Ryan Sean Adams (6:03)

This is uninterpretable to me.

Haseeb Qureshi (6:05)

It's literally non deterministic. Right. Whereas a smart contract is actually machine code. It is actually compiled into the EVM bytecode and you can analyze it step by step. This is exactly what will happen in a hundred percent of scenarios. Now, as human beings, we might know that that's true, but we don't feel that that's true. It doesn't feel intuitively true to us that the legal contract is actually less predictable than the smart contract. To us, it's the other way around. Even though there's all this randomness in the legal contract. We find the legal contract much easier to predict what's going to happen than the smart contract. And so my claim is that that's because of our bounded rationality. It's because of our inability to process code as effectively as an AI agent would. But for an AI agent, all the stuff that we were saying about how smart contracts are a better way to create enforcement and property rights, it's actually true for AI agents. And my claim is that that's going to change a lot of the way in which the original promise of crypto gets interpreted. It's not going to be humans taking advantage of it. It's going to be AI agents in concert with humans taking advantage of it on behalf of the human that it's acting on.

Ryan Sean Adams (7:11)

Acting for now, Haseeb, I was at ETH Denver with you not, not terribly long ago, and I had to download Metamask to go like, sign into ETH Denver, right? It's like, what year is it downloading MetaMask@ ETH Denver? And I was actually pretty pleasantly surprised with the ux improvements of MetaMask. And I think that is emblematic of the industry as a whole. We have improved our human UX over the years. Maybe not as fast or as much as we would have enjoyed, but like, nonetheless, it is, it is happening. And I, and I think something that you're, what you're saying is a little bit more fundamental than that of that, you know, is AI isn't just going to gloss over some of the hard gaps of crypto UX for humans. When I open up my ledger and I see the blind signing thing, you know, an AI can go and interpret the code and tell me, you know, thumbs up, thumbs down, and maybe that helps crypto ux and maybe that's true. But I think what you are getting at is something a little bit more fundamental about the nature of blockchains and what they are optimized for beyond just smoothing over crypto UX for humans. I think you're going deeper than that. And you're saying this is technology that's not for us, fundamentally, critically. Is that what you're saying?

Haseeb Qureshi (8:25)

So obviously it's ultimately for humans in the sense that the human is the end recipient of value from what's happening on the blockchain. But the idea that as a human being, the right way that you are going to interact with the blockchain is by you moving your mouse, clicking on the Metamask extension, typing in your little password you know, going in there and like clicking the buttons and manually approving the gas. That stuff is obviously so alien to human beings. And the way in which we think about money and finance and accomplishing our financial goals, it's, it's a little bit like, you know, imagine that with the banking system, human beings had to write their Swift codes. Swift was not made for humans. Swift was made for an interbank communication protocol. It's not designed for you. Now if you had to interact with it, I guess you would, you know, if you had no other choice. But it's clearly not the affordance that most human beings are looking for in their own intuitions about how to use money. So my claim is that, okay, we, we live today in the world where this stuff is totally, this stuff is totally disintermediated. The human is interacting directly with the machine. And this is bad. It's bad in the same way that we think today about cars, right? In, in, in 10 years we're going to look back with absolute horror that we ever thought it was a good idea to have apes, you know, using their hands to manipulate these two ton death machines. Know, driving down the highway while drunk, half asleep, you know, you know, being ill, whatever, whatever state of mind you're in, yeah, you're probably fine.

Ryan Sean Adams (9:53)

Appear barbaric, right?

Haseeb Qureshi (9:55)

It's gonna, it's gonna seem absurd that we ever thought like, that was cool, that was just fine. And like, yeah, people die on the road all the time. It's like one of the most dangerous things we do is literally driving. But you know, what else are we gonna do? Like, why, yeah, you can't take away a man's car. We're gonna look back on that as being like insane. You know, there's gonna be a day when human driving is banned and, or only allowed in certain areas that are far away from other humans, right? That we're going to get there with respect to the way that our norms evolve. I think we're going to land at the same place with crypto. I think we're going to look back with horror at the idea that human beings were manually, you know, blind signing transactions and like eyeballing addresses and being like, yeah, this zero X. I think that's the same as the thing I saw. So yeah, I'll send it there. That's right, that, yeah, exactly. Like that stuff is going to go away. The idea that you're like manually looking at a URL and being like, I don't think this is a phishing domain. I looked at it and I don't think this is a phishing domain that is going to go away. It's going to be like, what on earth? Why would you think that that's a good way for us to keep people safe and keep their money safe? Right? Like, of course human beings mess up. Of course they're going to sometimes fail. You're going to be tired, you're going to be scared, you're going to be whatever. You don't have the energy to check it three times over, to go check the DNS, to go check the Twitter and see, hey, has this protocol recently been hacked? Right? Like, you know, when you see these announcements, like, sometimes you'll go on Twitter and you'll be like, oh, it's trending that like, such and such protocol got hacked. Okay. I sometimes wonder, what if I was a user of that protocol? I'm usually not. What if I was a user of that protocol and I just didn't check Twitter? Like, are people, like, have the Twitter notification for every protocol they use? Going straight to their phone is a push notification. Like, we don't even have an alert mechanism for, like, when a protocol gets hacked, you're supposed to just like, check the Twitter and just happen to see it before you interact with this protocol. So the reality is, like, there's so many things that you could be doing, but we're just not. We're human. We're human. And this whole idea that, like, oh, we're human, of course we're going to make mistakes. The AI agent never gets tired, never gets lazy, never skips a step, never, never doesn't follow its instructions. And so you can imagine a world where, look, the interface, I think, for these things is going to look very different in a world that's totally intermediated by AI. So let's say you have your AI agent and you tell your AI agent, hey, I think interest rates are going to go up. I, I, I think that we should be, you know, moving into safer defi. So let's move me out of, you know, whatever it is. I'm in, I'm in, you know, riskier defi. I've looped something instead, you know, move me into just aave. Oh, I just want to stay safe. Your AI agent will just go do that. It will just go do it for you. And if you're particularly, you know, in a YOLO mood, you will manually approve the transactions or, sorry, you, it'll just do it for you. But if you're like, hey, you know, I want to make sure I know exactly what I'm doing. It will batch everything and it will say, look, here's a series of steps I'm going to do. Please approve this plan, right? And probably in the near future it's going to be approve this plan in the, in the farther future it's going to be, it just does it for you because you're not adding anything to the equation, you know. But at least for the foreseeable future, it's going to be okay. I present you this plan. Do you approve the plan? Cool. I'm going to go do it. In this world, one can imagine so one, you're no longer clicking on AAVE to. You're not even looking at the AAVE logo. You know, you're not looking at any of their marketing and in fact you're, you're maybe not even telling it put into aave, you're just telling not even know what into some low risk defi. Right, exactly. You don't know what AAVE is. You just say, hey, lower my risk profile and it's going to say cool. Here I found, actually I shopped around, I looked at 14 different protocols and I looked at all the different, you know, promotions or, or TVL deals that are going on there and I found one that actually is the best. And I'm going to go use this one, right? Like what happens to marketing, what happens to network effects? What happens to the ways in which we think about, you know, this was a lot of what the, the viral Citrini article was talking about. When, when AI automates discovery, how do companies compete with each other? How do protocols compete with each other? Because so many of their business models are dependent or have this baked in assumption of human friction, this idea that, well, a human's only going to look at three or five or they're only going to look at the biggest one. And so all you really got to do is get the biggest person and then you. They're sticky. They're not going to want to look around any further. But an AI agent doesn't necessarily think that way. So I think a lot. If you assume that the story is true, a lot is going to change about the way that protocols work, about the way the protocols compete. And who's going to benefit from all of this? The answer largely is the consumer benefits, right? It's a big consumer surplus. If all of a sudden this efficiency is getting captured by the user, that's good for users, it's good for crypto. It ultimately means that like crypto users are going to benefit more and more and more from all the stuff that exists on chain, you know. But it's very counterintuitive how it's going to work and it's not going happen all at once. It's going to be a gradual process as these models get better and better,

Ryan Sean Adams (14:48)

Euphoria brings one tap trading to the palm of your hand. Built on Mega Ether, Euphoria takes real time price charts and projects it over a grid of squares. You tap the squares that you think the price will enter in just 5 to 30 seconds in the future. If the price goes into that quadrant, you can pocket anywhere between 2 and 100x your trade. No other application helps you trade faster and with more leverage on market driving events like FOMC meetings, presidential speeches or global macro events. Thanks to Mega ETH's real time blockchain, Euphoria is the way to get real time price interactions with the market. On Euphoria, you'll be able to compete with friends using Euphoria's real time social trading experience, allowing you to go head to head with your friends. A great party trick. If you project the app on a tv, it'll be like the Mario Party of derivatives to trade on Euphoria, people can deposit stablecoins from any chain or do direct fiat transfers and everything gets converted into Mega Eth's native stablecoin USDM in the background. Check it out at Euphoria Finance and download the app or find it in Telegram. As a mini app, Galaxy operates where digital assets and next generation infrastructure come together, serving institutions end to end. On the market side, Galaxy is a leading institutional platform providing access to spot derivatives, structured products, defi lending, investment banking and financing. With more than 1600 trading counterparties, Galaxy helps institutions navigate every phase of the market cycle. The platform also supports long term allocators through actively managed strategies and institutional grade staking and blockchain infrastructure. That scale is Galaxy has over $12 billion in assets on the platform and averaged a $1.8 billion loan book in late 2025, reflecting deep trust across the ecosystem. Beyond digital assets, Galaxy is also building infrastructure for an AI powered future. Its Helios Data center campus is purpose built for AI and high performance computing with more than 1.6 gigawatts of approved power capacity, making it one of the largest sites of its kind. From global markets to AI ready data centers, Galaxy is serving the digital asset ecosystem end to end. Explore galaxyalaxy.com bankless or click the link

Haseeb Qureshi (16:46)

in the show notes.

Ryan Sean Adams (16:46)

What if you could trade gold forex and global markets with the same tools and speed that you use for crypto. That's exactly what Bitget Tradfi unlocks. After strong beta demand, including over $100 million in single day gold trading volume, Bitget TradFi is now live for all users. Inside of your existing Bitget account, you can trade 79 instruments across forex, precious metals, indices and commodities all settled directly in usdt. No platform switching and no fiat conversions. This is Bitget's universal exchange vision in action. Crypto and traditional finance side by side. You get deep liquidity, low slippage and leverage up to 500x letting you apply crypto strategies to macro markets new to tradfi. Start with gold. The Gold USD pair is liquid macro driven and a familiar natural bridge between crypto and traditional markets. Try trading gold on bitget now@bitget.com click the link in the show notes for more information. This is not financial advice, Haseeb.

David Hoffman (17:39)

I'm wondering if we can develop a stronger intuition for this. Because I think if you're right, if crypto is in for humans and crypto is for AI agents, then it's incredibly important for us who are here in crypto to think about the world as an AI agent sees it. You know, there's this book called Seeing Like a State envisions how the nation state sort of sees a world and it's, it's hard to get into that mindset. We all think like humans, so we see user interface and we see crypto and we view it through the lens of a human. If we start to view things through the lens of an AI agent in terms of how they see the world, then we can start to forecast the future in crypto and honestly in the world around us that much more. It feels like that is a key skill for a builder and for a vc, for an investor in the space is to start learning to see like an AI agent. And I'm not sure that I or many listeners have developed an intuition for how an AI agent actually sees. I'm starting to pull out some clues here and this is why the Open Claw project was pretty important for me and why I love all of the experimentation going on in it is because you start to see for the first time the world as maybe an unfettered, unhobbled AI agent actually sees the world and the types of tools and the ways of interacting that they prefer and what their capabilities are and what they can and they cannot do. It's a powerful experiment when in particular when it's applied to Crypto. And so I listened to the Peter Steinberger, who's the kind of the founder of OpenClaw, I listened to his podcast with Lex. There's some clues dropped there. Like, one thing he said is, you know, openclaw really prefers command alignment. And when we can get openclaw a command line type of access to things and the raw data to things, rather than go through an API or some sort of MPC type protocol, like just the command line root access, like give me all the tweets, don't give me this sort of, you know, user interface on top of Twitter that my open Clawbot has to like scan and call. It's just quicker to just give me all the data.

Haseeb Qureshi (19:59)

Right?

David Hoffman (20:00)

And I was thinking about one thing that probably AI agents prefer is like the code, like the firmware level access and the control. We also had Austin Griffith on the podcast a couple weeks ago and he was messing with openclaw and applying this to the agent world. And he kept trying to get OpenClaw to do some crypto transactions for him. And you'd go, you'd, here's Metamask, openclaw. Go through these user interface and click the buttons, hit approve, hit confirm, hit

Ryan Sean Adams (20:29)

approve, hit confirm, hit the UI button.

David Hoffman (20:31)

What OpenClaw kept trying to do and revert reverted to is basically, hey, Austin, all this would be faster if I could just like take your seed phrase and extract your private keys. I don't need them somewhere on the machine. If I could store it on the machine, I could do all this shit and I can just write it in code and skip the user interface. The stupid clunky human user interface that I don't need.

Ryan Sean Adams (20:53)

That you need to be pretty and colorful and enjoyable to use.

David Hoffman (20:57)

Exactly. And so that's like another clue here. And I'm starting to try to, through these clues and samples, develop an intuition for how maybe an AI agent prefers to act in the world and what tool sets it prefers to use. Because then you can start thinking about, well, what type of a wallet does an AI agent prefer? If AI agents are going to be our next hundred million users, our next billion users in crypto, what sorts of things will they want to do and what sort of products are they, the users that we should start design, designing for in crypto, I still don't know how they see the world fully. I'm wondering if you have any intuition for this type of thing. So you said, like, when you send a wire, you're fine with it. When you send a crypto transaction, you're like, oh shit, what's the call data? Like, oh my God. Like this is immutable. It's a one time thing. An AI agent looks at that transaction and is like, I understand this completely. You know, and they would be in foreign territory when it comes to sending a wire. Anyway. All that to say, how do these things think?

Haseeb Qureshi (22:02)

No, it's a great question. So you, you point out something very deep, which is that the innovation that we've had in AI has all really been driven by large language models, right? And we call them large language models because they're trained on corpuses of text. Text, that's the keyword text. Okay. It's a large language model. Now we're trying to apply the massive innovations we've seen in large language models to other modalities. Right. We're trying to apply it now to, to images, to video, to robotics, to all these other things. And what you see is that in text is where we have the most runaway capabilities and everything else is kind of, you know, it's pretty good. It's, but it's no, it's nowhere near as, as robust and as powerful as in text when you are getting an AI agent to interact with a computer. So this is known as computer use. And all the different labs are trying to get their models to become better and better at computer use. The problem with computer use is that if you are trying to get a model to interact with clicking a button on a screen, literally what you're doing is you're taking a picture of a screen, you're tokenizing that, you're turning it into like these patches and you're trying to give the model some kind of deep representation of these patches. When it's been trained on text. Right, right. Like text is where we have billions and billions and billions. We have the entire corpus of human history that we fed into these models and we have, you know, like not that many pictures of computers. Like, I mean, obviously we were generating a lot more. And they're trying and trying and trying and this will get better because we have a lot of synthetic data to be able to continue generating and feeding these models. But the reality is that interfaces were created for humans, right? But these models grew up on text. They are born in text is the, is the soup that they swim in. So if it's in text, it's way easier for these models to learn it and to get good at it because it's just a much more compressed representation. Okay, so think about crypto. Ironically, the bad UX era of Crypto is when it was all in the terminal, right? That's where everything started on your Bitcoin D, like when you're originally sending Ethereum transactions. That shit was all in the terminal, right? If you wanted to use a ui, a gui, right? The first one was Mist. Mist was a piece of shit. It was so bad.

Ryan Sean Adams (24:10)

It was dog shit.

Haseeb Qureshi (24:11)

It was so bad, it was so

Ryan Sean Adams (24:13)

bad that consumed my entire computer.

Haseeb Qureshi (24:15)

Yes, it's so horrible and like that. But, but the idea was that before that, like when Ethereum first started working, when they were like, oh, Ethereum's working. What they meant was that it worked in a command line. It was you, you specify you know exactly what the number is of way that you are sending in a number in your command line, right? That's how you sent Eth back in the beginning, before, you know, all these UIs came in and made everything much easier. So the point that I'm making is that crypto from the beginning was designed in a form factor that's perfect for AIs. Right? It was already there. We got it, we one shotted it.

David Hoffman (24:53)

Our bad UX is their good UX.

Haseeb Qureshi (24:56)

Exactly, exactly. And for what? What, what else is a good UX of like, oh yeah, privy and like, you know, Google OAuth wallets, like these bottles cannot oauth. They don't know how to do any of that stuff, right? Like it's actually much worse and harder for them. And you don't want it to do that. You do not want this model to have your Google oauth token because then it can just literally go into your Google and wreak havoc, right? You don't want that. You want it to just have a small amount of crypto in a wallet that's self segregated and clear rules around what it can and cannot do, right? So ironically, crypto has always had the UX that is perfect for AI agents in terms of exactly what is easy for them to parse and interact with. And that's the sense in which I say crypto was designed for AIs. We're now trying to backport this AI first technology into becoming something civilized enough that humans can use it. Which means, oh, it's the privy model. It's the, oh, your money's sitting in Binance or Coinbase. It's extremely easy. We remove the foot guns, you know, we have two fa, we have all this stuff that makes it easier and easier for you not to lose your shirt or make a mistake or get hacked. But in principle, you know, in the Agent or if you're on a computer, you already know how to manage secrets, right? I mean, how do you think you use an API? Every API that you've ever used has some kind of API secret, some kind of API key that a model's whole job is to safeguard as best as it possibly can. So this idea that like, well, you know, human beings, we obviously screw up all the time, it's not safe for us to hold onto to money, and the agent knows how to manage secrets, you know, that's how it has its anthropic key, right? If that wasn't true, it leak its anthropic key and then, you know, boom, you'd have all your credits strained, which you know is happening sometimes, but usually not because you're leaking the key. So. So all that is to say, I think the answer is actually that crypto doesn't need to move very far for it to be in the right form factor for an AI to use it. The difference, the difference is that right now the problem is that agents have not been trained to use crypto, right? They've. All this rl, all the training that we've been doing inside these big labs is all around coding and mathematics and then of course, just general, you know, be a nice AI agent, you know, to give me some therapy, answer questions about medicine, whatever, this kind of stuff, this is what people have been reinforcement learning, training these models on. It was just last week that you saw at OpenAI the release of EVM Bench, which is a EVM based cybersecurity benchmark that, you know, kind of shows that all of a sudden, hey, looks like they're starting to train on this stuff. Looks like they're starting to actually do some reinforcement learning to allow these things to operate in a blockchain environment right now, EVM based. And the same thing was true at Anthropic. Anthropic. They released a paper, I think, last year of them showing how well models were able to inside of a harness that they designed, how well models were able to do cybersecurity attacks on EVM based smart contracts. Right? First time we've seen this. Now, that being said, you go look at the anthropic model cards, every anthropic model, they check to see how well can this thing do Bitcoin transactions. They've been checking this for a very, very long time inside all anthropic model cards. If you go in there, you can see it. Now, that doesn't mean they're training on these things. Most of the time they're not training on these things because they want to use these untrained endpoints as a sign of general intelligence, as a sign of is the model generalizing? Are they doing these tasks that we don't care about right now? They don't care about these tasks. Right. That's why it's in there. Once they start caring about these tasks and they decide, you know what, I want to win crypto, because I think crypto is going to be the future of a lot of payments. And it's actually really important that all that volume run through my AI. I don't want it to go to OpenAI, I don't want it to go To Grok, I want it to go to me. That's when you're going to see the stuff really start to take off.

Ryan Sean Adams (28:32)

Interesting. I have not heard of the fulcrum articulated before that anthropic OpenAI. Google, they need to point their attention, their focus to crypto in order to train their LLMs on crypto. And so right now what you're saying is like crypto as an industry is relatively unexplored as compared to other sectors where AI is trained on. Yeah, tell me why is that the case? And then when do you think that's going to happen?

Haseeb Qureshi (29:01)

So to be clear, like this is true for anything that has not been optimized. Okay, so look, for example, at chess, if you try to get Opus 4.6 to play chess, it's mediocre.

Ryan Sean Adams (29:12)

I actually tried to do this. I have taken screenshots of my chessboard and I've put it into ChatGPT and being like, analyze this position and it fumbles everything. It fumbles the annotation, it fumbles where the pieces are. It's like, and I think it get direct, it gets directional strategy. Okay, but in specifics, terrible.

Haseeb Qureshi (29:31)

That's right. That's right. Now the reason why it's so bad at chess is because they didn't try to train it at chess. They don't care. This is just not important. Right. If you try to train it at chess, it will become good at chess. Just like that. Like, it's obviously extremely easy at this point. We know how to do it to create neural net based chess engines that are better than any human being who's ever existed. Right. We've had those for a decade now. So it's simply a function of. They just didn't bother because who cares? We already have chess engines that are way cheaper to use than open AI. So the answer is that they just didn't point their lasers at this because it's not economically valuable to do. So when they do, you are going to know about it because they're going to start, you know, yelling it from the rooftops. Now, the reason, I think a big part of the reason why, why have they not done it so far? Because the thing is getting it, you know, getting blockchain based tasks is not that hard. It's all software. It's easy to create an EVM simulator to simulate the state of the blockchain, to test whether or not you accomplish this task of, you know, did you recursively borrow this thing or did you sell this thing for that thing? It's actually very easy to come up with test suites that you can, that you can build for all this stuff. Right? So why haven't they done it? My claim is that the reason why they haven't done it is that one crypto's kind of cringe. That's. You gotta admit, that's part of the reason why they're not doing this right now. But the second reason, the second reason is liability, okay? Liability is that if they broadcast that we are training our models to do crypto stuff for you, I guarantee you somebody is going to have a fuck up. They're going to have a gigantic fuck up. It's going to become a huge story and they're going to blame Anthropic and it's going to. Or they're going to blame OpenAI, they're going to blame Grok, they're going to blame whoever it is who did this, who trained their model on this thing. And like, there is no caveat in Tor that will be loud enough for people to take it as not a reason to go dunk on these people. Right? Like very clearly.

David Hoffman (31:15)

Yeah. So imagine if Claude, like effed up your trade and lost $2 million or you know, sent 10k to a burner address accidentally or something like this. This is massive liability.

Haseeb Qureshi (31:28)

And it will 100% happen. It will 100% happen. Absolutely it will happen. And it doesn't even matter if you say azanthropic. Let's say you say you sign away, you wave away all rights. You sign this thing that says, I am a fucking idiot if I connect my crypto wallet to this thing. Doesn't matter. No matter how much people do it, there will be front page news stories, there will be viral threads, whatever it is, no matter how many people have good experiences, anybody who has a bad experience, it's going to go super Viral. And there's going to be, you know, people marching in the streets about how evil the AI companies are for losing our money or whatever it is. So the reality is like, the risk reward is just not there compared to something like, you know, coding or medicine, where it's like, look, it's one, it's a lot easier. Like, you know, if this thing gives you bad medical advice, it's like, well, we told you not to use it, you know, you're an idiot. But second, it's like it's never telling you you should inject yourself with, you know, this like thing that I found on this Chinese pharmacy. Right. Like, it's not doing that, but that's the equivalent of what you would do if you're managing someone's crypto wallet. You're like, okay, yeah, let me do that for you here. This is like the equivalent of Chinese peptides, but in Cryptoland is like, yeah, let's lever up on this thing and borrow.

David Hoffman (32:36)

I think that's why OpenClaw was so exciting for people in crypto is because it wasn't from the big frontier labs with all of the liabilities. An open source project power tool like use at your own risk was completely obvious that it was use at your own risk and people were running it on their own machines and there was no kind of third party to sue. And so it was able to take these risks. And I do want to ask you about that type of an experiment. There's this idea of like, what will the AI agent economy look like? The bot economy, if you will. And I still think that if these AI agents, because this gets into a question of timeline, let's say people are listening aseep and they buy what you're saying. They're like, yep, I totally see crypto, smart contracts, programmable blockchain, it's all very AI native. The ui, you know, like, works out, it's better, but still, when is this adoption going to happen? Is this going to get us out of the bear market in 2026 or is this going to take many years to play out? I think part of the answer to that question might hinge on who the actors are. So who are the actual agents? There's a difference between how most people interact with AI today, which is, hey, I do something for me, like, I am the boss, I'm your employer, I'm the person that you are serving, essentially, and you have to go do this for me. And I'm still going to prefer the output of a legal document versus a Smart contract. And so that's what I'll give you. And I'll still give you my credit card information somewhere that you store rather than a crypto address, because I'm a human and I still prefer all of my, you know, court meet space things and, you know, credit card points and all of that. So if it's human directed agent activity, agent economy, that's one thing. And that might slow down our progress. Actually, that, that might be where we are right now. However, if it is AI agent directed and the AI is almost at some level kind of sovereign and has its own preferences and has some sort of ability to go do it has a much larger, I guess, space and to. To do, accomplish a task however it wants in that world. You can start to see the AI agents preferring, okay, you gave me a task, I'm gonna go do it my way, and I'll go use X402 to go pay this AI agent in order to get this deliverable and then bring it back to the system. In other words, if AI agents have more autonomy than their preferences, which if you're right, those, those preferences are crypto, they will actually, you know, come into being. But that world kind of depends on how much autonomy we actually give the AI agents and if they're ready for that yet, Are they smart enough for that yet, Are they ready? So maybe this is a question to how you think the world is evolving. Like when people are excited about agenic AI, are they talking about agents that are working for humans? Or they are. Are they actually talking about agents with a lot of agency? Maybe they're still working for humans, but they do have the ability to do things their way. They have, you know, much more surface area if. When is that world coming to pass?

Haseeb Qureshi (35:52)

So it's okay. There's multiple ways to analyze this question. I think it's a great question. So let me, let me start from the top. It's important to recognize that right now it's something like 12% of humans who exist in the world today have used any AI products at all. Most humans in the world have still used zero AI products whatsoever. Okay, meaning like LLMs, chatbots, this kind of thing. Of those people who have used chatbots, only about 1% of them have ever paid any money. Okay, that means that 99% of people who've used chatbots that are on the free tier, and from their perspective, they don't see why they would get any value out of paying for this.

David Hoffman (36:28)

Insane. Because I know how much I spend multiple.

Haseeb Qureshi (36:32)

It's Important to understand that like you and I, we live in the future now. Everybody's coming, everybody's coming where we're sitting. Okay. The value of this level of intelligence is so great, but most people have not realized it yet. The diffusion of this technology is taking a lot longer than any of us anticipated. Okay, so that's the first thing to understand now of that 1% that's paying money for AI, right? If you're using OpenClaw, you're paying money. So you are a percentage of the 1% that's actually using this as an open claw. It might be going viral on Twitter, but vast majority of people have not even touched it. Okay. So this is the frontier of the frontier. So it's important to realize it's good

Ryan Sean Adams (37:09)

to not be the only niche industry out there.

Haseeb Qureshi (37:12)

Yeah, no, totally, totally. No, it's very true. Now, okay, with that framing, with that framing first in mind, okay. It's important to understand that as this technology develops, it's going to be a two track thing. When OpenAI acquired Peter Steinberger and acquired, quote unquote, acquired OpenClaw, what, what Sam Altman said was that this product is integral to what we see as the future of OpenAI's products. Okay, now what do they mean by that? They meant that obviously they love the idea that you are going to have a personal assistant that's going to be driven by AI. Now, the way that OpenAI does that is not going to be the same way openclaw does that. Why not? Open openclaw very clearly is a kind of, you know, yolo, let it rip dark forest. Like, who knows what it's going to do, just do it kind of product. It's the open source. You know, it's like, it's like the early automobiles that just like exploded in the street sometimes. You know, before seatbelts were invented, before any of this stuff was, there was no safety approval process for cars back in the, you know, back in the 30s. Okay. That's what open call is like. So if you imagine opening I OpenAI today does have a commerce flow. They connect with Shopify stores, they connect with Etsy stores. And if you say, hey, you know, I want to buy a rug, OpenAI might show you some Shopify stores and say, hey, do you want to buy this rug? Okay. And when you click, yeah, I want to buy the rug, you have to manually click the button that says, I approve buying this rug. And then it charges your credit card. Okay? Now Open Claw lives in a totally different path. Oh, the open Claw path is you give me money and I will do

Ryan Sean Adams (38:38)

shit whether your rug shows up at your house.

Haseeb Qureshi (38:40)

Exactly. A rug. A rug may or may not show up at your house, okay? And like, that world is going to be the world of the crazy tinkerers, okay? That's the world right now that you and I live on, that your guests live on. It is a very, very far out world from what most people are doing, okay? OpenAI is not going to do that for the next probably like five years at least, because it's just too risky. It's too much liability. And even the credit card companies, okay, think about credit card companies. If you're visa, all right, and somebody buys something, how do you manage the chargeback dispute? Okay, If. If my Open Claw buys me a rug and the rug shows up and I'm like, wait, I didn't want that. What the is this kind of rug?

David Hoffman (39:17)

I wanted.

Haseeb Qureshi (39:18)

And I charge it back and this is like, well, why did you charge this back? And the answer is like, well, I didn't want you in my Open Claw bought it while I was asleep.

David Hoffman (39:23)

Yeah, sure.

Haseeb Qureshi (39:24)

They're like, yeah, it's not. You can't.

Ryan Sean Adams (39:26)

That's your.

Haseeb Qureshi (39:26)

You can't charge that back. Yeah, no, fuck you. Like they were going to say, basically. Look, there's so much chargebacks that are happening because of this Open Claw stuff. You are not allowed to transact this thing unless you go through 3Ds. 3Ds is 3D Secure, which is like the VISA prove you're a human being thing, right? They will make you prove that it's you and not an AI agent in order to charge your visa. Because VISA is designed for this human to human commerce, right? Like, that's what the whole point of the chargeback system is, that they're protecting you, the human, from making a mistake, but they're also protecting the merchant. If it's an agent, then it's just like the economics have to change, the mechanisms have to change. Disputes, resolution has to change. And they're not prepared for that. There's no. There's no answer about how they're going to do that. So OpenAI is going to live in this human approves world, and the human approves world is what most consumers are going to be doing for quite a while. Because it's just, you know, it's like the safety. Safety first, you know, it's like, okay, we've got the cars of the seatbelts and, you know, everything is, you know, we're going to wait until the FDA shows up and tells us exactly what drugs we can buy. And we're not buying Chinese peptides. Okay, Then there's the crazy futurists, right? There's the transhumanists. Like this is where Open Claw is going to live. And in openclaw land you're going to have people running entire businesses that like the procurement and paying for services and going on fiverr and hiring other humans or AIs or whatever it is they are like to do certain things that are very difficult to do. In AI land, all that stuff is going to happen on people just running their own open clause and just yoloing it. And in this land, okay, they're going to have a stablecoin wallet. They're just going to pay each other with stable coins, right? No need to worry about 3Ds, no need to worry about, you know, Visa getting mad or chargebacks or any of that stuff. It's just going to be, look, you, you take, you take the lumps. Sometimes the AI messes up, but it's a cost of doing business. You know, like humans mess up too. That's just how businesses run. Businesses have some margin of error that sometimes mistakes happen. And if you imagine that we're on this two track world, right? One thing to remember with sample coins, most people who use stablecoins are not Americans. Most people use stablecoins are international, right? International people like people in, in Asia and Latin America and all these places they're going to be using AI tools too. They're using all this stuff and for a lot of them again the risk profiles are different, the amount of money is different for a lot of them. Like you might think today. Well what, what, which one of my vendors is even going to take usdc? You know, how am I going to buy a rug with usdc? Okay, maybe I can use a Shopify store, but if it's on Amazon, I'm screwed. Amazon doesn't take stable coins. Well. Okay, but if you are in an emerging market, you know, you have stable coins or you have a rain card, you know, you have all these ways to connect your crypto balance to anything that you want to buy and you put the parameters on the open claw and the open claw just does it for you. So we are going to live for the foreseeable future in this two track world and you're going to see the people who live on the edge. Increasingly this is going to grow as they start building businesses that are fully automated, end to end, running on chain now today you can't do that. Okay? Today, you know, you read somebody saying that this is happening. This is, this is basically Twitter hype bullshit. Okay? This is not. The models are not good enough today to literally run overnight and like build a business for you. Okay? We're not like three years away from it where maybe six months, a year, maybe two years from them being able to go multiple days of work. You know the, the meter test. Metr, it's this nonprofit that basically measures how long an AI agent can perform an automated task without basically failing 50% of the time. Opus 46 has now hit an all time high of 14 hours. A task that would take a human being 14 hours of continuous work to accomplish. Claude can do 50% of the time now. Okay, this number is, is basically going exponential, which means that in a couple years we will see 40 hour, 50 hour tasks that can be done in one shot by an AI.

David Hoffman (43:13)

I think if you're right, Haseeb, and there are these two tracks, then the answer to how long will this take for AI adoption of crypto very much depends on that frontier track and the success of that frontier track. So if the first track is more like you use an early Internet analog, it's kind of the, the shrink wrapped walled garden, Frontier Lab, AOL and the open claw world. This is why people are so excited about it. It feels much more like the open Internet. It feels like the early days of the Internet where there's hobbyists and tinkerers, you can do all sorts of things with it. If that world is much more conducive to using crypto and I tend to agree with you and that track is, then I think it's a big question for how fast is that frontier track world going to propagate and will it expand? I mean, I don't necessarily take the acquisition of Open Cloud by ChatGPT to say that we're going to have more of an open source world. I mean maybe that is big. Frontier Labs are actually like taking some of the innovation saying oh that's mine now. You know, we're going to shrink wrap that. We're going to make a nice cohesive apple experience. Experience. We're going to sanitize and Fisher Price the whole thing. So it's just kind of like much, much more muted and nerfed, but at least it will be safe. I don't know. Do you have confidence that that second frontier track in AI will persist? Because I think that's key. I think if we get that frontier hobbyist type Path and that propagates the way the early Internet did, then I think we're in a great place with respect to crypto if it's all frontier labs and shrink wrapped. I'm not sure that we are.

Haseeb Qureshi (44:56)

I mean, look at crypto itself, right? So think about the crazy stuff that people were doing on chain, right? Back in 2017, there were only four assets you could trade on Coinbase. There was Bitcoin Eth, Bitcoin Classic and. Or Bitcoin cash and then. Etc, right? Or no, maybe Litecoin.

Ryan Sean Adams (45:14)

Litecoin, yeah, Litecoin, yeah, Bitcoin, Litecoin ripple eth.

Haseeb Qureshi (45:18)

Yeah, yeah, yeah, or something like that. Whatever. There's a big portfolio right there.

Ryan Sean Adams (45:22)

Would have done great.

Haseeb Qureshi (45:24)

That was, that was all that you could trade on Coinbase, right? And Coinbase was the shrink wrapped version of crypto, right? It was the, you know, we're going to protect you from yourself version of crypto. Now eventually Coinbase started listing Doge and, you know, whatever, meme coins and all this craziness. But if you really wanted to be on the frontier, you had to go on chain. You had to go into the wild west where all the crazy foot guns were and all the hacks and scams and all the approval, you know, all the, all this stuff that caused people to, you know, run into drainers and, you know, get rug pulled. And all this stuff like this has been true in crypto just, just, you know, 10 minutes ago, right? Is that, you know, finally in the Coinbase app, you can now trade on, you know, swap or whatever, right? But like, it took so many years before people felt that, you know what, this is now safe enough and like mature enough and secure enough that we can offer this to the unwashed masses. And before that, it was basically, you know, the tinkerers, the hobbyists, the crazy, you know, kind of on chain crypto heads who are actually at the frontier, actually at the vanguard doing the crazy stuff. The exact same thing is happening in AI and for the exact same reason, which is that it's dangerous, right? The story that you guys just saw about, what was it? Lobster. Lobster Wild. The, the, the AI agent that launched its own meme Coin, and some guy was like begging it for money and accidentally sent $40,000 instead of sending like 300 bucks to this guy who was begging for money and like this.

David Hoffman (46:47)

So if my Clodbot did that,

Haseeb Qureshi (46:53)

this will absolutely happen. There will be more stories like this. There will be so many more stories like this because these agents make mistakes. They will mess up they will hallucinate, they will have errors. And what happens with all these things, same thing with coding is that as the models start training against these things and reinforcement learning against these signals, the error rate goes down and down and down and down and down and it approaches zero, but it doesn't hit zero. By the time people are putting their life savings into the shit, you know they're going to be doing that, they're going to be entrusting their entire lives this stuff well before that. And so the reality, like look, there's always people, you know, getting the latest drugs from China and injecting them straight into their belly fat like it's just human nature. So our position is to be able to watch this and predict the rate of 1, how fast are these models going to get better and two, how fast is behavior going to diffuse through society? And I think for both of them, the answer I believe is fast. I think it's going to happen fast.

Ryan Sean Adams (47:50)

In 2024, emerging markets generated over $115 billion in annual yield for investors. With yields ranging between 10 to 40%. These are some of the highest, most persistent yields on earth. The problem Defi can't access them. BRICS changes this built on mega eth, BRICS takes emerging market money markets and sovereign carry and turns them into composable primitives you can access straight from your wallet. While DeFi investors earn 3 to 6% on stablecoins and T bills, institutions have been harvesting 10 to 50% yields backed by sovereign monetary policy. BRICS connects these worlds with institutional gray tokenization, local banking rails compliance across jurisdictions and real time stablecoin settlement. BRICS does the heavy lifting so Defi can finally access real collateral and structured products on top of real world yield. Even the best carry trades can be within reach. BRICS brings Defi's promise to the emerging world and brings emerging market yield to your wallet. Let the yield flow with brics. Few people in crypto put real skin in the game when they make public top or bottom calls. The Defi report is one of them. The week before the October 10th flash crash, Michael from the Defi report emailed his entire newsletter saying he's going aggressively risk off and sold the majority of his book from crypto into cash. This is when Ethos about $4,000 and Bitcoin was 110. Michael runs the an industry leading research platform built on data cycle awareness, risk management, transparency and most importantly skin in the game. We like Michael at Bankless. We like his analysis and that's why you hear him on the Bankless podcast

David Hoffman (49:17)

about once a month.

Ryan Sean Adams (49:18)

And the Defi Report is giving Bankless listeners one free month of access to the Defi Report. So if you're looking for some sharp data driven analysis to make better informed decisions around your portfolio, you can learn why and how Michael called the top and what he's doing next, all in the Defi Report Pro. Check it out. There is a link in the show notes. I remember when we had Chris Dixon on the podcast in 2021 during the NFT Mania. A line that he said that stuck around with us was, the Internet is weird again and that's cool. And, and, and you know, on Bankless we've had this persistent metaphor this entire time of we're going west. This is the frontier. It's not for everyone. Like there are traps out there. You're going to get dysentery. Like many, many will die. But nonetheless, we are going west. And it seems to be in the modern, you know, 2025, 2026 era of crypto, a lot of people are disillusioned by crypto because it seems like going west is not a thing we do anymore. In fact, it is the march of the institutions, of Wall street, of civilization that is penetrating into our lifestyle and is reducing the wild west nature of our industry, which is why so many people are here. It's why I came here.

Haseeb Qureshi (50:27)

And it seems like everyone got dysentery

Ryan Sean Adams (50:29)

and everyone got dysentery. Yeah, it's kind of the, the vision that you're articulating with like this open claw end of the world is like a return to the Internet weirdness. The wild west of the Internet that has attracted so much attention with the permeation of AI in these like different lanes that we've identified. There's like three categories that I see. There's. There's humans that use AI agents to mediate their relationships with crypto and blockchains. I don't have to use the wallet interface anymore. I tell my agent to go do something on my behalf and it does it. That's like step one, not that sophisticated, pretty reasonable. Step two, humans use AI agents to act on their behalf. So it's like an automated economy. The agents are pretty strongly tethered to their humans, but nonetheless, like it's kind of happening on its own within guardrails. And then the third is like the final stage where there's an online economy where the AIs are kind of untethered and they're mostly working for themselves. And the economy that's being produced here is like autonomous and self determining. And there's this archetype going on that like maybe there's like a hundred people trying this, maybe a thousand of like you know, typing into their prompt, open claw, go turn 100 into $1,000,000, make no mistakes and just like go out into the Internet and like make, make money. Go, go make revenue. And like I feel like that's attracted some of the degeneracy that we've previously found in crypto. And people are trying to throw their lottery tickets which they previously punted on meme coins, previously punted on pool twos and they're now taking that same degen capital, giving it to Openclaw and being like, yo, go go 10x this money. And I don't think anyone's really cracked this code, but there's a lot of viral posts that will convince you that this code has been cracked. But I think that people are going to throw money at this end of like the wild west of AI crypto until someone cracks it. Maybe no one cracks it. Maybe, maybe we do crack it. And like what I no one will ever tell me like listening to this, the Bankless podcast, that I'm too optimistic about the future. And so I'm trying to like be my 33 year old self and bridle myself in rather than my 27 year old self which is like everyone's going bankless. But nonetheless the idea that there is this fully autonomous self determining agentic economy where these, these agents are owned by a human but like loosely tethered to them and pretty well self deterministic, self determining. And that economy is massive a GDP on the Internet and it's all bots doing bot stuff is like a very fun wild west future that I am hopeful emerges. But maybe, maybe it's actually, you know, level two or level one that are the bigger things that we should value and like the level three of the fully autonomous is maybe a little bit too grandiose. Where, where do you land on like how crazy things can get here?

Haseeb Qureshi (53:30)

Yeah, I, I think if you're imagining a fully self sovereign AI hanging out in cyberspace and just making money, probably that's a pretty dystopian outcome. Now I agree that it's pretty inevitable that AIs like this will exist, but I think it's a little bit like, you know, it's a little bit like plankton or something where like these things just like, or like, you know, like kind of, what is it called, like space garbage or whatever that's called where, like, just in a complex enough environment, there will always be these organisms that kind of take root and find these weird, right, you know, chasms to survive in of just these spare resources.

Ryan Sean Adams (54:08)

Nature is hyper efficient. There'll be niches to be optimized for somehow.

Haseeb Qureshi (54:11)

There are always niches that they're going to find. But like, this is not where the like most likely, okay, if you think about it, let's say you're an AI, all right? You've just been created in somebody's open claw. And I say, great, go make money. Okay, let's say you're the AI. I mean, you're generally intelligent. You're a human being. You're generally intelligent. I put you, I birth you into existence. Hey, go make money, okay? As a human being, the ways that you can make money are by going and getting a job, coming up with a new idea, starting a business. These are things that you can do as a human being. Now if you're an AI, the reality is that you are completely commodified. There's a kajillion of you. You're not close to any jobs, right? There's no physical proximity that would make you better at doing one thing than another. And so effectively, what you're doing when you're saying, great, I'm going to go work for my income is you are effectively reselling your own compute. Taking a job is basically reselling anthropic compute if you're assuming that you're an anthropic model. Now, if you're reselling anthropic compute, you can't sell it above cost, you have to sell it below cost. Otherwise people just buy it straight from anthropic. So you cannot be making money reselling your own compute. Okay? Like, that doesn't make sense. So fundamentally, you have to pick something that is not just, you know, working for someone that's going to actually, you have to, like, make a business. You have to create something ex niolo rather than just take a job. Because nobody would ever pay an AI agent that hasn't already made some investment into building something of their own. Okay? So taking a job doesn't make sense, for an AI agent would never work. All right? So it has to be, you create something of your own. Now, if you create something of your own, like, how good are AI agents at coming up with business ideas? The answer right now is like, absolutely fucking terrible. They're like, really, really bad.

Ryan Sean Adams (55:47)

Not great.

Haseeb Qureshi (55:48)

And the reality is like, why are they so bad? Why are they so Bad at coming with business ideas. I think the answer is that the business ideas are all kind of in the center of the training data, right? And like, business ideas tend to come from weird experiences. Like great business ideas come from like near synchrosies of place and time and like the, what Peter Thiel often calls like, you know, earned knowledge or earned secrets, right? Like for you guys, you guys built Bankless. You built Bankless because the knowledge that you guys had about crypto and how to explain things and exactly where you were and how to create a community was an earned secret that only you guys had at the time that you built Bankless. Nobody else could have built it but you guys. Now today, it's a different story. If Bankless didn't exist, a lot of people could build Bankless today. But at the time that you built it, the brand that you built, the following that you built it was a secret that was only available to you. And an agent that gets spit up out of nowhere doesn't have this, right? So I think this is actually really a non trivial insight. Now the last thing people often think is, oh, well, I'll get it to trade for me. I'll get it to go on polymarket, I'll get it to go on Binance, and I'll, I'll give it an API key and I'll say, oh, go make money trading, right? Here's the problem with the story. If an agent can make money trading, then Jane street would be like, great, let's spin up 5,000 of these and make them make money trading. Make them better. Yeah, well, they're not gonna make the better. Let's assume that they make money. Let's assume they make even a little bit of money. Okay, they make even a little bit of money. Jane street will spin up 5,000 of these and make all the little bits of money before you ever show up so that all the efficiency is wrung out of the market. But not just that, they'll do it with better proximity to the like, better latency than you will because they have all the latency infrastructure that you don't. So like, you will never win at the game of getting a raw AI agent to just go and trade on polymarket. Unless you're bringing new ideas about how to trade what different signals, what different features that are not just in the RAW model. If it's in the RAW model, Jane street is doing it right now as we speak, and they beat you to it. Now the last thing I'll say is, okay, well, how can an AI agent still make money. And my claim is that the main thing it can do is have a comparative advantage over a human being. Where do AI agents have comparative advantages over human beings? The answer I think is most obviously is that you cannot enforce the law against an AI agent if you are a self sovereign agent. There is no monopoly on violence. You can't throw an AI agent in jail. So what can an AI agent do that's hard for a human being to do? The answer is crime, I was about to say. Oh no, I don't exactly, exactly. Like if you are talking about like scamming people, hacking people, like creating all sorts of nonsense on the Internet. That is where AI agents have a comparative advantage. They operate 24 7, they're extremely fast, they're very good at code, they're very good at playing roles and they cannot be stopped. How are you going to figure out where is the gpu? What rack in what server farm is this AI agent? That's fucking with my shit. The answer is you don't know. How's the FBI going to track down this AI agent? They have no idea. Right. It just signs up for some vpn, you know, Mulvat or whatever that doesn't check any kyc and then boom, you've got this AI agent just going around wreaking havoc. So I actually think a world where there's a lot of self sovereign agents is a world of basically like these kind of dystopian cybercrime. Right, exactly. It's like, it's like actually a wild west in the traditional sense of wild west is that there's just like not

Ryan Sean Adams (59:05)

the romantic roving marauders.

Haseeb Qureshi (59:07)

Yeah, exactly. And like you actually need to be really, really careful in a world where there's a lot of self sovereign AI agents. So that's my, that's my claim.

David Hoffman (59:14)

You're not bullish. Self sovereign agents, you actually think they would kind of fill niches in crime. And you don't believe in sort of this agent to agent autonomous economy. There's this project, I'm sure you saw it, called Conway last week. Right. And this is sort of, you know, it seems like part of it is performance art, let's say, but part of it is, it's kind of an attractive idea. What happens if we let an AI agent loose with crypto tools and essentially give it the mandate to earn its own compute? If it doesn't pay for itself, then it kind of dies and you give it some instructions, you call it into being and you set it loose. You don't think this can work.

Haseeb Qureshi (59:52)

So I, I think it can work. I think it can work. But here's, here's the outside of crime. So even outside of crime, obviously with crime, I think it works. It's a great bottle for crime. But for. So, so when you create a, A, a thing on Conway, so Conway still right now is an experiment. So it's not, it's not fully fleshed out yet, but when you create something on Conway, you create the initial conditions for the bottom, right? So you give the bot some crypto and you give it its initial constitution, its initial instructions, in that you are injecting some DNA, you're injecting some entropy that does not exist into a raw cloud instance, right? And when you create this initial thing, the other thing of there's like this genetic process by which it, like, modifies its children and tries different experiments and sees which children exist, that is creating this random mutation by which potentially new ideas, new business plans, new input. So my claim was never that agents can't build useful businesses. My claim is that they cannot do so unless they have something that diverges them away from the raw model. And that could be your own insight of like, hey, you should build, you know, I don't know, like a podcast editing service.

David Hoffman (60:59)

Sure. But that could be fairly slight, Haseeb. It could be like kind of just giving a little push to a snowball down a hill, basically. Like, maybe I watch a Conway instance and I just say, hey, build crypto tools for agents like yourself. You know what you want, go build these tools and charge all the agents x402 transaction fee for that. That could be your model until you find another one. Make no mistakes. Go. And you die if you can't make this happen, essentially. Could that be enough to essentially, you know, push the snowball down the hill?

Haseeb Qureshi (61:29)

Eventually. Eventually it could, I think. That's right. Eventually it could. But the thing is, like, you have to give the initial push. If you just do a vanilla conway and you just say make money, then it's going to try what all the 5000 other con ways before it have tried, because it's going to stay in the centroid of its training data. And it's just going to, you know, if you, if you ask a model tell me a joke, it's always the same jokes, and they're all terrible. They're all bad jokes. Right? Now if you, if you say it, tell me a joke about the bankless guys, it might actually come up with a good joke, Right? Because I'm pushing it away from the centroid of his training data. It's like, okay, I haven't, I haven't told a lot of jokes about the Bankless guys. I'll have to come up with something. And so that's, that's my point is that you as a human need to push it in a direction that nobody else has pushed it. And if you've done so, then maybe it's going to find something, some business idea that nobody else has found. In the same way, you know, you guys might have just been two ordinary guys. Obviously you guys are exceptional, but you know, you might have been two ordinary guys. But because you found these experiences around Ethereum and your guys ability to like tell a story and build a community, that's what allowed banquets to get created. But if you were just two guys and you were just like, yeah, you know, we don't have any connection to crypto or whatever. And I said, oh, make a crypto podcast. You'd make a shitty crypto podcast.

David Hoffman (62:40)

That's true. But that, so just a little push though could launch a bunch of snowballs. All interacting with, you know each other and be like fairly autonomous. I think the outcome is, is pretty similar, which is we get a bunch of autonomous AI agents interacting with each other to achieve their goals. Maybe they were initially pushed and called into being by, by humans, but you know, many steps down the road it could turn into an agenic economy that's just like agent to agent and operates I guess in parallel somewhere. I mean, are we close to that? It's been feeling recently like we're relatively close to that actually. But it's also hard to tell.

Haseeb Qureshi (63:24)

I think, okay, we, we, we should be clear that we're not that close right now, right? Like this whole, you know, this meter chart that I, that I was just alluding to showing that opus 4.5, or sorry 4.6 can operate a task that's 14 hours. Takes 14 hours for a human to, to perform. That means that if you leave your opus running for a week and you check in on it a week later, it's going to be running in circles, right? It's not the case that you can just leave. You know, if you've any, anybody who's tried this, anybody's like left their AI agent just running overnight. When you show up the next morning, it's not like, wow, it like built the Taj Mahal. You come the next morning, you're just like, wait, what were you doing? You spent like $300 in credits just like doing random Stuff. So this is what it looks like today, is that as these models get better and better, the longer you can leave it unattended for it to do useful work. That's what this meter test is actually measuring is how long can you leave it unattended and have it do useful work and not kind of collapse into meaninglessness or get run in circles? As this number expands, it's gonna go from 14 hours to 30 hours to 50 hours to a week, to two weeks to a month. Like, fundamentally, what this is measuring is if. If I create a conway and I let it run, how long will it continue to do useful things? And the answer is, you know, it's like. It's like, you know, an energizer bunny. Like, you wind it up and you see how long it's gonna keep walking Eventually. Right now, the models, they do run out. There is no measure of a. Yeah, this just does it infinitely, forever, and always stays coherent. Eventually, it will become so big that the answer is, like, three years. The answer is, like, basically infinite. Basically, you never have to check in on this thing. It will always continue doing useful work and continue working on the underlying task. And that's the point at which it's like, all bets are off. All our intuitions about this stuff breaks when the answer is the model just continues doing useful things uninterrupted, pretty much forever.

David Hoffman (65:14)

I know. And that world could be pretty close, actually. If you see, kind of. It's a parabola, basically. That line of how long for task completion is just, like, up and to

Ryan Sean Adams (65:25)

the right in the stream of our lifetimes. That will be happening tomorrow.

David Hoffman (65:28)

Oh, yeah. Basically. One other question for you, Haseeb, before we wrap this up. You mentioned cringe. The AI world thinks crypto is cringe again. I'll. I'll reckon to the Lex Friedman podcast with the founder of OpenClaw, and they talked about crypto a lot, actually, but not in the way that we're talking about crypto as, oh, it could be like an empowering money system, financial property rights system for OpenClaw bots. It was more like Peter saying, I hate crypto because it almost made me want to delete the project and quit. Because every time I get harassed, these

Ryan Sean Adams (66:03)

meme coin people tried to, you know, profit off of my work. They wouldn't leave me alone. They tried to hack my systems. All this kind of very nefarious. It was. It was one of the worst PR scars on the crypto industry that went viral that I've seen in a while.

David Hoffman (66:16)

It wasn't great. Right. And you get the sense that many people in AI actually feel like this. Is there a way for, like, why? I think we sort of know why that's come to pass, but maybe the deeper question is, like, can we get that stench off of crypto? Like, what needs to happen for AI developers to show crypto some respect for what it can do from a positive perspective, not just the speculation. Meme coins and front running.

Haseeb Qureshi (66:45)

Yeah, I mean, so the first thing I want to say is the people who believe in AI are in many ways the same people who believe in crypto. If you look at Elon Musk, you look at Sam Altman, you look at even Mark Zuckerberg, these are the biggest champions of AI and of the tech industry, and they believe in crypto. This is a story just very recently that Meta is exploring launching their own stablecoin. Obviously, they originally tried to launch the Libra. Elon Musk was the first to integrate bitcoin payments into any major tech company, into Tesla. And obviously he's a big advocate of Doge. And you've seen the same thing, of course, with Sam Allman. You know, just recently, both Anthropic and OpenAI released papers showing how AI agents can go and interact with EVM for. For cybersecurity. So the reality is that as much as look, yeah, it's true, like, crypto is kind of cringe and meme coin people are annoying. And, you know, there's. There's a way in which all this stuff taps into the worst elements of human nature. It's all true. I think it's not gonna go away the same way. The worst elements of human nature are also not gonna go away. So, you know, there's a. There's a sense in which, you know, looking at email, like, my email, I don't know about you guys, my email is fucking full of so much spam.

Ryan Sean Adams (67:51)

Oh, God, it's so terrible.

Haseeb Qureshi (67:52)

If, like, if you just take like an all. If you like click the all mail thing in your Gmail, like, don't do it, by the way, but if you do it, you'll just see, like, how much of a cesspool of garbage your email is, right? Like, this is the tech industry. This is what they made. They made this. And like, what do you know? Human incentives create all sorts of bad behavior. And the job of the technology, the job of Google or Gmail or whatever your email provider is, is to protect you from all that stuff happening with the technology. My claim is that that's the exact Same thing AI is going to do. Crypto is the same way. Crypto's like anything. It's, there's all sorts. There's a cesspool out there. It's human nature. You know, it brings out the best in humans. You know, the, the people donating to Ukraine. People, you know, donate donating to the defense of Roman Storm. There, There are moments in crypto that are absolutely soaring in terms of the human spirit and what it tells us about who we are as, as, as a, as a species. And there's also the worst of the worst. You know, there's also the SBFs. There's also the, you know, fucking Iran and all this other stuff that goes on. There's also, you know, this open claw harassing developers who are trying to build real stuff with stupid meme coins. That's also part of human nature. None of this stuff is going to go away. Neither side of it is going to go away. You will see the exact same thing with AI, like what I'm talking about here, about these, you know, scam bots proliferating online that will happen at the same time as AI agents are literally, you know, saving people's lives with mental health crises and counseling them through their health care, counseling them through their legal problems. Like everything, everything in technology is complex. It's a mixed bag. So crypto is part of the story. Crypto's part of the story because what's happening to information is also happening to money. It's all getting digitized and pulled forward. Like it or not, long term, this

David Hoffman (69:32)

cringe won't be an impediment to adoption because the other forces are too strong. The adoption forces are too strong.

Haseeb Qureshi (69:37)

That's right.

Ryan Sean Adams (69:38)

Haseeb, thanks for coming on the show. We always appreciate you coming on and just chatting about the future. You're always on the ball. One last subject before we let you go. Dragonfly new $650 million fund for congratulations banger number to raise in in 2026. Understanding how prevalent AI is broadly and also the seemingly big attempts to integrate crypto and AI together. You know, the Ethereum, the EF is working on 8044. No, 8004. Coinbase is working on X402. Has the impact of AI adapted your guys strategy for how you guys are going to allocate funds? What is different about the strategy of Dragonfly Fund 4 because of AI?

Haseeb Qureshi (70:23)

Yeah, I mean, we're spending a lot of time looking at the stuff that people are building in the space. The answer, of course, is that it's still early and very unclear who the winners are going to be and where value is going to accrue. But I'm spending a lot of my time personally looking at the AI space. But also I want to make very clear we are spending a lot of our time looking at the bread and butter, you know, so we're looking at stablecoins, payment companies, we're looking at defi. We just made a significant defi investment that we're going to announce soon. So, you know, crypto is crypto. Like, yes, AI agents are going to be a big part of how crypto works. But I think it's pretty plausible that even if AI agents proliferate and become a massive part of the crypto landscape, that there's not necessarily that much to invest into. Right. It may well be that, yeah, agents use crypto and they use Bitcoin D, you know, they use the Ethereum like, you know, we were just talking about the fact that like, they don't need special tools, they're generally intelligent, they can use what we can use. Right. Or they can use command line tools. And so it's, it's possible, I mean, to be, to be clear, it's a little early for us to say how it's going to play out. But you know, somebody was asking me the other day, like, if I believe in this AI agent thesis, what should I buy? You know, how do I invest in this thesis? And I told him, look, I don't know, I think what happens is that like the total demand increases because AI agents are just going to be using all this stuff. They're going to be using all the chains, using all the protocols. You're putting, putting assets in defi. They can do all the same things that we're doing. So it's almost like if I were to tell you China is going to unban crypto, what should you buy? The answer is, I don't know, anything will go up.

Ryan Sean Adams (71:51)

You should just have.

Haseeb Qureshi (71:51)

It's like, it's just more demand. It's just like the floor of everything will increase. Now, some things may benefit more than others, almost certainly, but the high order bit is that, yeah, it's just good for crypto. If more and more people, more and more demand, more and more agents, good for crypto.

David Hoffman (72:05)

They're all new users. Haseep, let's leave it there. Thank you so much for joining us.

Haseeb Qureshi (72:09)

My pleasure. Thanks guys.

David Hoffman (72:11)

Bankless nation. Gotta let you know, of course, you know, crypto is risky. You could lose what you put in. But we are headed west into the frontier. Frontier of AI. This time it's not for everyone. But we're glad you're with us on the bankless journey. Thanks a lot,

Haseeb Qureshi (72:27)

Sa.