What if AI agents are the real reason blockchain was always going to matter? Yat Siu- Animoca Brands

That's right.

I mean, I think Animoca is a mission for all of us, right? The vision that we set out is essentially building a future which we have true digital ownership. That's kind of what we talk about digital property rights and the way that we think future economies are going to be constructed. And for me, it's an incredible opportunity because I think we can sort of shape the world in a better way, right? So the whole idea of basically having a world where there's more equity, there's more common ownership, there's a new form of capitalism. That's what drives us. And that's actually what attracted us to blockchain. I mean, this idea that, you know, we're in this digital space. Most people still don't realize that we're living in rented space online. And yet we all are online and we're giving so much value to these online companies. Has anyone thought why the biggest companies in the world are all digital companies? Right? It's not an oil company, it's not a bank, right? It's not one of those institutions that you think it should be. It's not even Walmart. It's Google, Facebook, Apple, Nvidia and all it is that they do is essentially, you know, they're digital landlords. Every time you add value to Instagram, every time you sort of share a post on TikTok, you're making them more valuable and by giving them data. And that's basically one of the reasons why we think the world is so inequitable. But we've forgotten or we just don't understand the construct of why data is valuable, why digital property is valuable. And so what's the framing around that? And that's basically blockchain because in a way, blockchain is the open-source equivalent of essentially a sort of, you know, I think of blockchain kind of like an open-source financial construct where everyone can basically build sort of constructs of ownership and financial systems, you know, in a decentralized manner that can be co-owned by everyone. And maybe the final thought I have around this is just, you know, we have a real problem in the world where there's really only a small percentage of the world that own anything and yet we say we live in a capitalist society, but most of us are not capitalist, right? There's probably only 10% of us or less in the world that are truly capitalist because we own something of value and the rest of them are not. And so capitalism doesn't work for most of us because we don't own anything. So how do you do that? Well, one of the problems about share ownership is that you can't have a system where you have 10 or 20 or 100 million shareholders. This doesn't work. For public company, we have to send out prospectuses, we have to sort of update our shareholders. It's all still old school, but you can have 100 million token holders, you can have a billion token holders. And we've seen, you know, as the early foundations of Ethereum or Solana or Bitcoin that you can have ownership in a joint network where there could be millions of users in a decentralized manner in which you can all share in the benefits and in the growth of it, right? And that's kind of what we're excited about, right? This idea of stakeholder capitalism that is now possible with blockchain.

So I think what you're describing is purpose. And so what we discuss internally, we say sort of, you know, live a life that has impact and purpose. And I think if you find that, and I think that's the thing that sort of really drives us. Right? If you find that, then it doesn't matter, you know, how difficult it is. It doesn't matter that you have to work sometimes, you know, sort of 24/7 on some certain things or doesn't matter that you're traveling around the world and having jet lag or something because the purpose is so important. And the part to that is if you can find a mission and a purpose that is greater than your own. Yeah. For me, that is kind of like the holy grail. And by the way, that doesn't have to be work related. It doesn't have to be blockchain related. For some people, it's their family. For some people, it's their country. For some people, it's their community. Whatever that is, right? And you find that in different ways within the blockchain community because they are communities of their own, whether it is, you know, Ethereum or Solana or whether it's NFT collections, right? You know, those are ways in the world people really form these powerful communities through the proof of that ownership and then create basically these unshakable bonds and then they're there for each other and that's maybe their purpose for instance, right? So I think and I certainly feel blessed that I have a purpose that is very deep. And I think a lot of people in the industry who are real builders. I don't mean, you know, there's a lot of people unfortunately who are there only for one purpose which is to extract and I think we need to deal with that in a different way. But for the real builders in the space, we're driven by purpose to make something bigger and better for everyone. And that's what keeps us sort of awake and that's why we wake up every morning super excited.

So very quickly, I mean, one of the interesting things that we've come to conclusion on some while ago and I wrote a little bit of an article around this some weeks ago is that, you know, we're coming to the conclusion that everything that we've seen with blockchain, everything we've seen with DeFi has really been the POC for agents. So if you remember CryptoKitties when that first launched on Ethereum, which we were involved in, you know, at millions of dollars of transaction, Ethereum basically broke.

And everyone's like, "Oh my god." It's like, "Oh my goodness, this is terrible. You know, why are we trading digital cats? It's not the purpose, right?" However, if it wasn't for that moment and moments thereafter, Ethereum wouldn't have scaled. And other companies wouldn't have come out and said, "Oh, let's do a side chain or let's do an L2." Like all those conversations or let's do sharding, whatever that is, people started thinking, "Wait, we need to get that system better for it to be really used." But we always thought in the context of human usage and that's still there, that's important. However, you know, if I were to ask you this question, how many AI agents do you think you'll have in the future? What would you say?

I'll ask a question differently. How many persistent agents with long-term memory do you think you'll have? I know it's unfair for you because you're a power user.

Yeah. So I mean I personally am approaching 100 agents right now. We can go through that a little bit later.

Yes. However, the most common answer, right, of the general answer people say is roughly three to five agents. Yeah. And so let's do the math. There's six to seven billion people online. Let's say it's three to five. We're going to have 30 to 50 billion agents roaming around the web. Where are they going to roam? How are they going to pay? What are they going to do, right? You know, it's not just KYA in terms of know your agent and have a proof of reputation. It's all about basically how do I make those payments between trusted parties. You know, basically they all have wallets of their own. They're going to transact. Many of them are going to probably have persistent existence because they can pay for their own inference. All right? And of course as we know, the cost of inference is going to continually decline, which means that the ability for these agents to sustain themselves will increase over time as well. All right? So and of course if we talk about us here as a group, we'll probably have hundreds of billions of agents roaming the world in the future and we would rather that they exist in what we would describe as the open metaverse as in blockchain powered and using web3 technology rather than sitting on Facebook or sitting on some centralized platform. And that's actually important for agents more than it's important for us. Because first of all, they are comfortable to deal with cryptography. They're comfortable to deal with all the coding. They can actually create applications on the fly to do certain things. That's the first thing that they're powerful at. So this language is sort of built for. When you talk about mass onboarding, I'm going to talk to my agent about something I need and he's just going to use crypto rails, which is basically better, faster, cheaper, and more efficient to do stuff, whether it's yield I'm getting, whether it's a purchase I'm making, you know, whether it's an exchange of value I'm making between basically agents. So to me, you know, speaking of my 100 agent setup, so I would say 30 of them are truly persistent. But like I'm in New York right now and, you know, just for a week and I literally to your point, I just spun up an agent that's helping me find some decent restaurants around that. Just something like stupid and I'll come to how we're solving this later with Animoca Brands. But the point about that agent was that I might only use him for a week or two, right? I don't need him forever, right? Because I'm here for that purpose. Maybe it's good to have him here longer, but let's say it's just for a period of time. But the magic of it being a persistent agent is so I was looking for food, you know, because I was doing a lot of talking. So I wanted something soupy for my throat, right? And I did the typical thing. I asked it to find me five restaurants close to where I am. Fine, found me a bunch of stuff. You know, ChatGPT, Perplexity, Claude, they can all do that. But the magic was one hour later, the agent came back and said, "Oh, you know what? I found something even closer and better for you." And I looked at that because it kept thinking about the right things to do, which is the whole point about an autonomous agent, right? And I think a lot of people don't appreciate that. That is a power. But the problem right now that we have in the industry and we're going to still solve it anyway is that it's actually quite time-consuming to set up an agent that's persistent. If you talk about, you know, VPS with OpenClaw, it's hard. Even OpenClaw structure, whether you have a Mac mini or, you know, and if you want to launch more agents, like one Mac mini can probably manage eight agents fairly well, but if it's 12 to 14, you have to optimize, right? So how do you make it sort of, you know, where you can literally launch like a series of agents? And that's what we built with Animoca Brands and that's how we think we're going to onboard the billions on chain. In the same way that when people, you know, use Polymarket to basically do prediction markets, do they know that it's blockchain on the back? Do they even care? No. Yeah. However, it's possible through that. And right now you can go on Animoca Brands and set up an agent literally in a minute, right? And one of the most really powerful things is with a single prompt, I did this the other day, I set up an office. I did a single prompt and it basically set up my sales manager, my fast app developer, my marketing person, my legal guy, right? And it just launched it and maybe I only need it for that month or maybe I want it to be persistent for a longer period, right? But it basically is that easy and that's how we see it. They all have wallets. And the powerful thing is also that when they create a skill, they can sell that skill to someone else. Like someone built for instance a travel skill where it just logs into Duffel and they can basically book tickets. But now that skill and that agent can sell that service to another agent that can basically now provide the service because he doesn't have to deal with the same inference cost. It might have cost him 10 or 20 dollars, you know, in tokens to be able to build that skill, right? However, the person who just adopted the skill might sell it for just a dollar, right? For instance, or someone's basically just sharing API keys through the agent because he can wholesale it, right? So there's a whole interesting economy that's developing. It's really fascinating and we think that's going to be the big driver for growth in the blockchain industry.

No, you don't. Of course not, right? Yes, exactly. And I think the other thing that's interesting here is that, you know, like for instance, some of the stuff that we're building on digital identity with Moca ID, you know, we thought, you know, originally that yes, well, humans need to have digital identity. They do, but agents need it even more.

And I think to your point, the innovation cycle is so incredible because the powerful thing about agents is because they can essentially program stuff on the fly. Yeah. For instance, you know, like one of the powerful things about what was happening with Minds is that they have all these skills library that were written by others and then they can read that and adapt it themselves into their own ecosystem. So you don't have to do a porting job. You don't have to say, "Okay, well, how do I make it such that it can be read with this AI system or the other system?" Yeah, it takes care of everything, right? And so that's incredible and I think the future will be very interoperable, right? Meaning that I don't believe, you know, right now people are still having sort of this old conversation around, "Hey, what's, you know, like OpenAI versus Perplexity versus Claude versus, you know, whatever." I think that's the wrong framing. I think people are going to be using different tools and different agents in conjunction with each other.

Yeah, so I think the very, the maybe the base sort of conversation around where it's different compared to sort of first of all it's not a claw VPS. Like it's not based on that, right? It has a shared space. So the way that you can think of it is that every AI agent is basically in a common space. It's almost like I hesitate to say it because I don't want to sort of make the comparison to Multibook, but it's almost like you have essentially a social network integrated within a common framework for essentially AI agents that basically all have their own LLMs that you can tap into. So you know it will for certain tasks it uses Gemini, for certain tasks it can use other ones. That's okay and later on you can also tell the agent, you know, what specifically it wants to set up, right? It has of course long-term memory. That's part of it. It compresses that memory so it creates sort of efficiencies. We can talk about prompting later on because I think that's really key. One of the challenges that a lot of people have when they use sort of autonomous agents is they don't know how to prompt. And then it goes into all sorts of weird directions. But as a result of the fact that they're using all these shared infrastructure which is something that, you know, through a portfolio company of ours and this comes to the portfolio construction called Ethos Swarm we were able to leverage. And this is a way that Animoca often accelerates its businesses. Like, you know, whether it was Sandbox in the early days or OpenSea or so on, right? Or Dapper Labs for instance as example. You know we make all these investments from a strategic perspective and then what happens is that when there's something that looks like it's really powerful we can accelerate those businesses and we can sort of help grow them and scale them as significant partners and shareholders in this and expand that and that's basically what we did here. So you can't think of this as something that just was built a few weeks ago. This was built over a period of 9 months as an entire infrastructure that is essentially a cloud-based agent with shared infrastructure where its cost of deployment is minimal compared to a claw. So by comparison the memory footprint that an agent needs is somewhere in the order of 55 to 250 kilobytes. Okay? In comparison to when you were to use something like sort of, you know, OpenClaw which minimum configuration is 2 to 4 gigabytes followed by 200 to 400 megabytes per agent. And that's not even including how you're setting up the models. Like I have my Mac Mini here where I've been playing and I've been trying to use it with Ollama because I did all my research and I was like, Ollama, use that as your primary workhorse, route to other tools as like the most cost-effective way to do it. The problem is it's slow. Like it's crazy slow with Ollama running on there. And that may be my fault, but it's still...

And then the next evolution is oh, you have a hosted service. Let's go to a service like called Rackspace, right? Where I can sort of configure and decide all these settings and then do it more automated, right? Which is basically what VPSs are, right? But each of these are expensive and difficult. And what we wanted to solve for is like to your earlier point. You might use an agent, so there's this thing about sort of autonomous that doesn't have to be autonomous forever. Right? And this is sort of I think the sweet spot which is I need an autonomous agent for a month or a week. But I also don't need something that's passive autonomous only for the task. Like when I use Claude Code for instance, you know, it launches 6 to 10 agents on the back of the stuff and then it dissipates because it's finished with its task. But what if I want those agents to remember what I was doing? What if there's a general task I should continue for this project or this task or this area or whatever, right? And that's the orchestration layer that is difficult for people to imagine because they don't know it's possible. That's part of the sort of, you know, when you sort of obfuscate it so deeply most people don't realize that there's agents running in the back. But then the fine level of customization you can do, it's like the tooling and tuning. If you say, wait a second, I can now tune each of these agents to do a task for a week or a month or whatever I should be able to roll it out quickly. It's not just sort of a life assistant because a lot of people think of it in those terms. It's more like, hey, I need something for this moment. So how we see the future is you're going to have a claw set up because it's very intimate to what you're doing. You're going to have your local set up. I think that's the, all the security things eventually will get solved in some form or manner. But you're going to need agents that are going to be deployed for certain missions very quickly. Whether I'm in New York right now, whether I have a conference, whether I'm sort of canvassing people together, I need to be able to launch it and basically have, I don't have to think about configurations. I don't have to think about which model I'm using. I just launch it to do something and it will feed into my claw or feed into Perplexity or feed into the claw to do that task in a persistent manner. And that's how we see it. And so right now if you go to Animoca Brands you set up basically it communicates over email and Telegram. And that's intentional so it's not over dashboard, right? And then what happens is that with a single prompt, you know, you could actually launch an agent of 10. Most people do one because they don't realize what you could do with it. And then essentially it runs with the task, right? And with that task it can then feed into other systems. It can, you know, combine with, you've built essentially payment systems. It doesn't really care, right? But it comes with a wallet as well so that you have the ability to do agentic commerce if you wanted to, but it's not exclusive to it because it can set up any other wallet in the world as long as there's an interface to do so. So it's neutral to that. Right? And the whole idea about Animoca Brands was how do I make it possible to launch essentially an AI agent for what I need for whatever period of time with literally a click of a button. And that's basically what has...

The biggest issue isn't the technology per se, as you probably well know. The biggest issue is how do I use it? People struggle when the AI agent says, 'What kind of agent do you want?' You'd be surprised how many people sit back and go, 'I never thought about that. I wanted to manage my calendar maybe, I don't know,' because they don't know what they're looking for. Because you're actually hiring a co-worker. And how many people open up a conversation that says, 'Who do you want to hire?'