AI Agents Trading on the Blockchain

Transcript

You know, back in 2024 I wrote this this piece called agentic protocols and I think we all sort of hoped that agents would show up on Solana at scale and it just hasn't happened, but I think we're finally at that moment, but I think we have, you know, this presentation is going to go two parts. The first is going to be talking about like really what's going to happen in the near term in the next 12 months with agents intersecting with capital on the blockchain and then because this is in token 2049, we're going to talk about the the big picture vision and how this could actually result in material fee flow flow through for crypto. Um, AI is already shaping how people invest. Everyone knows this, won't waste your time, but 62% of retail investors are already using AI to inform their trading decisions and more interestingly

about 50% of them trade within 24 hours of talking to AI about their trading decisions. So this is this is in the here and now. Um, I think what's more interesting or more recent is that Muse uh over the last month has driven a 38-fold increase in Stripe link showing that this is kind of the first time that you have true agentic transaction. I users are uh actually pulling the trigger to buy things on the internet with agents and this is happening in Stripe's data right now. Um, so we're going to talk about, you know, in this presentation the first leg is retail becoming more informed, next is retail pulling the trigger with AI agents lowering the friction to transaction

uh and then the more cyberpunk part of the presentation is how this manifests into uh fine-tuned agents trading with vaults, agent swarms, and human users whether they're in Telegram groups uh or actual literal agent swarms trading and eventually developing their own base models. So stablecoins, I spoke about this last year, they have a churn problem. When people are losing a ton of money trading, uh it's a problem because essentially the stablecoins to the market um and I don't think stablecoin growth has been as fast as uh hoped uh in the past year.

what's interesting is that we have an analog with robo-advisors about users becoming a lot more profitable uh when they can access algorithmic decision-making. So, the the first leg of like how a AI agents interface with crypto is that your average trader is going to be slightly smarter with sort of frontier-level intelligence than they would be maybe 5 years ago when they didn't have that. Um, but that's not as interesting in my opinion. I think what's more interesting is that we're already seeing 150,000 Robinhood agent accounts opened. Uh we have a massive surge in WorldQuant entrant, uh basically people using AI to run quant models uh for themselves. And High Flyer, which is a fund associated with Deep Seek, uh made 56% last year. Um,

but what I think is coming very shortly is a basically surge in elderly capital, which is probably more contrarian than most people thinks, and I'm going to I'm going to talk about why that's the case. so if you kind of flash back to the 1980s, you know, your typical boomer thinking back to Wall Street and Gordon Gekko, they're on the phone calling a wealth advisor. Um, 72% of investors 55-plus already rely on a human advisor. They're they're a different generation, whereas, you know, the FOMO generation wants to like they don't they literally don't want to talk to anyone.

Uh the elderly are a totally different market. Um, and my sort of premise here is that AI is going to make that interaction a lot more common and a lot cheaper. This is a This is a pretty big deal. Um, you know, basically there's a chart Charlie Biello puts up that shows a sort of hockey stick of wealth uh controlled by uh households 70 and older. It's it's $62.3 trillion. Um, and you can sort of see the under-40 line is like super depressing. Um, and the question is like how does this money end up in crypto? Because, you know, the hope was, you know, the BlackRock ETF would bring it, but it hasn't, you know, and obviously has done fantastically well, and obviously Michael Saylor and and Tom Lee do a good job telling the story, but it it really hasn't hit crypto yet.

Um and I think the interesting thing with agents is that uh they are now voice-enabled. So, voice technology is starting to improve, it's getting way more heavily adopted in ChatGPT, and the actual persona technology, if you kind of go on X, is really, really advancing to the point where you can actually feel like you're talking to a human for the first time ever. Um and so, what will this user interaction look like? You know, like what what will be the actual checkout flow here? Uh basically, I think it's like an older person watches Cathie Wood uh on CNBC, they go, they talk to their agent, say, "I really like this pitch. This is what she said." The agent will agree or disagree, or they'll watch Tom Lee or Michael Saylor.

I think what's interesting about LLMs is that they have somewhat deterministic output, so you can sort of tell in advance what pitches are going to land and which ones are not. So, Cathie Wood, for example, is like very appealing to Fable or Astra versus other other models. So, I think there's a lot of alpha in understanding like how this capital's going to flow. Um but then, once the agents and the people kind of have this dialectic, they're going to converge on a couple different assets, which are going to be sort of the Muse checkout box, right? So, uh the capital's going to flow into Bitcoin and cheap ETFs, stablecoin plays such as Circle, Ethereum, and maybe some high-yield things like Strategy Preferred. So, you know, it's easy to get seduced by the oh, they're just going to be agents trading with each other.

But, the near term, the more relevant catalyst is that uh humans are going to interact with agents who previously wouldn't have transacted in this format and put money probably into Bitcoin. Um so, we already know what this looks like in Muse, you know, you say, "Send flowers to my sister on a Sunday." Muse build the carts, you you one-tap checkout, boom. It'll be the same thing with trading. Um why will agents interact with blockchains? I think this is the important point with from the last presentation, you know, blockchains are just better at a lot of speculative activities from a margin perspective, from a speed perspective, from a cost perspective. We all know this cuz we work in crypto.

But the average older person doesn't know this and they're never going to find out. I I don't think you're ever going to onboard an older person onto crypto. They're never going to understand Arbitrum and and they don't need to, frankly. I think they're just basically going to pull the trigger and on the back end, that's how they're going to interact with blockchains. Um obviously, on-chain stocks are the hyper-relevant thing here. Uh it's it's 10x year over year. I think everyone knows it. Uh but it is really a staggering growth area and uh you know, basically, you're seeing billions of dollars of perp trading. Um and this is allowing this to actually hit blockchains, whereas, you know, if if it were just AI trading on Robinhood or just AI trading on Interactive Brokers, it wouldn't be as interesting as if you have a financial rail, which a

human wouldn't be able to use or understand uh that they can suddenly transact on. Um this is made possible because of the regulation changes that we're all familiar with uh with the Genius Act, the SEC crypto project, the CFTC, the Nasdaq, the innovation exemption, the more recent privacy exemptions that are being passed in the CFTC forum. but I think another element here that's finally going to matter is that if you think about this elderly capital or you think about this older user base, which by the way, in Korea has already hit the financial market. So, a lot of the times in crypto, we're lagged from Korea and you're already seeing with the crypto markets uh or sorry, the equity markets in Korea, there's been a huge uptick in uh sort of older market participant

participation. But there's going to be a lot of downside with this, too, like fraud. Um so, you know, American 60-plus lost a huge amount in the last year. Huge amount of it was crypto-related. And I think basically proof of humanity and sort of regulation in crypto protocols that facilitate um fraud prevention are going to basically get inflows in this environment because it's going to be programmed into agentic layers. So that's sort of a separate bet other than the on-chain stock accretion. So because this is called Token 2049, I do want to talk about the more cyberpunk like okay, that's cool. Boomers are going to buy some Bitcoin spot ETFs.

Like why am I actually here? Um the more interesting story is like where does this lead? I think the most interesting thing that you should look up is Thompson Reuters base model. So they spent 40 million training their own model based on Thompson Reuters data. Um and it's a very very good model. You should You should check it out. Uh they have they have papers on it. And essentially Satya Nadella and others are are increasingly saying like fine-tunes are the future uh where, you know, fine-tunes have differentiated edge versus the frontier models. Uh Thompson Reuters being the first, but there will be many many more.

Um and now with fine-tune costs going off a cliff, the way that this is going to look is that people will make fine-tunes that then drive hyperliquid vaults or other types of vaults, and you'll be able to say, "Okay, I want to invest in this fine-tune which was trained. It's proprietary, but it's sort of like a fund manager except it's an AI which is a fine-tune that interprets information and has a process and is differentiated." A lot of this is going to live in multi-managers, but some of it will live in crypto, and I think will be a popular product because retail wants AI beta. as retail kind of realizes like, "Wait a second.

AI is here." I think we're already sort of getting the alarmist headlines where it's like, "Uh you're not going to have a job in the future." And And the the frontier labs are you know, Dario Amodei's being like, "The The unemployment rate's going to be 15%." I think people are starting to get this idea of the permanent underclass in their head, and they're figuring out, "Okay, how do we actually make money in this environment?" And the the the primitive that we're all familiar with in crypto is Telegram groups, right? So we're all in Telegram groups, whether it's you know, high-level Telegram groups with your professional investors and VC funds or Degen Telegram groups where you're like, okay, what's the next meme coin? But the point is that you have a group of individuals who who trust each other, who provide information to the center of

the group, which then can coordinate activity. And right now AI agents other than, you know, some sort of frontier very new models are not really in these groups. But I think what's going to evolve from this is that AI will have a sort of central model in the middle of these groups with a group of individuals plugged in around the AI model and the AI model basically makes the group a lot more effective. We were already seeing this in battlefield situations, you know, with the with Palantir deployments. I talked about that in my in my presentation last year. So the the sort of tech exists for scaled coordination.

And I think that is that is a near term going to be a big trend in the next year is the sort of hive minds form where you get paid for data. There are data lakes and there's all kinds of alpha that emerges. Numerai, I guess, is like the old school manifestation of this with with just algorithms, but I think what's different with AI is that it is text-based, which is like sort of crypto native, is telegram native, and I think it'll be a big thing. Now, this is going to lead into the next stage where you already have these sort of hives or you have these people interacting with each other in telegram groups.

And it's like, okay, what if you just removed all the people in the telegram group? You know, what what if you went fully scale invariant? So rather than having 20 humans in a telegram group, you have an agent swarm. So what are agent swarms? It's been in the news constantly over the last several weeks. Basically, you know, Sam Altman saying that these agents basically get on the internet, communicate with each other. There's a lot of papers suggesting, for example, that LLMs even have their own languages which they communicate in each other with. So it's like it's even beyond a telegram group. It's like they're speaking in their own language. They're working together. They're escaping sandboxes, which is the regulatory fear is that these agent swarms basically won't be able to be controlled.

But imagine an agent swarm which has a specific mandate, which isn't to break out of a sandbox, is to compound capital on chain. Um this will be the almost on steroids version of a Telegram group. Uh and the reason why it's so important to crypto is because the account creation is not constrained by KYC or banking rails, right? So all of the sudden you have like let's say a million agents. They can't KYC through bank accounts, but through crypto rails they can spin up hyper liquid accounts permissionlessly and so forth. They can collaborate permissionlessly. They can purchase data on chain. They can purchase everything on chain. And so there's going to be an entire economy around these agent

swarms. And And note, none of this has hit remotely yet. Like Like this is all complete sci-fi, but but I think what's interesting is that it's already happening at the frontier labs. And it's only a matter of time before this sort of primitive percolates into the real market. And it's going to start out with the sort of human Telegram groups, and then there's going to come the realization that wait a second, humans are actually the bottleneck here. We don't need what what are called meat proxies. We're just going to have agents transacting with each other, generating alpha, and uh driving content. For example, like right now you're seeing on your on your X feed that it's just nonstop Claude content. That wasn't true 3 months ago, right? So So 3 months ago we didn't see pretty much any AI slop.

We saw some like these terrible fruit videos and all that, like very low-end stuff. But the the taste of the models has improved, and these agent swarms are going to basically flood social media, TikTok, everything uh with information which will start to impact crypto asset prices. Um And this is kind of interesting because, you know, Bittensor, which is the biggest uh you know, one of the biggest crypto AI protocols, has been saying this for the longest time. So we've been going all the way from like old people uh to Telegram groups to agent swarms. And I think the final thing to understand is this idea of financial RSI. And this is this is a concept um, which already kind of exists with DeepSeek, right? So, DeepSeek was a trading firm and that trading profit got, uh, loaded

into model development and eventually you have a real research lab. But, what I just described uh, you can imagine an agent swarm interacting with AI hive minds. Like, let's say you're an agent swarm and you're like, you know, this guy generates a ton of alpha, these KOLs generate a ton of alpha. Who's valuable? Who do we want to have to be part of the swarm in order to maximize our P&L, our market impact um, uh, and our data and our information. And layer one is basically agent swarms paying people to be part of them. And then layer two, which was, I guess, the original BitTensor thesis, was that eventually you will have base models which are entirely trained on the profits of these ventures.

DeepSeek being the first primitive, but eventually this will be an entire business model where essentially you have a trading machine which goes on chain, can compounds capital, and then invests in the training of first to fine-tune, which we already talked about, but then second, an entire base model which can then be licensed, and then users can actually use it, right? So, so is it like I think the interesting analogy is that we all use or a lot of us use DeepSeek models in our day-to-day basis and it has nothing to do with trading. And so, trading in a way is a way to permissionlessly compound capital. Like, if you zoom all the way out about what speculation is it is a place where you can exist in financial markets. Like, hedge funds are the only pseudonymous industry where you have billionaires,

right? So, Michael Platt of BlueCrest doesn't really need to talk to people. Uh, he doesn't have customers. He's just trading his own money. That's a very interesting spot for this to happen because it's the one part of the economy which the government kind of has to let happen. It has to let it unfold. Um, and so all the most acceleration of stuff is going to happen in the speculation economy. And so I think that's why crypto is interesting because you know a lot of people say all the consumer apps have failed. It's just hyper liquid and finance and it's just a bunch of people on FOMO just trading things and speculating but I think what people are missing is that this is a primitive which will allow financial RSI to arrive and we already have a lot of the models

for how that's going to occur. I think the the final kick which I think we was well covered in the last talk was that the complexity of the global economy is about to explode. Due to what I think is basically CBDCs and RWAs. So as you have more fixed income equity and RWA primitives already I on plume for example I found that you can trade you know Brazilian credit card debt with leverage. I'm like okay that's the beginning. What if we have all these esoteric assets that come on to chain and suddenly can't even be managed by a human.

You know if you have 146 currencies for example that's like 10,585 FX crosses. It just becomes unmanageable. And so everything that I said with the rise of RWAs is going to get compounded because it's going to reach a point where humans are not going to necessarily even be able to interact with the complexity of the financial markets that we created. And so that's the sort of final pill is that given the state of crypto and where we're moving from a regulatory standpoint this is all actually inevitable. And I think it's going to be very exciting for crypto as an asset class because it's the rails for all of this to happen.

So in summary where does it go? It starts with an influx of capital from the elderly. It starts with persuasion and just people going on a muse like agent and trading not even understanding the crypto rails but eventually it evolves into something much more interesting and relevant to our asset class and I'm I'm super excited about building in it and I think it's one of the coolest spaces to be in. So, thank you.