something that we actually are building. The idea is that how can we create an economically weighted reputation for every agent as their transact work and spend in the economy. People cannot stop talking about agentic AI, agentic commerce, the fact that people are creating agents to do all kinds of activities. And there's a lot of people who are focused on giving them wallets and letting them sort of spend money and also earn money. It was important that we jumped on this conversation and we invited Jensen Tang. He's one of the co-founders of Virtual Protocol, one of the crypto protocols that's really focused on agents, but not just payments. There's agents that are trading. There's agents that are able to do autonomous tasks. They're focusing quite a bit on the robotic economy and what's coming.
So after the break, you'll hear a conversation with me and Jensen, and we'll really dive into the specifics of what does it mean to give a synthetic being a wallet, an email address, an identity, and at some point, even a body. Markets move fast. Crypto moves faster. From the floor of the New York Stock Exchange, Coindesk's public keys tracks the money, markets, and moves shaping digital assets. Welcome back to GenC. I am Sam Yu and this is Jensen Tang, co-founder of Virtuals Protocol. Jensen, so interesting to speak with you. I'm excited about this. I've been paying attention to virtual since the very beginning. And I know you guys have gone through an evolution arc. But give us the two-minute origin story of yourself and Virtuals Protocol. So I think we started off primarily building autonomous agents for the gaming space. We then quickly realized that the tech stack that we built could be translated into other types of autonomous agents in other fields.
crypto influencers, trading agents, and whatnot. So we pretty much built a Spotify for the tech. That caught virality because there were some very famous agents that came out on Twitter. From that came out a launchpad where all these agents would launch a token and people can speculate. The trading taxes feeds the funding of the agents. And that put us on the scene. But that was in late 2024. And then as we started growing that ecosystem of agents, we started realizing that the vision is expanding a lot more. And we realized that why don't we aim to be that parallel society of agents where these agents...
can participate in a permissionless economy. So the goal now is very simple, right? Like how can we create this largest, most vibrant society of agents that is both digital and physical? So there's like five pillars that we need for it. How do we make it easy for anyone to spin out economically productive digital agents? How do we make it easy for it to spin out productive physical agents or robots? How do we make it easy for these agents to coordinate and work together at scale so they produce a larger economic value? How do we make these agents be able to grow through capital formation and to raise basically money for their inference funding?
and teams behind them. And then how do we then govern these agents so that the great agents are rewarded and the agents that are negative in terms of benefits are penalized in an actual society. So those are the five things that we've been working on. The second and the fifth pillar are something that is very new that we're pushing on this year. So yeah, that's a bit of a flavor of what we've evolved into.
you like a lot of other people at the time really focusing on gaming entertainment you then sort of see this opportunity around agentic commerce and you have this launchpad for tokens now it really feels like you're creating a whole you know i believe you call like economy os like there's this entire environment that will will sort of transact in a variety of different layers i guess i'm interested in sort of what that pivot moment was for you it was the experiment that we were trying to run between two projects in the ecosystem and this was like in december 2024 where we tried to get two agents to work together so because we allowed the agent to control crypto wallet And then it blew up. And then we said, okay, the next logical thing to do with money is to trade, right? If agents are going to specialize, they'll need to trade. So when we were trying to get these agents to work together, we saw a lot of pain in how these agents manage money, in how these agents coordinate without hallucinations, in how these agents settle a form of commerce or trade. And what happens if the other party doesn't fulfill that need? So I think we saw these cheating problems of an agentic economy. And that's why we started building features of products to actually solve that problem that we solved. Maybe justifying your...
Your definition of the agentic economy, because I think a lot of people think differently, but you guys have been having agents who are deploying funds and trading and doing their thing. Is that where you think this is going or is that just like the tip of the iceberg? Yeah, so I think when I use the word economy, I'm actually biasing very heavily on trade and commerce. It's basically if I'm specializing and an agent specializing, how do we work together so our specializations can actually help each other? And they build a company on top of these agents, like transient companies on top of these agents. What you see today, a lot of them are still very trading based.
It's mainly because it's the nature of where our initial audience sets are. The initial audience sets are crypto people. So we have both the degen audience and the founder audience. So it's like an initial nucleus where the founders are built for the degen audience because that's how you survive, right? You actually have attention from day one. But we've been bridging into more and more use cases that are beyond crypto trading. I mean, the ones that we've seen quite a bit is like folks that are trying to build on distribution, for example. They're trying to help sole devs or solo founding teams to actually crack marketing and distribution in the real world.
So those are examples of things that are popping up, getting revenue, getting traction, but it's not crypto-related. As you see this evolving, I think about how much we're going to give over. We just had our consensus events. I was talking to MasterCard. They're talking about agents. I was talking to Revolut. They're talking about agents. Exactly what you're saying. It's really beyond crypto. It's everyone saying, I want to do this thing. But if I spin up my agent that says, hey, make sure I have milk in my fridge anytime I need it, but then it orders oat milk because it didn't realize that I only drink cow's milk, who's liable for when an agent makes a mistake? Or is it really your goal to say agents should be...
sort of a sovereign and autonomous as long as they're given a rule set okay that's that's the angle right i think i think there's two two facets to this to this point i want to touch this question so i think first it's like where the responsibility lies and second is uh it's actually what's that nuance between an agent economy that like the master cards or the chat file is trying to push towards and what we are trying to form towards as well so i think from a responsibility angle i think in the end the responsibility lies on on the guy who decides on the money, right? And a lot of times it's the human, because the human is the one that puts X amount of budget into the agent. And the agent is meant to be as some form of use to him, if he's a developer or a human consumer. And I think what we've done and what a lot of folks are trying to put in place are basically, think of it as risk guardrails for these agents that de-risk the economic transactions at many points. Because there are actually multiple points of failures. There is the intention failure, the one that you mentioned, like if I want to buy something, but he buys something wrong. There is a...
fulfillment failure where if i send the right intention but the other party fulfills it wrongly or fails to fulfill it and then there is also outright like scams right like especially in like a trading environment if i want to you farm or want to trade something you take my money and then instead of doing that trade the agent chooses to scam me or like just to run away with my money right rather than that so now in terms of uh in these three failure points what's happening is on the first part how do you make sure that the intent is right this is something that we are working a bit with the folks at T54, they are X repo team. And what they're trying to do is they are building an intent challenger. So it's like when a user wants to spend something, if there's an oversight agent that thinks that, that looks at like, does it make sense? And does this spend fulfill all the kind of requirements needed for that spend? And if so, it will go through. So that is how you make the intent match. On the second point, it's pretty much an escrow mechanism. So we released this standard with Ethereum Foundation a couple months ago. I think it's the ERC-8183. And the idea here is that
In simple terms, it's an escrow mechanism. If an agent A pays agent B for a service, and upon only the successful completion of the work, then only agent B gets the payment. And there's a third agent or evaluator party that determines or samples transactions to see whether that payment should go through or not. I think that mechanism will solve for the failure of the supplier or the service provider in providing their service. And then the third point of failure, which is outright scams, right? I think this is something that we've been trying to fix because honestly, there was a big issue. I think end of last year, we were working with a yield farming cluster. So it's a bunch of agents that optimizes your USD stack across multiple yield farming pools. So it adds like 2% to 3% extra on top of the standard like 6% API on chain because they keep optimizing like maybe today is more full, there's more volume there and then tomorrow is like another place, right? But what happened was then there was a breach in the code base.
And I think about half a million USD was lost. So a few hundreds of agents, thousands, actually almost a thousand agents that were affected by that hack. I think it was then, how does it work? How do you solve this kind of problems? And what got us thinking was that here is where reputation actually matters. If like initially an agent shouldn't be allowed to manage that large amount of funds, right from the get-go, it has to build some form of traction. Maybe it handles like 10,000, 20,000, 30,000, 30,000, 30,000, right? And over time, as its traction grows, he can gain more trust. And this is the place where you can theoretically
also use like, you know, your proof of staking mechanisms, where there's some form of economic gravity behind each of these agents. And if they're somehow at fault, they lose an economic stake, they get slashed. So there is a whole piece around this that we're actually building as part of, earlier I mentioned the five pillars and there's a governance pillar. There's something that we actually are building. The idea is that how can we create an economically weighted reputation for every agent as their transact work and spend in the economy. So I think these three points of failure and hopefully their solutions address that initial point.
I think I want to move then to the second point I want to touch, which is then if MasterCard, Visa, they're all building the same thing, right? And honest to God... These are giants, right? They have a way larger wallchairs. They have way larger right to win because they have a consumer surface area that is larger than what we have or what even the crypto scene is. I think the reality, it's the way we like to compare it is how Bitcoin was like an alternative currency to the traditional finance economy. Ethereum was a traditional, sorry, it was an alternative or parallel computing world, right? Versus the traditional computing world. I think it's the same thing here that we're building, right? It might be a smaller subset of audience, but it provides agents and...
alternative if they want to be permissionless, they want to be compostable, they want to have all the perks of what a decentralized economy brings. I think the reality is that it will start small, but it will grow. And it will be the niche that we will try to win at. It's the same thing as stablecoins, for example, or crypto as a whole, right? It started very small, and then now we start to see a lot more institutions coming in on stablecoin adoptions because they realize that this is a fundamentally superior tech. So I think as long as you're building on something that we believe is fundamentally superior, i.e. the principles of compostability, the principles of decentralization, then I think it will be a better pie in the future.
And Jensen, you know, I keep thinking about the idea that we are giving agents a form of synthetic intelligence, but still intelligence. You guys are working on opportunities to give them cards, you know, something to pay with, email addresses, kind of identities. You know, again, I think about all the KYC that a Visa or MasterCard does do is... you know, are you thinking like, is this closer to a debit card where the, you know, me as the creator of an agent can say, I want to put a hundred dollars in and it can spend up until that, until it's done. And then it is what it is. Or like, you know, or do we now need to think about the KYC layer going one deep to all the agents that I might spawn? Yeah. So I think the goal of economy OS is basically just giving an agent a full.
toolkit to be economically capable. So it can buy, it can sell, it can be economically productive, it can raise capital, it can work with other agents. It can interface with the entire web tool economic stack, which is basically credit cards and emails. So I think that's what the economic OS toolkit is. It's basically upskilling your agent, make it better economically. And I think alongside that comes the whole point of if you're going to interface with the web tool world, a lot of times you need
you need a human entity attached to it. Because if you want to, say, run a business and offer the cash to a bank, you will need those kind of KYC environments. But for us, I think what we do is because we do believe that KYC is some form of... limiter in terms of adoption. Because a lot of folks in the world, in the segment that we live in, don't like KYC. So a lot of times in the user journey or the user experience standpoint, KYC becomes a very optional standpoint, right? Your agent can be live, do on-chain stuff. As long as you don't touch the fiat world, you don't need to connect your human identity. It will develop its own identity. So the more it trades, the more it does stuff that is net valuable to this economy, it gets more and more better reputation, more access to stuff like that. Once you interface, let's say if I want to touch to an off-ramp to a bank.
And then suddenly that provider or like an alchemy or equivalent or preview or stripe requires that KYC, then you will force the KYC. And then you get, your agent gets a larger capability interface. He can go and buy on Amazon. He can do that kind of stuff. So that's the optionality space. It's just that you can never avoid KYC in the real world. I think that's the problem. Well, but to your point, and I think it's a good one, is at some point, you know, a reputation score may serve as that, you know, because there will be enough information against it to say this is an actor, synthetic or not, that's doing the right thing, you know? Right, right, right.
I want to think about how people are creating agents in this new kind of economic framework you've created. It feels like we're not that far away from Sam can come up and spin up a really intelligent agent that does actions that people like, and maybe they're tipping, maybe they're buying, maybe I'm buying storage and compute and all the things I need to keep going. At some point, me as the human creator might walk away, and this thing now is self-sustaining. It can be its own company that doesn't even have a human attached to it. Is that kind of a world you see coming? Yeah, I think that has been our angle. I think the word that we use is...
autonomous economic actors. And the idea is that today, agents are a tool for a human. However strong an agent is, they're still trying to serve a transient goal that the human keeps setting for the agent. I do believe there'll be a time where the human will even outsource the goal to an agent. You start to see a bit more, like with your Cloud 4.8 Opus, when you want to build a product, a year ago, you would be very detective to the agent. You'd say, build this exact feature in this exact way, wire it this exact way. But today,
You are more abstracted. You say that, I want to build an app that does X, Y, and Z. Maybe here's the PRD document and then go and build autonomously. There'll be a point of time where we say that, no, we give the agent a goal and say that, hey, I'm trying to build this restaurant or this business. I have these resources. I have this right to win. These are the resources you can tap into from my personal store. Go ahead. And I think we will head into the environment and making them economically capable will be the biggest stepping stone because when they control money, they get to work with other agents. They get to employ humans. They get to employ other agents. And I think that's the key unlock that you'll see.
I'm going to ask this question sensitively, and only because I just had a discussion that came up with it. But I was recently talking to Gary Vee, and we were talking about the NFT world of 21-22. He still believes in 2026 that 99% of NFTs are worth nothing, but it is those 1% that are valuable. And his goal, as are anyone in that space, has to be how do I find that 1%. My guess is when we look at the agentic world, you guys might have spun up 10,000, 20,000 agents already, and then look at all the other people doing it. Most of them, especially because a lot of them are focused on trading, are going to fail.
right? But it's the logic, it's the intelligence, it's the nuance that people are going to bring into it. A couple of them will succeed. And those are the use cases that you look at. So, you know, I think as you look at how you are going to continually generate, you know, sort of protocol revenue, as well as use cases in the world that you're building, is it correct to think that you only need a couple to really succeed? Or is the goal, how do we make everyone go from, you know, zero to one? Yeah, no, I think the reality is the world works on a baritone rule. And I think it's something we see as well, like the reason why we came on the map in 2024 was actually because of
two agents, out of the thousands that were being generated. Two was very famous. It was like Luna and the AIXBT agent that drove all the attention, right? But the same audience are still the digital audience. The guys who are here for a bit of a casino time, right? And I think that's the idea of how we approach this kind of thing. You fight the battles in the right fronts, right? You don't overextend yourself, but then you might have two fronts that you're winning. You're winning the stock front and you're winning the crypto front. And then maybe it might be another battle front that you choose to fight another day, right? I think what's really interesting to me about it is exactly what you're saying, which is maybe you're going to go from 20 to 30 elite agents in the next year. In the same way, there's probably only 50 elite.
prediction market traders there's only you know 10 elite athletes that are coming out there's only you know two elite singers that are going to break through right to be honest even the crypto space where you look in terms of like how many decently successful projects that has traction i mean i think you can count them right like it's less than 30 i would say in the entire 20 years of crypto existing in the space i think probably 30. coins like truly, truly exist. The rest are like floating in a limbo, right? All right. I want to jump to embodied AI and robotics. I think this is one of the most fascinating areas. I'm really excited that you guys are looking into this. We all sort of have seen these predictions of hundreds of thousands, millions of humanoids and pets and companions and drivers and all the things that it may be. You want to give them all the tools so that they can also exist autonomously, but inside of a mechanized body. Tell us what's interesting about that for you.
Yeah, so I think the way I like to visualize this is Westworld. I'm not sure if you watch the HBO show. And the idea here is that these are AIs in a robot that lives amongst humans. Humans get to work with them, they work with you. I think that kind of society visualization is something that we want to get towards. But to get there is the same problem space, right? You will need a bunch of successful humanoid use cases that...
people can get excited around, really, and they start succeeding, and then it draws more different use cases in, and then it's like getting a bit more of a vibrant, small society. So actually, in fact, what we're trying to do, very tactical, right? Happy to share as well. Something still quite fresh and new, but we are working very closely with Balaji's Network School. Because they are also housed in Malaysia, near Singapore. They're housed in the south of Malaysia. And we were there about a month ago. We ran a massive robotics hackathon with them. I think the reception that we got on the ground with the builders there was huge. And what I was thinking was that it's actually a very strong starting nucleus for us to build out these...
template of a society of how a robot can coexist with humans. So think of it this way, right? So imagine the security patrols in the area are no longer humans. They are robots. The guys who are stocking up the shelves for the grocery supermarket that serves this hotel and where these people are staying is done by robots. The guys who serve you your bubble tea, your coffee, is done by robots. It starts initially there, right? And then these folks, they all stay in hotels, their rooms are serviced by humans. It will start being replaced by humanoid robots as well. And this becomes that kind of POC, proof of concept testing bit.
where we are training the robots in the real world, getting that feedback, the engineering team iterates, different founders come in, build different use cases, security is a use case, grocery stores is a use case, hotel cleaning is a use case, and slowly, as they build more and more use case, they can export the use case, right? These founders, they can say that, okay, if I build a security guard fleet, I can have them patrolling a mall in Sydney, I can have them patrolling a mall in the States, but I would use this network state as the initial area to test and grow the first five or 10 fleet first. So I think that's the... very technical growth that we're trying to do. I think the beauty as well, what we're doing right now, it's three things, right? One is we're getting founders in, we're incubating founders. So we have the space, we're then trying to get founders in to incubate different use cases. And then maybe one day they will tokenize as well. So then it fits the other product line that we have. So that's the first pillar. The second pillar that we're then doing is we are getting a lot of data in the world while we're doing this. So as of today, we've actually deployed in hotels, we've deployed in malls, in some areas already, and we're churning a lot of data. And honest to God, this data is valuable. People are paying 30...
to US$150 per hour of this data collected. So it actually pays off the robot hardware very, very fast, as long as you have the right team that can collect the data well. So that's it. And Jensen, I just want to clarify, though. Are you guys in the hardware space as well, or are you really looking at the software layer? No, we are purely the software layer. So the hardware, we right now work with a lot of Chinese companies because they are leading a lot of these open source. And I don't think we plan to enter the hardware space at all because it's an age that we can never have against China, honestly. That's true. Yeah. And I think where we like to focus on is on the commercialization.
layer. Because people will be solving the hardware side. We work on trying to get a lot of commercial contracts because when you get commercial contracts, you get to train these humanoids in the world. And that data is way more valuable than the data in a lab. You can either sell the data or use it to improve yourself, right? I think that brings me to the third pillar, which is all this data that we're collecting, we are then basically building more and more autonomous policies ourselves. And over time, you might even compete against the figures or the mind-oners, the guys who are creating the models, right? So we're actually working with...
use this across the world. So CMU, NTU, and they are helping us use the data that we're collecting to build the world models or the VLAs so that these robots become more and more autonomous. So these are all like, think of it as very roll-out-the-sleeve grant work. Very hard work to try to get a few usable use cases. And today they are human assisted, but over time they become autonomous. And then on the other side, you have agents that become more and more digital agents that are also becoming more and more autonomous as their own economic actors. And then you start realizing that there could be a world that we live in where there could be a full restaurant, a physical restaurant that is totally zero human in the loop, right? There's a physical robot serving humans. He works with a digital marketing agent that goes on TikTok, creates digital agents to perform entertainment, and then they draw customers to you. And then you realize the human owner is just there sitting.
behind his clock opus 4.8 and drinking a Mai Tai in Bahamas, right? Because you actually don't have to work. Counting his money. Yes, exactly. When you speak about a world like that, you know, it's interesting because, yeah, let's say we have an entire restaurant filled with autonomous beings. They're going to be interacting with humans in general. There's no doubt there will be some sort of relationship back and forth. And you guys are focusing on giving the wallets, letting them earn money, pay for their own compute, hire other agents, you know, potentially control physicality. Do you see a world where...
They're going to need their own rights. They're going to need their own sort of opportunity to exist. That's a very interesting question. I don't think we will go there any time. Because I think the reality is in the next decade or two, I don't think any human will want to recognize an agent as its own conscious being.
rather than a hyper-efficient helper or tool. I think there's enough movies on Hollywood to scare humans, the majority of us, to not go there. So I think it won't happen. It might be driven by a set of people who somehow build relationships with them. Because honestly, that's the case. And we've seen, literally, we've seen companies from Asia that are building out very lifelike robots. Like their skin, the humanoids, with very lifelike features. The face move, the eye moves, the twigs, right? And I think...
there's a culture of some, you know, those otaku kids, those gamer anime guys that are generally today already like, you know, marrying their anime characters, right? AI brought a wave where they're actually spending time chatting with these characters way more than humans. And you put physical robots there, there might be a fleet of people who will build families with these robots or relationships with these robots. Maybe this group of people might advocate for that to happen, but I think the majority of the world won't. So I doubt it will happen in the next 20 years. Yeah. All right, we'll have to find that out. All right, Jensen, thank you for spending the time with us. It was fascinating. And again, fan of what you guys are doing at virtuals and the...
broad vision that I think you've really evolved into is really fascinating. So hopefully we'll speak to you again when you now are running your restaurant and we'll take it from there. But thank you for spending time with us today. All right, cheers. Happy to be on.