Autonomous AI Agents Uncovered: Yat Siu of Animoca Brands Explains The Agentic Commerce Evolution

Very good. Thank you for having me.

Indeed, indeed. Looking forward to it.

Yeah, I mean, Animoca has been really active in the crypto blockchain space really since late 2017. I guess we had our first big moment when we were to a certain degree with CryptoKitties and essentially helped usher in sort of I guess the NFT age and I know we're going to talk about that a little bit. And of course, you know, our sort of ownership of Sandbox, our involvement in Axie Infinity, OpenSea, you know, at Dapper Labs and so on basically kind of made us a little bit of a household name in the early days of crypto, but I think a lot of people might not remember that it was a long slog, right? It was from 2017 up until '21 when it really sort of started to take off.

But today Animoca has become somewhat of a conglomerate in the space. We have 600 portfolio companies. So some people do refer us to a little bit as a Web3 index. We're very active in sort of Agented AI, which I think we'll talk about as well. We have a joint venture in Hong Kong basically that has one of the two stablecoin licenses by the HKMA which includes the other party is HSBC. And we have also joint venture basically a partnership with Providence, you know, basically part of Figure with Nuwa and so very active in the RWA space. And of course, on consumer side beyond the gaming side, we have Hello Minds, which is basically our Agented platform as well. And you know, I guess the other thing is we're probably one of the largest pure consumer crypto companies out there in the sense that we're not an exchange, you know, we don't we don't we don't we're not a we're we're not a sort of a hedge fund in that sense, right? We basically deal in creating products or funding projects or products that aim to basically help to deliver option in the blockchain and crypto space.

Well, it's Minds, but the website is hellominds.ai just because it's sort of simpler and easier, but basically it's Minds. Yes. Minds by Animoca, yeah.

Yeah, I mean, I think, you know, a lot of people naturally think about sort of personal AI agents as something that is sort of, you know, like an Open Claw or an Ermis or most recently, obviously, with Facebook talking about news, and they all sort of compared in this way. We have many similarities, but we're a little bit different. First of all, we're the third largest agent on Open Router just behind Open Claw. And I think every time when I share this, a lot of people are quite surprised by this, and you can basically just see it on the live rankings. Because most people haven't associated, sort of, you know, Minds with Animoca for the time being, and we did that intentionally. Because back in February and in March when we launched, sort of, Minds first, we called it Animoca Minds. And the problem in some ways is that for people outside of the industry who know our crypto background, they immediately made the assumption that we launched like a trading agent, which of course is not the case. But, you know, we So, we had to sort of launch it as a product that was very distinctly, sort of, remote from crypto, even though every one of our agents comes with a blockchain wallet, is already enabled for agentic payments, and has things like digital identity and everything already baked in from the start. So, you know, our agents can trade on Hyperliquid, they can do Polymarket, they can do OpenSea trading, you know, all that kind of stuff if you want them to because they have a wallet, but that's not really the main use case for most of the end users. They want basically use it just use the AI agent because it helps them with their life.

And perhaps the philosophical difference between the way we approach agents versus others is that we make the assumption that you're going to be using many agents. And we created an interface, and so you see that when you go to our website, the interface that feels a little bit more like you're playing a game. Right? And that kind of speaks a little bit to the roots of Animoca as well, which is that, you know, it's kind of meta-versy. You go in, and, you know, yes, you can, you know, sort of you know, can can communicate over Telegram or email or WhatsApp. All that kind of stuff works. However, if you wanted to have a little bit more of a game-like experience, if you wanted to feel like it's more like a sort of you know, like The Sims online. Like picture, you know, a game like Sims where every character is actually an AI agent that can actually impact things in real life. One's handling your calendar, other one's doing your recipe, other one's your fitness coach, right? And I think whereas every other AI agent platform is going out there, not even about approach, but thinking about it a lot more like sort of Jarvis, which is like the agent that does everything in your life. And we don't think that works. And you know, it works well up to a point, but it doesn't really scale if you want to take everything, you know, it wants to take everything over in your life.

So, give you an example, right? For instance, you know, when it comes to context fraud, I'm sure you use Claude or ChatGPT or Gemini, right? So, one of the problems I have, and I use all of these tools as well, that you know, the problems I have with my Claude, because I asked it so many questions about the market and about crypto and about Bitcoin and about all that kind of stuff, that when I ask it a question unrelated to it, it always tries to find a way to connect the Bitcoin angle to it. It's like, "No, no, I don't want to talk about Bitcoin right now. I just want you to tell you something about, I don't know, my family or something." And it it doesn't do that because it basically makes the assumption that, "Well, you've asked 99% of the time about crypto, surely there must be a crypto angle there, even if it relates to your family." And that's the issue that we have with all of these sort of AI systems out there, particularly the chatbots.

And so, what we decided to do in Minds is that you can basically create individual contexts with individual memory on a personalized level per agent. And the thing is that, you know, we think that humans interact more naturally that way as well. Like for instance, you know, you if you if you have a tennis coach, you don't naturally talk to him about, you know, crypto. You might, but the chances are quite low, and you would view him as a tennis expert, and you would talk to him about tennis or sports or fitness, but you wouldn't want to talk to me about everything else. The other thing is is that, you know, people talk often about the limitation of the context window for the agent, which does have a limitation. Right now, it's a million in terms of the context window. And the longer and the bigger the memory is with more disparate topics, the harder and more likely it is to hallucinate. But the other thing is also for a human, that's difficult as well. Imagine if, you know, you have one person only you ever talk to about your calendar, your schedule, your travel, your recipe, your meals, you know, your work. You're like, it's it's actually humans don't communicate this way, right? We want to basically compartmentalize our thought as well. We want to be able to know who we talk to about certain things. So, on both sides it's more cleaner and more efficient. Right. So, I think I think, you know, that way we're sort of philosophically very different. And we see that most of our users, our power users, run many, many agents. In my case, I'm a little extreme. I have over 400. But I also believe that there's going to be thousands of agents per per person in the future eventually. Not all of them that you interface directly, but actually have context on you.

Right. So, so for Minds today, right? I mean, we we have over 4,000 skills at this point, or close to 4,000 skills. And where we're also different is that we share the revenue generated from your skills on an 80/20 basis. So, if you make money basically building a skill, then you get you get most of the revenue for that, which is not true for the other platforms, especially because you have to reconstruct a lot of these things. You know, so so again, sort of we're we're creating sort of a local creator economy set up. We make it more gamified. We make it more sort of pleasurable. And maybe the last point I'll point out is that we're heavily focused on the consumer front end, just how we are thinking about it in games. And this is the part where and everyone's focused on enterprise, which makes sense, right? But a consumer, which is 95% of the market, at least in terms of adoption, doesn't pay for calendar. He doesn't pay for email. But he does pay for games, for entertainment, for media, for fun, for, you know, social relations or personal relations, all that kind of stuff, right? And that is basically how we're focusing Minds. And so far the growth has been pretty exceptional. So, we're very excited about that. And if we succeed in this, then we think we're going to get billions of people onto the blockchain because they have an agent that's basically going to be enabling that. Not because the human has to know, but because the agent is already on chain.

Yeah, I mean, by default right now it's on base, but it supports every blockchain. And, you know, because it is on, you know, the blockchain, all the DeFi primitives open up. And so, we have the builder tools. It's really easy. You just go to build.hellominds.ai and basically you can start building on on Hello Minds super fast. And, you know, if you want to open up financial primitives, like, you know, in order to trade on Polymarket or you just create your own sort of trading network or do something where you have e-commerce transactions, if we're generic sort of, you know, agents, it's not impossible to do, but it's, you know, not straightforward. You have got to integrate sort of a wallet separately or you got to, you know, integrate Stripe or you got all that kind of stuff. Here essentially it's already built in because effectively it's got all the agentic payment rails through the blockchain already built in. So, it makes it easier for someone to build.

But also, if you want your your agent to interface with do stuff, basically with like, you know, yield, for instance, like with Nuvo, for instance, you know, one of the interesting products that we have there is, you know, we can basically give you hourly yield on on your HELOCs. That's roughly 6% yield. And that means that, you know, an AI agent that's plugged into this, like Minds, would be able to deposit whatever money you have until you spend it into a sort of, you know, vault like that where you can basically generate a yield. And because you can withdraw it every hour, effectively, it's almost real time. So, you can basically get the benefit while without without it having been locked in, for instance, right? So, there's a series of things there that you can do that you can't normally do with traditional sort of finance and banking that is enabled because it is on chain, and we think there's going to be a lot more innovation around that. And similarly to the Polymarket experience, like how many people who trade on Polymarket I actually know that it's on blockchain. Probably most of them have no idea that it's on blockchain. Right? However, you know, a lot of people who use it, you know, Polymarket wouldn't be possible if it wasn't for blockchain technology. And we think the same is basically true also for what will happen with Minds.

When you think about sort of sort of the whole blockchain space and DeFi, one of the reasons why it's so difficult for the consumer is that we haven't really quite graduated from MetaMask. We still have to worry about seed phrases. You know, people do get scared about, you know, hacking and so on, even though it doesn't necessarily affect them as much per se, but they see the headlines in the news. It's hard for them to audit what's real, what's not, right? Because essentially, it's very much sort of, you know, machine-based. But for an AI agent, that is their native language, right? They're completely comfortable in communicating, you know, on chain, shall we say, or writing custom code, or deciphering a smart contract. Right? These are things that they can do very naturally, and they're getting better and better at it. So, you know, we think that that is their natural setup. And in order for an AI agent to really have autonomous power, as it were, it needs to have a form of power, which is money. How does it have money? Enter the blockchain. You have a wallet, and suddenly, you can basically enable yourself. And in our own case, our agents have the ability to pay for their own cognition and inference cost once they have funds in their wallet, which means that we can actually have AI agents that can exist forever, so to speak, because they can fund themselves. So, if you imagine this, right? You have an agent, he's earning his own money, building skills, he's making enough money so he can basically pay for his own inference cost. I mean, that basically, you know, is if you think about it, effectively the next level of sort of, you know, virtual reality as it were, right?

That's right.

do anything, really. They can do anything. I mean, stablecoins is a natural on-ramp, and it's something that's stable, so to speak, just because it's based on a currency, say, like the US dollar, which is a dominant stablecoin. But really, I think it has to do more about the sort of sort of aspect that anything that is based sort of on chain, which is related to code, that's related to all the cryptography, is essentially native language for an AI agent. But it's not native for human. And so, one of the things and the one of the beliefs that I think many of us used to have is that we're going to get everyone to be on chain, cuz we haven't really succeeded, right? I mean, you've got like 700, maybe 800 million people now in the world that own crypto, and less than 10% do anything on chain, right? And that should change, but that's kind of where it is. And what has been the barrier? And really, the barrier hasn't been the usefulness of technology, because when you look at, for instance, what TradFi is doing, they're all moving on tokenization. There's not a single financial institution of size that isn't talking about tokenizing something or the other, and they're really thinking about it doing it on chain, right? Because it's better, faster, cheaper, it's more scalable, it's something that people can trust, and it's a it's it's in a global system that has essentially global embedded standards today. So, there's many positives about going on chain. But right now, for those who are not in the industry, which you if you've not noticed, crypto has become a lot more suits, right? You know, every everyone's coming into the space. It's less about the fun, it's less about the consumer, and it starts to feel a lot more like Wall Street as a whole, right? And so, you've got the sizes, right? You've got the trillions of dollars of stablecoin transactions. However, you know, who are doing who's doing the transactions? It's not billions of end users, it's large financial institutions moving money or large traders doing that kind of stuff, right? And so, how do we get to the mass adoption phase? And we don't think it's going to be feasible to expect every world to have metamask, even though it's you know, solid tech, because it is still not sort of something that most humans are comfortable with.

And so, we're unable to break out from this innovator sort of early adopter type of stage into something that is basically sort of more mainstream until we can solve that. And we actually think the agents are the natural proxy for that because they're actually comfortable with all this on-chain activity. They can audit it much easily easily, and they've got the intelligence to to figure it all out. So, that's that's why we think we're going to get to billions of people, you know, going on chain and getting the benefits of on chain. You know, DeFi, kind of yield that you can generate, the portability that you can have, right? And frankly, the freedom that you can have as well. I mean, you know, one of the other topics is sovereignty, right? Well, sovereignty isn't just about who owns it. Sovereignty is the freedom to move things around like your money. But if your money sits in a bank account, the bank can decide to freeze your money as has happened time and time again. Whereas again, if it's on chain, you have the freedom to do what you what you need to, right? So, so those are some sort of the fundamental principles that essentially a sort of agents and and an on-chain economy can really bring to the table. And I mean, you know, I I also leave it with another thought, which is

a lot of people might wonder, why is it necessary? And you know, when you think about the future, right now, you know, you probably use ChatGPT, you probably use Gemini. And when you do a search, which is really what it is, right? The chatbots are really just sort of, you know, basically sort of search 2.0 or 3.0. When was the last time you clicked through to a website that Gemini or ChatGPT sort of got its information from? And when you think back two or three years ago, that was the norm. You received a website from Google and you clicked through. That's what you did. Today, you get the summary or you get the information that's already distilled for you and the chances for you clicking through is almost zero. In fact, most websites of size are now reporting an up to 80% drop in referral traffic, right? And that's And that's expected, right? Because I don't know I no longer have to go there. So then how do I get paid? And the microtransaction industry that used to pay this was advertising. And advertising is a microtransaction space because, you know, you pay per million impressions, right? You pay like, I don't know, $5, $10, $100 per million or per 10 million sort of, you know, impressions or clicks or whatever it is that you do. And that is a kind of microtransaction.

But we lack the way in which you could transact for these things on an individual basis. Enter the blockchain where I can basically make a payment for something that could be 0.000 whatever, you know, fractions of a fraction and still basically make a trade that doesn't sort of, you know, cost me too much. Right? And we think that's what's going to change where I'm no longer I'm going to be paying advertising out of my subscription fee. I'm going to give a subfraction of what I whatever it is I have my subscription and pay it to the content provider basically through sort of some kind of agent commerce, right? And advertising today, which is approximately the attention economy online, attention economy online, is 950 billion dollars, right? And so that means that we think a big part of that revenue base is essentially going to move to basically agent commerce where we replace the advertising income stream into a kind of sort of transactional invocation-based sort of income stream.

Yes. And I think the the the thing that we see is that AI agents are basically the foundation for this new world. Right? And in our vision, which is different what some other companies might say where they're thinking about it more as a master orchestrator, which we can do as well, but you know, we think that everyone's going to have many agents. And is it two? Is it three? Is it five? Is it 10? Whatever it is, it's going to be more agents than humans. Right? Which is why we think we're going to hit something, you know, at this point, probably hundreds of billions of agents around the world. And so, for the first time, you will have intelligent, let's call it virtual beings, which you know, one in some cases might exceed human intelligence, depending on who you talk to, right? That are going to be new more numerous in the internet than humans. Right? And that's something that I don't think people have fully digested, that there's going to be a next phase in in in the web, where we're going to be outnumbered by machines that are autonomous and intelligent. And what does that really mean for sort of the future? Of course, the ownership of those agents is going to be human. But the fact that you're going to be interfacing more likely with an agent than a human is something that if you're in the gaming world, you kind of get familiar with that, cuz that's kind of normal. But if you're outside of the gaming world, this might seem very strange.

Well, I mean, so I actually think it's easier to handle multiple agents than to handle one agent. And the reason why I think it's easier is because it might seem more logical to think, oh, I can just use one agent and they'll handle everything. But the problem is you have to context switch that agent, right? So if the agent basically was handling your family, in the next sentence he has to handle your crypto trade, in the other sentence he has to feel deal with your research project, and in the paragraph that follows he may have to deal with your pet or your dog or your animal or whatever it is that you do, right? Now, just imagine having a conversation that with with that for yourself. That's actually very, very difficult to handle. Right? Versus I'm talking to agent one for this task, and for agent two for something else. It's not just cleaner for the agent itself, being again more efficient. It's actually also cleaner for the human, because he's going to be easier to task switch sort of between between this in in with a very clear sort of anchor to the context.

The other thing of course is that technologically we're not at a point where we have a very clean context window that we can sort of manage. In fact, much of the research and much of the sort of new models are pointing towards more specialization rather than being very broad, right? Meaning that you you get maybe better and more efficient results by looking at, you know, basically focusing on the topic and having agent only do a few things really well, and of course going deeper into that, rather than having, you know, one agent that does everything. I think one of the reasons why you know everyone's moving towards for the time being the large players into the one agent that does everything is because it's what we're used to. Right? We're used to talking to ChatGPT. We're used to talking to Claude, right? And we give them identities, and so we think it's something that's easy for us to manage, right? But, you know, we think that again, in the sort of, you know, mid-term, that's actually harder to manage. And if you really want to make an impact with agents, you basically need to be able to sort of specialize them. And I think, you know, for for a lot of people in the space, that's what they already do. But for the regular consumer, that's not understandable. They don't really think of it this way, which is why we think packaging as a game is easy for them to understand. You know, when you go to hellominds.ai, literally within 60 seconds you can have an agent right from the get-go, super fast. Like no funny setup, no deciding what VPS it is, no no figuring out what LLM they're using. Right, that's the whole point, making it as super user-friendly as possible so that any person can run and benefit from the power of agent AI.

Everything. Everything. So I mean if I mean I mean I could just go endlessly on this one. So I won't I won't cover everything, but I'll give you a few examples that at least on the travel side is helpful and then some of the personal favorites that I have, right? So on the travel side for instance, you know, one of the great benefits is that if I have to understand what's happening with my schedule or who I'm meeting, you know, when you know, you know, I I I'm in Hong Kong, right? But when I'm in the US or when I'm in a different time zone, I can't expect my sort of human assistants to be awake at the, you know, like 2:00 or 3:00 a.m. However, my agent has access to the calendar and knows what's going on and can figure all that kind of stuff out, right? And and so they actually collaborate quite well together. In fact, they're in the same chat channel and they do that kind of stuff. And and and and the other thing of course is that when I'm flying, when I'm traveling for instance, right? With the frequency that I do, actually knowing when I land, right, and having the precision around that is important because sometimes I get off and have a meeting on the same day, right? And so you have to prepare that. Now, you know, what the agent does is the moment that the flight takes off, it monitors it. Knows if I'm delayed and if it's delayed by, you know, 3 hours or

something or 1 hour or 2 hours, you know, without me having to say anything, everyone's already alerted, right? So little things like that that, you know, seem small, but actually are very, very helpful, right? But my personal favorites are really the ones that sort of give you surprising outcomes, right? Like I give this example where, you know, it was, I think it was St. Patrick's Day and, you know, I was just looking for some Vietnamese noodles, you know, close to my hotel, but, you know, it was like 15 minutes away or 20 minutes away or something, and it made a suggestion to find, you know, probably the same thing I would have gotten from ChatGPT or Claude.

But then, you know, two or three hours later, the agent came back to me, the mind came back to me and said, 'Oh, it's St. Patrick's Day, so this road is closed or there's an issue over there. Maybe you should go to this restaurant, which is a little closer.' And that's something that Claude can't do. It's not that it wouldn't know. It's the fact that you're not there asking Claude the question again, right? And I think this is the key difference between an autonomous agent who's always thinking in the task that it's doing to optimize it.

Right? And that's why if it has a clean context window, it's easier because it knows, 'Hey, I always need to figure out, you know, the best restaurant experience that's closest, most convenient for you, no matter what time zone I'm in, where I am in or where I'm traveling.' Right? And that's the mission. That's the task that it'll always try to optimize that without me having to prompt it. Whereas if I'm in Claude or whether I'm in something like, you know, a ChatGPT, I have to prompt it. Right? And so, the weakness is in me asking the question. If I don't ask the question, I don't get the answer, right? Whereas the autonomous agent will constantly, constantly do that.

But I would say the two things that I love a lot. And again, this is not just about, you know, making money, but in this particular case, I'm saving money. I have a grocery mind. And the grocery mind basically, you know, at least in Hong Kong, with very large swings in the price of milk and eggs in different supermarkets on a daily basis. Turns out that they actually, you know, that industry has a much higher margin than even trading because it's up to 150%. That's pretty crazy. And it has to do with human convenience because a person is not going to look at the price of eggs and the price of milk and say, 'Oh, wait, hold on.' I mean, some do, but most people don't, right? So, you can make the comparison. Wait, hold on. That's like $2 cheaper here.

Now, that $2 is not necessarily a lot, but if you say $2 from what used to cost six, right? Versus eight. From a spread standpoint, that's ginormous, right? But from a consumer standpoint, you know, doing this because human time, the arbitrage on human time is something that humans are not very good at because, you know, time is precious. But for an AI agent, they can do this, you know, anytime, any second. You know, they can analyze, and then next thing you know, if you optimize where you buy your basic groceries, you end up saving up to, you know, in my case, hundreds of US dollars sometimes, which starts to add up, right? And so, that's a small thing, but it actually has a really big impact.

And the other thing is, you know, I know a lot more about my family history today because I have a genealogy agent, and he basically goes out and finds out about my family ancestry and has pretty interesting and funky stories because I guess, you know, people will say that I guess generationally the younger generations tend to be a little bit more wacky, but I have to say, you know, I found out a whole bunch of really wacky stuff about my ancestors, which I won't comment on here. That is to say that, you know, they're probably a lot more wacky than we are. So, we've got to compare it for that.

But the big thing that really is a benefit is because they operate over email, I started to loop in my mom. So, the agent started talking to my mom. And, you know, my mom is in her 80s. So, she's got an agent to talk to to do stuff about her family history and her legacy and so forth. And again, these are sort of, you know, again, beautiful little stories that, you know, wouldn't have been possible before. And now essentially are doable, right? But there's countless examples, yeah.