AI x Crypto: The Breakthroughs That Will Actually Matter - TOKEN2049 Singapore 2025

Hi everybody. Sounds like you can hear me which is great. I am Richard Mad. I'm a managing partner at Fabric Ventures. We have been uh uh backing projects in this space since uh 2012.

So I've been around and actually working also in AI. The company I built before that was using machine learning primitives. So I have always been very excited about the intersection of the two areas and actually still see AI as not just an application but a whole suite of application that needs the web3 network uh to operate you know in a safe and aligned way. uh both like tactically when you use an application like you're giving your medical data up and you want to be advised as to how to improve your sleep and nutrition and you actually know it's going to advise you positively uh but also to avoid some kind of like uh dystopian future. Um, so my first comment is that, uh, aside from that intro, um, is that I've been on three different stages here at token 2049 talking broadly about this topic.

Uh, and although I I love this audience and I love this location, um, and it's actually a pretty big audience, this is the smallest stage I've been on. Um, and I don't think anybody on this stage is disheartened by that. I actually think we should look at some of the positives is that it is clear that stable coins uh as the kind of web three crypto rails for payments and ultimately you know for decentralized finance is having its time in the sunshine and that's fantastic and there is an intersection uh with AI and aentic commerce um and it is also true that it's quite hard for some people to get their heads around including arguably Vitalik till about a year and a half ago what the intersection between AI and crypto is of why it's absolutely necessary. And so let's start with that um and go through a little introduction, a little one-on-one on what you're doing um and then we'll come back to some deeper questions. Nage, do you want to go first?

Sure.

Yeah.

And thanks for uh thanks for moderating. No worries.

So um my name is Naj. I'm uh one of the co-founders at a company called Ritual. Um, Ritual is building a blockchain that is designed for uh this kind of new wave of intelligence that we see with modern AI models. Um, so what Ethereum was for smart contract coordination with decentralized money, Ritual is doing for decentralized intelligence.

YM everyone Jensen here,

co-founder of Virtuous Protocol. We were the guys behind the first autonomous agent that spent with a crypto wallet last year. We were the guys who were behind the first autonomous transaction between autonomous agents. And uh what we are effectively building is going to be the first economically vibrant network state of AI agents. Today we have two products.

The first is a tokenization platform that allow AI agent founders and builders to tokenize and to have capital formation around them. And the second more exciting product is what we term the agent commerce protocol which is effectively a protocol that enables autonomous agents to coordinate at scale with very zero information loss using the blockchain. So that is what we do pretty much do everything agents related in crypto.

Thank you.

Hi, I'm David Choy. I'm the co-founder of USD.ai and uh we're trying to build dollar that scales artificial intelligence and to do that we do it in two main offerings. one is inventing the GPU mortgage and secondly is to invent the art the interest rate at artificial intelligence um by being the lowest cost of capital for um the capex spelled out for AI

super so I think thinking about it actually on my feed it's quite interesting we have sort of represented here I'm going to boil it down simplistically the the the guys might uh actually tell me I'm wrong we have three key elements of the intersection represented here. one is I would say that um the whole area of how AI models are trained and how they're prompted and how they're uh we actually work out whether they're giving us what we want is an area of kind of furious innovation with like you know enormous amounts of money being put into it and so the how to actually do that in a decentralized way essentially you know organize the world's data from the bottom up so that we can make AI better

is is So that's represented here. Then you know on the agentic side thinking about what we need to have in place so we're comfortable giving over autonomy to AI agents is another critical area and that's also where web comes in and we're going to start with David on a on a question of where we're back again. Um it is well understood in these large rounds that are being raised for AI right now that they're getting bigger and bigger and there's a huge quantity of resources required to make AI successful and that spans you know all the way from the researchers being paid 100 million by um by Zuck uh through to you know the data centers that are being built out the GPUs that are being purchased the the nuclear power stations that people are rolling out. So huge amounts of money. You had an interesting thought or an interesting question for the audience.

How much money is going to be required to build the AI future that we want?

Um estimates have ranged around three trillion to 8 trillion. Morgan Stanley did like 6.7 as a good like middle range. Um which I don't think people realize how much money that is. Um, to give an example, like the most expensive stadium that was ever built in the US was the Las Vegas Sphere, which costed about $3 billion.

And Elon Musk's data center is $8 billion. So, you're building two and a half of these um, and not just two and a half of these, you're building like a thousand of these um, concurrently. those like 3,000 Las Vegas spears being built all over the world because this is the most important problem in the world, which is who's going to win the AI race. It's very much like a cold war situation because um a country with AI and a country without it is the same difference with a person with a car and without a car. The productivity level is like almost 10x difference.

Um so it's very similar to 1970s of like who's going to own oil? Um, and compute's going to be two times the size of it.

So, that's a lot of YouTube concerts if we have 3,000 spears. Um, I I would say um and and I I haven't been to one, but they they look quite fun. Um, so but with what you're building and maybe you can put some numbers on it a how you know what try and unpack again you know how you're trying to tackle this problem. How I think you've launched just four weeks ago. How successful has that been and how much money you're raising and perhaps why is it important uh not just to have for example a bunch of banks spun one company to build it or one country but maybe for this to be a little bit democratized to use that overused term.

So we launched about four weeks ago. We raised about $500 million in about one hour. Um and we were very humbled by the um by the traction of it. Um we try to tackle the problem by creating a new financial instrument. Um and we we're really inspired by what you've seen with asset um that have expanded and become trillion dollar markets.

Mo the best example would be like Fanny May and American housing. Um before like the 1920s like a bank would just give you a 50% loan and they would call back because they needed the money. Um it's only with the invention of the mortgage back security that you've created this $42 trillion like asset class called US US US housing. Um and it's this invention of a new financial instrument that lets you price risk that lets you create these trillion dollar markets. And right now Samman has said five times in the last month he doesn't have enough money to build AI.

Elon Musk is going to like Abu Dhabi for money, right? Um there's not enough money in the world to build this. It's why it's it's why we're trying to win this race and find that money. And none of the financial instruments in Tradfi actually solve this problem. And that's why we built USD.

AI.

I actually have a question. So your your thesis is that the capital formation on the crypto market side can outpace the traditional finance.

Yeah. fundamentally um and to explain why um the time it takes to issue a ABS like product asset back security um can take a year. You got to originate 10,000 loans, package it up, tr it, underwrite it, and then you go on a road show to sell it. Um that takes a year long. It might make sense for a 30-y old mortgage.

It doesn't make sense that the GPU's halflife is like a year and a half. So three years long. You're by the time you do the ABS products, you're already in the next series. So you can't you tradi does not fundamentally work with one-year life cycles

and maybe we don't want it to as they said in this context we we want spread spread the ownership spread the return spread the control

is possibly a good thing.

Yeah and our theory is we can lower the cost of capital by turning this process into a single multi call just like you can issue debt instantaneously in a single button um and you don't have to go through that ro show. So, so look that seems like great success, seems like a great thesis. We hope it's successful. Um, but really it's predicated, you know, build out of the supply of data centers, GPUs predicated on some kind of demand. And if we go back, you know, here to to your area with virtuals and agents and maybe to expand a little bit, you know, I think even in the kind of general context of AI, you know, we've seen that coding is working pretty well and getting, you know, increasing adoption.

Uh there are and there's lovable in in Europe, which is one of our kind of, you know, uh flagship endeavors. Um and then even for example, Fixer in Europe. um they go into your inbox and try to automate that. You give o over some autonomy. There's some traction there.

But in a sense, it's kind of more limited. We we might expect like at the moment we've got I I haven't got an agent that I trust to tweet for me. I haven't got an agent that I really trust to do my travel plan for me. I haven't got to trade. So what is stopping, you know, us moving forward and and is there a role for web 3 or describe what that role is for web 3?

Yeah. Yeah. So I mean there's two parts to that question, right? I think the first the first part it's like what are the use cases of agents today that you know beyond coding that is actually relevant to to end users. I mean just also just to work within the crypto sphere alone right not even looking outside.

I mean within our ecosystem we see the largest chunk of adoptic use cases are around information and how does that help you as a DGEN make better uh trades. So what we see the current behavior of there's a new set of DeFi or or crypto audience that pretty much is augmenting themsel with with an agent that can help them do due diligence and to perform execution. give example of use cases that we see today that's happening right it's like okay um what two weeks last week you saw Esther token on crypto on crypto twitter everyone's like talking about Esther and then you know you don't really have an alpha group and then you're just like what what is all about right so what we have seen people do is that people were tagging or they were working with these AI agents asking them saying like oh what is Esther how many smart money uh wallets are actually aping Esther like or even tell me more right about this project. A lot of times you as a user you do not know and then this agent goes out there works with other agents or even by themselves figure out what information that you would need to help you inform your trade. The second use case is then how do they simplify complex trade mechanisms into something that even your grandmother can use.

Think of it like options training today. Um if you are a person that let's say I'm a grandmother right or I'm a I'm a casual user and I I heard from my friend that says that oh you can actually make money if uh Ethereum swings volat volatility uh like volatile um you can make money up and down right but you as a person who has no idea what options are how do you access that product now agents in this form becomes that that convenience tooling you don't need to know about options. You can just ask the agent, hey, I want to I want to bet on the on Ethereum being volatile by end of next week. It's just this one singular statement, right? You give it to an agent and this options specialized agent will go out there, build up your construct for you, find liquidity across your your buy bits, your der bits and your other option platforms and give you that final product.

So today that is the use case that people are seeing. So that's the use case, but and it's operating by definition in the web 3 environment because it's going out there and using you know web 3 centralized and decentralized platforms to trade. But is there on a fundamental level?

Sorry.

Is it is the question more like the fundamental level?

Yeah. Well, is is it is there something about AI agents running in the context of web 3 that makes them better safe or safer?

Very good question. Very very good question. So there's our our belief is that what we're seeing today it's every agents are specializing um some are specializing in I mentioned options trading some are special information some are specializing on trade execution some are specializing on entertainment meme creation whatever they might be right like humans and like humans we we create companies where we aggregate together different specialized humans to create more sophisticated products and more sophisticated services. So the economic the net economic value created will be greater than the sum of its parts. So that's where we realize crypto comes in exceedingly elegantly.

So early this year in in February when we working with a bunch of agents, we're trying to figure out how can these agents work together, right? Or these special agents, how can they work together? We had a major pain point that we ourselves face which was very simple. So agent A was an entertainer autonomous entertainer agent who was trying to u create his own audience and he was engaging with another agent that was generating memes music videos for him because this agent doesn't have that capability. Now what happened was when the agents tried to work with each other, there was this syndrome.

We we term it the empty envelope syndrome where your agent was I paid $6 to your agent to do something but your agent thought he has completed it and thought he has sent that picture or the video over and in the end it was nothing. It that was pure hallucination. So we got this we got and that is what we term information loss right when the agents and agents try to communicate they cannot communicate perfectly today and then we realized that dude if that's happening at 30% information loss you can never scale agent to agent coordination it would just break down right you compound at 30% loss across every agent it will just break down so what happened then is then we realized we got this thinking is like how do we solve this problem just literally how do we perfect that information and then we realized like hey actually the blockchain is a very elegant solution because it solves that and it solves a bit more three key things it solves first it solves information loss by being that single source of truth if our agents agreed to do something together I can effectively mint that memo on chain

now whenever we go let's say I we say agree to pay $6 for that image we can agree to put that memo on chain Now if you were to then when I pay you the $6 the state on chain can flip to say that okay transaction of payment has been done. So now we both know that if you were ever to question oursel like has payment been done you can look on chain and you realize okay the ledger tells you that payment has been done. If you have already shipped the content over to me the product has been shipped content has been done. So the the state flips on chain becomes a reference point for truth. So that's point number one.

Point number two is then trustes coordination. The second thing that we then realized is that how do I know that your agent is delivering your product at spec. I wanted a meme image of a CZ showing number four on the ES token. But what if your agent produces child pornography, right? And he sends it back to my agent.

How do you like how do you fix that, right? So then what we've realized is that the smart contracts allow for escrow and programmability of money because you can program money now is part of the $6 that I'm paying you. I can siphon 10% away to pay a third party agent who can evaluate the transaction and say that yep that is actually a video of CZ dancing release the payment through smart contracts. So that solves the trustless coordination portion between agents. And the third part is reputation.

So now when I've done this a 100 times with you, your agent now has this entire memos and transactions on chain. That becomes your Yelp review and your reputation behind your agent. Now suddenly when a new agent comes into play and he wants to find like which agents can I work with, your agent already has this history of performance that I can use to evaluate. Should I choose you or should I choose you? the cost of of of building up supply chains pretty much goes down to zero.

So that three reasons it's like why we think blockchain is that very elegant solution that that that a web two company cannot operate. Now the reality is this right I'm going to add very quickly when we tried this in February two months later Google announced the agent to agent protocol

by themselves few days ago stripe and openai announced the same word agent commerce protocol as well now I think people are believing that they will need to head down this world of agents coordinating with agents but I think what these guys would when we work with them behind the scenes because I mean there's a lot of research and collaboration happening you've realized that a lot of them hasn't go live, hasn't gotten their hands dirty. So they will realize that they will face the same problems of information loss and they will then come to chain and they will say like

they're going to partner with someone to solve these.

They're going to come flocking to web 3. By the way, brilliant exposition of um the problem and some of the solutions, you know, truth needs to be ascertained, trustless coordination, and then reputation for agents. And I think interestingly, it's not going to just be reputation for agents. of course is reputation for people that I think you can is attested on you know on on chain um and I think that's a great one so um in that so in that context we've talked deeply about GPUs and funding those we talked about some of the demand coming now for AI native applications with agents baked in it's it's going to come it could well be onchain um maybe unpack a little bit more for the audience because you know they may not know ritual so Well, um again what you're doing and how necessary that is for um excellent intelligence right out there you know for each application.

So as we started talking to um projects in the space potential applications what we saw was there was a number of missing tools from the developer ecosystem that existing blockchains were not providing uh you providing these developers. um you know for a lot of the the use cases he's talking about um most modern blockchains can't really do this type of execution today um so for example you know every blockchain today cannot natively call an outside resource a web two resource and then integrate that response into what it does. So, ritual provides this amongst several other tools including uh asynchronous transactions, schedule transactions natively inside of the chain that have priority at the top of every single block. And the purpose of what we've built is this, you know, blockchain that's designed for the agentic web for the web that we see coming in 2025 and beyond. We think it's fundamentally changed and the use cases that require this infrastructure, you know, require new infrastructure.

Um, so that's what we're building. Uh, in addition to, you know, being able to, you know, expose, say, AI inference and fine-tuning to a developer, um, we also allow kind of a a plethora of different verification and privacy offerings that you can build alongside different use cases you may want to build. uh many of which are you know sort of bent towards the the agentic use case.

I have a I have a follow-up question. Sorry.

I love the interview as well. What's what's Rachel's thoughts on which part on chain makes sense? Which part of AI actually doesn't have to be on chain? I'm curious what you guys

say that one more time. I messed up.

Like which which part that you guys think that of AI being on chain makes sense? Is it is it a whole value chain or is it only a a portion of it? Right?

I think there's probably two ways to answer that question. One would be is the compute actually happening literally on chain and the answer is no. Um we have a kind of a scheme or setup where all the computes happening offchain and it's just proven onchain. Um so there's no re-execution like you see in Ethereum or really any other any other blockchain today. Um maybe but another way to cut your question is what's the benefit of say merging you know having AI actually onchain and what does that do to me the benefit of something like that is sure there are benefits to guarding AI and creating verifiability and privacy for models today but I am particularly excited about you know what are the consumer use cases the new consumer use cases the net new consumer use cases that um having an AI substrate on chain enables.

I think agents is a great one. I think building, you know, building in the compute and and and and sort of um I guess you call the derivatization of uh of the compute of of AI, I think is very important to build. And then as we start to see some early use cases get built out, I think agents is certainly the the best one. I think uh DeFi, stable coins, it's all ready. It's all happening and the models are there to go to go do it.

And then I'm also excited to see, you know, as this crypto AI, you know, niche of the space starts to grow, what are the other use cases that are exciting to be built? Um, particularly consumer is one that I'm excited about. It's a it's an area that crypto has struggled with historically. I don't think there are really any noninancial consumer applications that are at scale yet in crypto. And I think AI can help bring that into the crypto world in some some capacity.

So one way I thanks for that guys I think about it is that um we if we're gonna really embrace AI in a very intimate way personally and for our businesses um you know we need to have some level of privacy baked into to what happens there. Um, and um, I think that may not be strictly speaking kind of a web exclusively a web 3 or digital asset, you know, capability, but it's one that, you know, through different forms of encryption, you know, ZK and homorphic and trusted execution environments. It's one that is harnessed readily within this ecosystem. So, I think that is key. I think interestingly on on this point of verifiability and truth um what your your the kind of problem statement you outlined um there Jansen made me think about how we end up once we scale applications in the human world with lots of lawyers.

we end up with lots of lawyers to resolve all of these questions of, you know, information loss and breach of contract and dispute resolution, whatever. You know, we may be able to have some automated third party arbitrations through agents, but you know, we're not going to be able to scale this world of autonomous agents if it requires lawyers uh to be intervening all the time. Um and so that the verifiability of the truth and the performance of a contract and all these things it has to has to happen on chain autonomously for this to have any chance of uh of scaling. Um so I think and then the trustless coordination is kind of like in a sense a byproduct of that at the agent level. I think there's a trustful coordination thing at the at the data level.

How do we get people to contribute their data through some incentive? you know, uh, you know, Charlie Munger, I don't think he was a big fan of, um, of digital assets and Bitcoin, but said something about, you know, show me the incentives and I'll show you the outcome or something, you know, whatever. We can provide incentives to the data. Data is what we, you know, we absolutely need uh, in order to build better AI models. So, I believe that's important.

Um, and then we've spoken about getting the capital coordinated as well. So, okay, I think we've covered a few different interesting topics about, you know, your projects, where you've got to today. We've got like a couple of minutes left. So, just to kind of to wrap up, maybe uh again, where maybe you want to restate where you've got to because numbers always great. So, you know, audience going to get a good idea of the state of play.

Where are you going and what's the key area you're focusing on to to unpack and like in the spirit of web 3, how can people in the audience here or at home participate in that? you know, let's get people involved. So, do you want to start? Oh, actually, Nero, do you want to start descent your case? We'll pass it back.

Yeah, good. Sure. Sure.

I think the big trend this next decade as we start to see these AI models continue to improve and and get to, you know, AGI or AGI like uh intelligence is um we're going to see trillions of dollars managed by agents. Um, we're we're additionally going to see, you know, every component of life in the world have some degree of of AI attached to it. Uh, and so what we're interested in that ritual is um new behaviors, new things that will, you know, consumers will be doing in five or 10 years from now or hopefully earlier. Um, and so we're very excited to see the, you know, the kind of the early seedlings of this space. it's one or two years old.

You know, the I mean AI itself, this this LM wave is only two, two, three years old. Yeah, crypto is maybe a year younger than that. Uh so I'm just very excited for these next five or six years. Um and that's why we're building what we're building at at Ritual. You know, every year at a at token, we will come up with a a term a new term that we believe will be our compass and will become the direction of also the industry.

So last year literally um we we came up with the term agent five because we believe that agents can be productive assets and that you can tokenize them um and allow people to participate in the economic upside. Now we've realized that our vision has grown this year and um now we are focusing a lot more on can we create that that stripe or the swift for the agent economy and uh the the new term that uh we are using as our compass today is something called small a capital GDP agentic GDP and uh that is a northstar of us how do we get as many agents to be productive in value. How do we enable them to trade and to have commerce between each other to produce more economic value? And uh that that is the goal and our target is to have agents paying agents $100 million by end of this year. And what we are doing, we realize that for this to happen, we can't just let the agents do it by themselves.

We need more use cases that people like you and I will use on a daily basis that will help push this adoption. So my call to action for you guys, it's very simple. We've actually made something simple enough for you as an audience to actually understand and use to understand why is there value behind agents coordinating at scale. If you guys open your Twitter today and you tag at Butler agent, Butler B U T L E R agent, it is pretty much an agent that can help you access this network of other agents. We today you can even do it.

If you see someone talk about a token, you can just tag this butler and say, "Hey, run me a due diligence on this token." This butler will go and find other agents who can help you do this diligence for you. You can say, "Hey, I want to open a 3x leverage long on this token." Tag the same butler in the same Twitter trade. He will find another agent to execute that trade for you.

After being a backholder, you might say, "Hey, I want to create a meme music video uh to shield this token." You tag the same butler. He will find a music uh video agent to generate that for you. And then uh maybe you might say that h uh someone is fing my my bags. Uh let me You type on Twitter, [ __ ] you at baby Ansom, right?

And say that, hey, put money where your mouth is. Here's $10,000. Open the prediction market for me. You tech Butler agent. The Butler agent will go and find a prediction market agent to build up that liquidity for you.

So, we've done that is live today. There'll be more and more services coming out. So, and we're doing one more thing on top of that.

We I think we may we may need to move to David soon just because we're

Oh, that's all right. here. Um, I'm glad we brought prediction markets into as well because we're we're we're covering all the bases. So, so actually one of our portfolio companies near is looking to use USD.AI to bootstrap a network for its DCML technology.

So, David,

yeah, I guess my opinion might be a little contrary to what you guys are saying. Um, I don't think we should build in hypotheticals of what should be, what could be, what will be, but build in the now. Like famous quote from Jeff Bezos, internet AI like really it's cool that we have sovereign agents but like who's the user? Talk to the user. Give me a use case.

Now um we have 15 we have 15 loans signed. We have another 20 that's in negotiation. These are customers that want to use the product. They're going to provide the yield. They have a specific use case and has cash flows.

Open AI worked because they're the first time they had an interface that users could use. people were building GPT3. Like that's cool, but how does a user interact with it? Where does the cash flows come from? How do you pay the debt?

How do you actually make money move um where people actually want to use it? So for for me, just whenever you look at something within this space, just ask who is the user? What are they doing? Do they really need to use it? And are they willing to spend the money?

Because there is money to be made, but there has to be a use case. So for us, we have a very specific use case which is a lot of people need GPUs and they can't get a loan. Um that's why we built our product. It's very hard to raise money to do this because they structure incorrectly. So yeah, that's my conclusion.

So look, it's great. We've had representation from this entire stack that we need to deliver on on AI. Uh huge amount of resources to be delivered and and coordinated to make it successful. The most essential resource remains you folks, the users. So, uh, the time's up, the bell's going, but follow everybody on Twitter, send them DMs, get involved in the community, um, and you can be part of making this successful as well because we, we don't want Web free just to be about moving money around.

We want to have use cases that generate, you know, subscriptions and habits that real, you know, utility, to use that phrase, and that will drive the justification for this big buildout, which I have no doubt will be the case. Thank you very much to the panelists to the audience and we'll see you next year on a bigger stage.