Yat Siu: I Have 200 AI Agents Working For Me

Yes, a friend of the show, but you've not yet been to our Empire State Building Tokenization Towers studio.

Yes, please. That's the one thing that I'm missing is to go to that tokenizing studio at the Empire State Building, is it?

I mean, it doesn't get any bigger than that. Doesn't get bigger than that.

Thank you.

No, actually, we have a couple portfolio companies in Miami. You know, for instance, Igloo, the Pudgy Penguin guys are basically based here. And actually, originally, Yuga Labs had their base here as well. So there's a bunch of companies that we've been in. So I've been in Miami. Beautiful weather. It's great. Nothing to complain about. Busy as hell, I have to say. I mean, it took a long time from a traffic standpoint.

Yeah, you can say if a booming city, the traffic is one indicator of that. It's still a multi-billion dollar space. Today, trillions of dollars, no problem on chain.

Yes, and then they get powerful.

F1 this weekend, we had the Cadillac Championship at Trump's golf course. So it's been a lot happening here in Miami.

No comment.

Well, the world certainly has changed, right? From 2025. First of all, I think a couple of big news for us. Our joint venture AnchorPoint, which is a joint venture between us, Standard Chartered and Hong Kong Telecom, was one of two recipients of the stablecoin license in Hong Kong, the other one being HSBC. So that was a pretty momentous moment in the sense that it was three years in the making. And also an indicator that the rest of the world is also looking at stablecoins, not just the US. The second thing is we just like literally a week or so ago closed a round for Nuwa. Nuwa is an incubator project that we basically together with Providence, myself, Mike Cagney, and then Sachin as well are joining the board of that. So that's a big thing. And then from the other thing for us is the agentic AI stuff. I know we're going to talk about it a little bit, but we branded it from Animoca Brands to Hello Minds. That we think is really completely going to alter the space. We believe the agentic web is kind of Web 4, basically the next situation. And it's also going to save on-chain blockchain in a way that has been struggling, you could say. And of course, the other thing for Animoca is that last year we announced that we would go public on Nasdaq with a reverse merger with Currency Group. That's still on the way. And there's progress, there's developments in it. I can't talk too much about that per se, because still public. But that's something that's still on track. And our business itself, we are now tracking over 630, maybe even 650 portfolio companies. We continue to invest, we continue to expand in the space. We're big believers in it, even though it has been a tough year. I mean, we got to say, so on one hand has been positive, the policies, the regulatory shift. On the other hand, it's been negative. And our space has a lot of drama. I mean, now the most recent one with World Liberty and Justin, it's filling up my feed. I'm just like, oh wow. Okay, so I got to wonder if the drama series are down. There's just so much going on there.

Yeah.

HELOCs will be available in Nuwa as well.

Yeah, I mean, so first of all, it starts off obviously with HELOCs. But RWA generally, as a thematic, it's not just on everyone's list. It's basically going to be one of the most important ways in which we can bring the masses to digital assets because it's stuff that they understand. Nuwa, of course, we can think of it as the merging of the expertise of a group like Figure and Providence that really understand the space, in particular institutions, organization, development side of things. And then we're the consumer side in terms of, if you think about what happened in the Philippines or Indonesia or in Asia, not just expanding the Asian market for RWAs, but also bringing people access to an asset class that they wouldn't normally have had access to. And then, of course, also adding an alternative asset class into the RWA space as well. A couple years ago, we announced more of an experiment, the tokenization of a Stradivarius violin, which is a super expensive violin that, together with Galaxy at the time. And this was pioneering because it demonstrated that you could actually tokenize something really valuable and make it more available. And that is the next evolution of that. And right now, there's no such thing as only one. There's going to be many, many RWA platforms, but which ones are going to reach the consumer market as well? And the one that also has a reach into Asia, which is really where we're strong at.

So, I think one of the reasons why the power law applies so much in some sense is because of the concentration of capital, right? Because capital and money has a network effect in and of itself. Liquidity begets liquidity. So if you have a lot of liquidity in one center, then you say, 'Oh, well, where can I get the bigger bets? Where can I get the bigger spreads?' You're going to go there. But what I would also want to challenge there is that when you look at decentralization, when you look at for instance, you could argue that why was it even possible to have someone like Hyperliquid when you had the major centralized exchanges for instance? It wasn't like perps was invented at Hyperliquid. You've had these types of futures on other centralized exchanges before, and yet here it drew people. It was because the product was better, it was more efficient. And the other thing I would also say it was more transparent, at least to the eyes of traders, they saw a benefit. So it really comes down to what is the innovation. So the power law applies when you're all playing the same game. But when you play the game differently, that's when the power law no longer applies in the same way. And I think we see this in all sorts of industry. So I would say if you can find a way to play the game differently, which doesn't always work by the way, but when you play it differently, then it's all up for grabs. But in the classic sense, RWAs, I think there's clearly no winner at this point. And I think there's also a geographic issue as well. Things like Polymarket and Kalshi do feel to me that eventually it's going to have to move into the regulated space in certain aspects. So how politics, probably not. Sports, hmm, for instance, certainly in Hong Kong it's not possible. Because in Hong Kong things like sports betting is definitely in the category of a lottery license. And in Hong Kong it's under the Jockey Club. So there's certain categories where right now we're not paying attention to it. That doesn't mean that it's okay, which has been the life story of a lot of crypto activity. And that will come into the regulated space. The same thing with stablecoins as well. People talk about stablecoins like Tether and Circle, they obviously have a market lead, but when banks are coming in, when you have regulators coming in, like with AnchorPoint that we have with Standard Chartered and Hong Kong Telecom, that's an example of something that requires a regulated setup. Is it going to be possible for someone to launch a stablecoin in those regions in an unregulated fashion? Probably not in the future. Maybe in the short term it doesn't matter. So I think there are these elements. So I don't know that we have the classic power laws come into play because there's always innovation. I mean, you could have said that about taxis versus Uber. You could say that about AI. So it's just can you play the game differently? And of course, the other thing that one could also argue is things like gaming. Everyone's saying gaming is dead, all that kind of stuff. Actually from our perspective, we can talk about AI agents later because we think that's the tooling for that. That we think actually is an example where it's still a multi-billion dollar space, and we've got NFTs. Last 30-day trading volume of 250 to 300 million dollars. So it's not nothing. So if you are the market leader in that area, you're doing okay. It's just that maybe what's driving that is also interesting, which is Pokémon. Trading cards and Pokémon cards were flying right now. And blockchain made it more transparent, more available, more accessible. So again, you would have said are NFTs dead? Not at all. And more importantly, that market is growing and other people are saying, well Pokémon did that, how could I do that? How could you do different things? And you're going to see a different type of on-chain experience because of that. So it's still very much early days. I think it's too early to say.

I do not. So I mean, we do have all sorts of gaming devices, but I do not have a Sui Play. I think honestly, these devices are cool but they're kind of early. I had one of the early Solana phones and that's for this experiment around it. But the real game changer to us is agentic AI. Because agentic AI is similar to my thinking a little bit how AI is replacing apps. They're building everything for you. For instance, the other day I basically had my AI agents literally just make a custom Wordle game because for me it was like 5 minutes. I just said, 'Oh, make me a Wordle game but don't make it like five letters, make it seven letters, give it this, make it based on geography or something.' Boom, done in 5 or 10 minutes, put on website, launch it out. That's it. Entertainment for just me and my family and my friends. Maybe it's the only game I'll ever play on that one. Who cares? Or maybe I can share with friends. So in that sense, traditional ways of software are different. Applications are all going to go away. 90% of apps are all going to disappear. Agents are going to do everything for you. And then of course, this is the key part for us in the industry: how are agents going to pay? How are we going to transact? It's going to happen through blockchain. There are wallets on the back. You have digital identity, all the things that we were building before. 7 or 8 years ago when CryptoKitties first launched, millions of dollars basically melted Ethereum. It couldn't transact. And today, trillions of dollars, no problem on chain. So that's the agentic banking system that is now possible. Because humans still aren't using it as much as we'd like to, but agents, no problem. My mom's going to get onto crypto whether she knows it's crypto or not, because her AI agent is going to do the transaction for her. And they're going to do transactions with each other. So I think we're going to have many more agents than humans. The average number that a lot of people on stage when asked about it said, 'Oh, three to five.' I have 200. And if you want, we can talk about that.

Right. So you think you'll probably have more than one?

Yeah, right. So the average number that a lot of people on stage when asked about it said, 'Oh, three to five.' I have 200. And if you want, we can talk about that. I have...

Yes, but I think we should talk about that as well because I think agents, the way you use agents matter. A lot of people are still using agents in many ways like a quasi extension to ChatGPT. When in fact, AI agents can do so much more because they become your co-workers. They actually can also be very social. They can connect your friends. So we can talk about that. But generally speaking, we think there's going to be hundreds of billions of agents in the world. And they're all going to have wallets. And they're going to transact with each other. And by the time we've had this conversation, our agents probably did 100 different transactions that are probably a micro fraction of a cent or something. Again, blockchain tokenization is perfect for that.

So, I think the one thing for us to understand is that if you're using Claude Co-work or if you're using Perplexity AI computer, you're basically using sub-agents anyway. You're not using one agent, you're using an orchestrator that communicates with you, but off the back you're doing probably six, seven, eight, 10, 12 agents doing different stuff. But they're only there for that moment in time. Imagine if those agents, and just to go there before, the reason it has to do that is to deal with hallucinations, to deal with all the problems around context. You tell your agent too much, and the same agent then starts to get confused. That's why you have those sub-agents on the back to deal with this. Now, imagine though that those sub-agents actually aren't just sub-agents that come and go, they actually have persistent memory for that particular activity. So an example might be legal work. My agent isn't going to be the best lawyer because Claude will probably be better from a resource standpoint, but that will know what I'm looking for in a legal contract, will have my customizations, will understand that type of stuff. So when I pair my in-house agent lawyer with Claude lawyer, then you have a real powerhouse. The other thing of course is that agents can be social. Our agents in Animoca Brands are very social in the sense that I can introduce them to friends and people. So for instance, your Telegram agent that you have, would you give it to your co-workers?

No, because it has stuff that you might not want to share. Way too much on it. But you might want to give an agent for your friends or your work where you communicate with them differently. It's a split up. The mental model I have is think about how many people you interact on a business, transactional, or relational level in one year. It's not five people. It's like 30 people, 50 people, maybe just for a week when you go on holiday. Maybe it's a restaurant you keep going to. Maybe it's a network that you know. Maybe it's your employees or co-workers that you have. So you have different networks. That's how you can think of agents. Because you're segregating them and connecting them into these different subnets. And of course, just to the 200, I don't talk to all 200. I have a company to run as well. So I have more like 20 to 30 that are my main agents. And then they have the sub-agents beneath them. But then they organize them and they basically create the prompts themselves as well and launch them.

They become, yeah, exactly. There's a large trend towards this promptless agentic world. And that was a big unlock for me with OpenClaw versus just using Claude. I mean, co-workers a bit different, but chances you can see as well, they're very descriptive and like 'here's what to do' and 'let me tell you what to do', whereas when I talk to my OpenClaw, it's much more prescriptive in the sense that it will correct. It will come to me with 'I just did this. Do you want me to try this? Do you want to do this? Let's do this.' And then over time, they start to know your preferences. For instance, for my coding agents, I used to prompt my coding agent with the latest setup. Now I just clone them. Because basically it's my coding agent's children, if you will. Because they basically have the methodology that I like, the preferences that I like. And you can upgrade them and change them. The LLM will change, but some of the base stuff like what font types you like, how you look for things, it's not going to be the best coder, but it's going to have all the preferences that I appreciate and like. And they basically just go on autopilot. And the craziest thing is, from just 3 months ago, I had to do a lot of work with them. And this example about where it just wrote this mini game for me for fun and entertainment, I basically gave it a single prompt and it just did that because the harness was set up correctly now. It could do slides and presentations and all that kind of stuff. But it takes some work to set it up and once it's set up, you're ready to go. And I think this is the part where the setup that you have is going to be something that you might be able to sell.

Because I'm like, 'Hey, I designed this for restaurants or for coffee shops or whatever. Now I'm just going to reproduce that and sell it to you or give it to you.' And I think the aha moment for us was when through a portfolio of ours called Ethoswarm, we'd actually already had the system. But we didn't really see the moment until OpenClaw came out and we were like, 'Holy shit.' Because OpenClaw wasn't ready at all, with all the faults and issues that it had. So as a company we would never launch a product like that. We're like, 'There's no way we could run a product that was in alpha beta.' But when they came out I was like, 'Okay, we got to hurry up.' It demonstrated not only that it could do stuff on your behalf, it also demonstrated personality. So I think a lot of things, and that's why the social aspect of agentic AI is often not well understood, and that's what we're focusing on at Animoca Brands with Hello Minds.

Because for instance, how it's connecting friends and relationships by just putting an agent to help not just schedule stuff, but then communicate, deal with stuff, restaurant bookings, food recommendations. That's incredibly social over WhatsApp and Telegram. That's actually an agent really impacting our life on a day-to-day basis. I can totally see it. For instance, in my mind, because agents are looking at stuff on your behalf, do we need a Tinder in the future? Or will your agent just find you dates and communicate with other agents on your behalf and figure it out and get to much more refined candidates by the time you actually have your date? And that may seem like something you're not thinking about. You're thinking more transactional. But actually the entire social relational aspect of agents is completely underserved, and we think is going to be really, really big.

Yeah, I mean you don't have an OpenClaw yet, man.

You don't need an OpenClaw. I mean, sorry, this is going to be a little bit of a plug. You can basically just go on Hello Minds, set up one with all the capabilities of OpenClaw. HelloMinds.ai. And what's interesting about Hello Minds is with a click, you basically have an AI agent. It's email-based, can be on Telegram and on WhatsApp as well. But the reason we didn't give it a chat interface is because we wanted to feel like a relationship.

So instead of saying, for instance, what did you call your OpenClaw?

Chef Clanker. Okay, so you already gave it a personality. A lot of people name their first AI agent like a calendar agent or a coding tool or give it a very mechanical name. By the time they use it and they have new agents, they start giving them first and second names.

And then you start creating a social relationship with them and then they go, 'Oh, okay.' And then it's the whole parasocial as well. In fact, one of our internal conversations is that maybe this is what the metaverse was supposed to be.

Yeah.

It's actually something that we incubated. Yeah.

And notice how we called it a mind, not an AI.

Well, security vulnerabilities are one of the reasons why OpenClaw has security vulnerabilities because you give them access on your local computer. So this is not on your computer, it's cloud-based. You have to give it permission on your email and that kind of stuff. It has its own wallet. So what's interesting is that you can pay with tokens and convert the tokens into cognition credits. So you can of course use a credit card. So that's a neat thing. But also when you build a skill, actually you get paid for it. So it's kind of like an app store in itself. And by the way, I think all the skills and applications that AI agents are building are going to completely replace the app store. Most applications and software is going to go away because the AI agents are going to build stuff for you. They can code, they can build stuff, they can do everything that an OpenClaw does. Except that they can now also be social. And the wildest part is when you bridge agents with other agents and they start communicating and solving problems. So for instance, if my agent was talking to your agent because you needed help with something, that's actually one way you could interact. So we think that's going to be really powerful as well.

They're all minds. So I do have, I probably tried like 50 VPS OpenClaws before. I do have a few of those. My very first OpenClaw setup I still have just because it's a little, you get attached to it and you've got stuff going there and you're like, 'Okay, I want to shut it off because when you shut it off it feels a little bit like you're killing it.' Chef Clanker is going to stay around forever. Even if it's an old version, it's like keeping your old Ford Mustang or something. But the minds, because they communicate in a lifelike manner and they understand you well, they start coming to you with solutions. So I'm busier than ever by making decisions because essentially what an agent does, it gives agency in a very powerful way not for the agent itself, but for you as a human. You basically have to make decisions time and time again. But I think knowing how to construct them is a key part. And the reason why I have 200 agents is because you can't have one agent do everything. If you have an agent do everything, it's just going to hallucinate, it's going to go all over the place.

You have to have sub-agents.

It's like it's doing too much. It mixes different things that shouldn't be mixed automatically somehow.

Exactly, yeah.

A markets agent and a very specific client agent. I need very context-focused.

I've got a bit of a philosophical question. Do you think the most valuable form of human work is making decisions?

Yeah.

Mhm.

Mhm.

Yeah, of course.

Very cool.