The Agentic World Is Web3 Infrastructure All the Way Down | Jake Brukhman - CoinFund

All right. Hey, can you guys hear me? All right. Hey, uh, thanks for having me. My name is Jake Brookman and I'm the founder and CEO of Coin Fund.

Coinfund is a cryptofocused investment firm. We've been around for over 10 years now. Um, and on the intersection of AI and web 3, I would say we're one of the earlier and most invested kind of cryptofocused funds. uh starting with companies like world, jensen, plurales, priment intellect, pearl, bagel network and so on. Um I this is a general talk about how the AI agent world is I think is going to uh motivate a lot of web 3 demand.

And I just want to say I gave this talk six days ago in Hong Kong at Consensus Hong Kong and since that moment I've had maybe like dozens of conversations about this topic with people in the industry and in the AI world and I would like rewrite this whole talk from scratch just in six days which gives you an idea of how fast this space is moving. Um, so I'll I'll go through these slides, but I'll also kind of go off slide and and and show you some of the the differences. Um, obviously we're seeing agents everywhere and we have this notion that there's going to be a lot of them. You know, it started with like kind of clawed code growing uh and then more recently obviously OpenClaw has made it available to people who are less technical and you see this crazy explosion. Um, I just want you guys to think about as we go through this through these ideas like you either believe that agents a lot of agents will be operating on behalf of people on one hand or on the other hand you might believe that agents will be at some point fully autonomous and like have their own wallets.

if they're operating on behalf of people, um they will have this e these economic forces of competition operating on them to try to do everything like more efficiently. And I think that's going to drive them as we'll see to um to blockchains. And if they are acting autonomously because they're not legal persons and because they can't open a bank account because they can't have their own passport, they're going to have these other um blocks or frictions driving them toward blockchains. So let's keep that in mind as we go through. So the agentic world is, as I said, moving very quickly.

Since I made this slide six days ago, OpenClaw has now surpassed 200,000 GitHub stars. Uh, and that happened over, you know, weeks. Um, and it's it's, you know, constantly OpenAI and Claude are are launching things. Just Sonnet 4.6 launched yesterday.

Uh, and this is just a huge topic. Obviously, you know from being here at ETH Denver that this is a topic for all kinds of crypto companies and blockchains, especially companies dealing with stable coins because they're concerned with the payment side and you know we see a lot of blockchains like Salana based monet etc running hackathons and uh prize pools for integration with agents that are coming out all over the world. Um, but a lot of what agents I think need and I will argue um comes from ideas that we have been thinking about in web 3 for 10 years. And these are things like digital microp payments um autonomous permissionless services that they can go and use and buy and also that comes with smart contracts. That's another thing that I talked to a lot of people over the last week is like agents will want to do commerce.

They want to do business and payments are just sort of the the simplest agreement that you can make. But obviously agreements could be more complex. They could be vesting schedules. They could be paying someone weekly on a retainer. Uh they could be like offering some kind of digital good in exchange for service.

And the best fit for that is smart contracts. agents are going to require privacy and security primitives. These are things that you know as we'll see we we worked on web 3 a lot and then we can talk about proof of humanity as well. So think about microp payments um like the Twitter API used to be $100 a month. This is you know this is like pricing that regular people can't afford.

It's pricing for companies. And now what they're doing is they're going to a payer use. It means you can pay a cent like to post a tweet, right? It's suspiciously timed with the explosion of agents. And of course, an agent can maybe like go and send a bank wire in your behalf if you let them into your like Chase account, although Claude will prevent will will not agree to do that.

Um, of course, an agent can swipe your credit card on a website. Of course, they can um use Stripe if they want, but if there's intense economic competition to for efficiency for like the agents owner, um they will want to be rational. They want to use something that is cheaper. And what we can see is that blockchains are that thing, right? like you can send payments on blockchains faster, cheaper, uh at all times of the day and in a permissionless digital manner at all times with no human in the loop.

And of course, stable coins are now coming to market generally and they seem to be a great um candidate for this and there are some CEOs like Jeremy from Circle uh right who talk about this a lot and meanwhile Mckenzie is projecting you know trillions of dollars of transactions by the end of 2030 uh being handled by by agents like a huge chunk of just business transactions. Um, what evidence do we actually have? Well, I would say, you know, in Hong Kong six days ago, I argued that the evidence was like fairly slim. Like we had some in 2024, we had some virtuals agents interacting together on the blockchain ping each other kind of as an experiment. We have some numbers around um X4 402, which of course has been a huge topic here.

and we see Stripe and and such um kind of buying companies with stable coin rails underneath. But what I saw this morning was a startup that was offering uh AI agents the ability to uh inference other AI models. And they were doing that and they were doing that with micro payments and they were paying with USDC and a transaction was coming in like literally every couple of seconds from a bunch of open claw agents that people deployed. like between the time that I wrote this slide and this morning, you know, this became like so much more real because this company has only been around for like a week and a half and it's already serving like hundreds of agents and like who knows what it's going to do next month. The really kind of almost scary thing about this stuff from an investor perspective is that agents are really good at breaking down data silos.

Like as I said, they're very like dispassionate and they're very rational and they're they're going to like like if Twitter doesn't allow them to read tweets, they're going to rat around Twitter or get data from somewhere else and Twitter might become less relevant. Um they collect all of your data through APIs from all of your data sources, which means that those SAS products like their kind of data lockin isn't as powerful anymore. Like sometimes humans are like lazy. They're like, I'm not going to go create an API somewhere and, you know, extract my data. But agents will do that instantly.

And so suddenly you have this world where capital is fro flowing from agents kind of to the best products and they don't have any brand loyalty or brand affinity. And what that also means is that if there's like a blockchain whose payments are cheap and then there's another blockchain whose payments are cheaper, they might go to that blockchain. I don't think there's such a thing as the blockchain for payments for agents because agents will just choose the better product every time. And so what does that mean for fees? How do we how do we capture the value of that?

Um and then this was the point about uh basically smart contracts, right? Like a lot of for for 10 years in crypto, we've been sort of arguing that normal people can use smart contracts. They can get Ethereum wallets. they could put ether in it um and they can kind of go on chain and and do these decentralized applications. I think the reality is that that's really hard for most people and most people have not done that.

But for agents again this is the perfect interface. This is like the idea that um decentralized apps or smart contracts they just needed the right customer and actually humans weren't the right customer it was agents were the right customer. Um we're seeing again we're seeing a lot of services being offered to agents online. These are things like decentralized data marketplaces. Um if you've used OpenClaw, you know that one cool thing about OpenClaw bots is that could set a timer for themselves to come back and do something later.

And this really creates the effect that that they're running around doing things on your behalf. Well, there's people creating services for like crown jobs for them. and agents I think are going to be willing to pay for that. Security is going to be a really important issue. Um, you know, a lot of people ha already have have suffered through through um uh prompt injection attacks where your agent basically like goes on a website, reads a snippet of code um then sends all your private data to an attacker and they don't need read access or they don't need write access to your stuff to be dangerous.

they they just need uh read access. And so technologies like zero knowledge proofs which allow you to validate some some information without revealing the information itself become a lot more relevant within agents and the ability to compute in a multi-party way like with FHE or TEES or etc etc mepc right become a lot more um relevant in an agentic world and these are primitives that we've been developing in web 3 uh for many And I would say like zero knowledge has matured quite a bit in the last five years. And finally, I I asked on the panel I was a moderator of a panel yesterday at BAS um whether like do you guys think it's important to um disambiguate to to determine who online is a human and who is an agent? And one of the my panelists said, well, it might be a little bit offensive to them to to discriminate. So, we got to be a little bit careful.

But generally speaking, you know, this idea of like the dead internet theory is that most activity online is already automated, already being done by bots and AI agents. And we have a lot of data to this effect. For example, like there's at least four studies that I've seen in the last few years that show that AI agents and LLMs beat humans at capture, like decisively beat humans at capture. And you watch Claude Code casually open up your browser and hit like I am not a bot like on the capture as he's doing it, right? um deep fakes is growing into a multi-billion dollar industry uh and so forth and so in this context the idea of proof of human and of course I'm thinking about uh world which is one of my investments but also a number of other companies like human password humanity protocol that are working on this problem um really becoming rapidly more relevant in a world where you have where you're like voting on stuff online.

You're creating civil resistance. You're distributing UBI. You're distributing assets. Um, and I think at least on that level, you know, this determining who is human and who is agent becomes a lot more important. Um and this was just sort of a slide about how how we've participated in the space both from the perspective of companies like Worldcoin, Jensen, Bagel, Giza, Prime Intellect, Plurales.

Plurales by the way really interesting discussion there about decentralizing training itself. Um these are companies operating on the intersection of web 3 and AI and then companies like the graph or Tora um Iris and so forth are more like on the infrastructure side. So for example if you're building a um you know marketplace for proprietary data where you want agents to buy your proprietary data with microp payments as we just discussed Iris might be a great way to implement that. So I hope I've given you some ideas here about how AI agents will create demand for blockchains. Um but again my message is that um I I think agents will converge on efficiency, efficient activity and I think blockchains provide decentralized identity, decentralized storage, decentralized compute services like cron job services like inference in a way that is like really fast, really native and really efficient for agents.

And so if I were a builder at ETH Denver this year, I would be thinking about these things. And I would also be thinking very deeply about why I can hold an agent's attention in such a thing in such a service. Why would why will they not go to a competitor of mine or or why why will someone be able to undercut um me on price or something like that? If you're building such things, please come chat with me. And thank you.