web, whereas I think most products in our space trended towards specific integrations and saying like you have to do that and it sort of is like building a dumb phone when Ethereum is all about uh extensibility uh being touring complete being maximally uh programmable. And so we think that the AI the the AI agentic space it needs to not be confined to simple use cases. It needs to give access to the users across the entire decentralized web.

Um I think that is all the questions I or time I have. I uh yeah uh I have 30 I have one minute. Uh was there another question? Uh we'll wrap there. Uh thank you so much everybody.

Um come talk to us. We have a booth that is uh right in front of the the DeFi stage um uh the New France stage uh right over to your right. We'd love to talk to you. Uh we'd like to talk to other builders. We'd like to talk to users who want to try this uh critique it, break it, and uh let's build together.

[applause]