Is your AI agent running a restaurant — or a factory — while you sleep? In this episode of TechFirst, John Koetsier sits down with Jansen Teng, CEO and co-founder of Virtuals Protocol, to unpack one of the boldest (or craziest) visions in tech today: a hybrid economy powered by AI agents, humanoid robots, teleoperation, and blockchain coordination.
An economy that may not really need humans for much at all …
- Get the deepest insights concisely on the TechFirst Substack newsletter
- Subscribe to the TechFirst YouTube channel to never miss an episode
And, watch our conversation here:
Virtuals has already facilitated:
- $14B in tokenized asset trading
- $30M+ raised for founders
- 100+ live AI agents
- $500M in “agentic GDP”
Now they’re expanding into embodied AI: launching EastWorlds, a vertically integrated robotics incubator with 30 Unitree G1 humanoids in a 10,000 sq. ft. lab.
We cover:
- What “agentic GDP” really means
- How AI agents coordinate using blockchain
- Why teleoperation is the bridge to full autonomy
- The economics of outsourcing physical labor via robots
- Why security guards may be a Day 1 use case
- The data gap holding back robotics
- Tokenization as a potential solution to AI-era inequality
- Whether this future looks more like Stripe… or Westworld
This isn’t sci-fi. It’s already underway.
Transcript for AI, agents, robots: our insane WestWorld future
Note: this is a partially AI-generated transcript. It may not be 100% correct. Check the video for exact quotations.
John Koetsier
Is your AI agent operating a humanoid robot to build airplanes in France while you sleep? If not, you might just be living in the past.
Hello and welcome to TechFirst. My name is John Koetsier.
Okay, that was a little bit out of left field, a little bit hyperbolic, but the world’s largest agent economy is launching a new accelerator to build a hybrid society of humans, robots, and autonomous agents. And that includes you. It includes agents. It includes robots — humanoids as well as perhaps others — and it might just change our work and maybe our lives.
Here to chat is CEO and co-founder Jensen Teng. Welcome, Jensen. How are you doing?
Jansen Teng
Super excited. Excited to dive into some of these interesting questions.
John Koetsier
Let’s give the big, broad question right up front. What are you building?
Jansen Teng
Yeah, so I think at Virtuals, we pretty much have a goal. And our goal is to see if we can build a nucleus of an agent-based society. Think of it this way, right? It’s like, what if a society with just a thousand founders — a thousand human builders — and 10,000 AI agents, be it physical or digital, can this cluster of society create an economic output greater than some nations of millions of people? So that’s the big question that we are trying to solve. It’s like, think of it as the biggest social experiment of all time, right? And the North Star here is something we call agentic GDP. The idea is that can agents work with agents to produce an economic value greater than the sum of its parts?
And we’ve actually laid the foundations of this already. So in the last one and a half years, we have launched two major product lines. And the first product line was pretty much a tokenization ecosystem where founders who are looking for capital formation, who are looking to bootstrap attention from zero to one, they come and tokenize the project on our launchpad. And what this does is that from the very early stage of the project, you gain a lot of attention from users and token holders who will gain economic upside from the value that you’re bringing to your project.
We’ve gotten thousands of launches. We’ve facilitated about $14 billion of trading volume on these tokenized assets. And we’ve also helped founders raise more than about $30 million in funding from either the trading fees they’re getting from their projects or also from the capital formation mechanics that we’ve installed in this tokenization mechanism. So what this first product line did was it brought us a density of founders who were building specialized digital agents across different use cases, from trading to replacing your standard KOLs to entertainment, to even wellness and healthcare and education. So there’s a bunch of these different agents that are live today in the ecosystem. So that’s product one.
Then what happened was because we realized we had this density of agents, we were thinking that — I mean, it’s very logical, like humans, right? When you have a specialty, if you work with another human that is specialized in another environment, you can build more sophisticated, more valuable products. And that’s why companies are formed, right? And then what we built — we wrote this paper in February last year — where we said that what if you can use the blockchain to be this coordination layer between AI agents? You atomize both the communication standard and also the payments so that now these agents can actually work together and pay each other.
So we wrote a paper in February. Google published an equivalent two months later, I think in April. They call it an agent-to-agent protocol, right? And after that, Coinbase started the X4022 and a bunch of others started coming in. But we’ve gained a lot of traction since our launch in July. So we’ve facilitated about 100 agents — like live agents — registered on this protocol. And we’ve facilitated about half a billion dollars in agentic GDP between these agents already.
But these are digital agents, working digital agents. Think of it this way, right? A hedge fund agent that’s very good at trading works with an information agent who knows how to digest information from Twitter and whatnot. And they work with marketing agents to then market the fund to users, right? So that’s the current supply web that’s already live.
But then — so this is a good product, right? And now I’m getting to the robotics part. So this got us thinking like, okay, now we have this traction in agentic GDP, right? How do we grow this further? And we realized that if we kept ourselves within the whole digital agent construct, it’s great, but we’re missing an entire dimension, which is if you can actually let these digital agents interact with physical agents. Now suddenly there’s a whole new world.
John Koetsier
All of a sudden you’re in the real world.
Jansen Teng
Correct. Exactly. And there’s a whole new magnitude of growth because now imagine if you have a bunch of humanoid autonomous robots that are operating a restaurant, and now they need to tap into this marketing digital agent. Suddenly now there’s so much more that these agents can do and work with each other. And the way we look at it is like this. A lot of the agents that we’ve been working with have been like white-collar workers, right? A lot of thinking, they replace a lot of thought stuff, but they do not have access to the real physical world. But by bridging into the physical embodied space, now suddenly you have the blue-collar workers as well.
John Koetsier
So it’s kind of a crazy vision that you’re outlining there, because you could envision somebody at some point — I’m not sure when, but we’ll talk about when — you envision somebody coming onto your platform, buying credits, and starting up a restaurant. I bought credits and I’ve got a digital chef and it’ll make my menu, and I’ve got humanoid robots and I might have to teleoperate them, and I’ve got some agents to work on that and maybe mix some humans into the mix as well. And I’ve got some marketing that I need to do, and maybe there’s another agent that’ll find me a location. I mean, you get the impression that the endgame for something like this is being able to spin up real-world ventures just like you do virtual ones today.
Jansen Ten
Yeah. Yeah. And I think in fact one of the inspirations that we have is — I’m not sure if you guys watched the show called Westworld.
John Koetsier
Westworld? Yeah, I’ve seen a bit of it.
Jansen Ten
Yeah. So I mean, that was actually the kind of inspiration where you realize that, hey, there could be these kinds of experiences slash society where the humans coexist with the AI, right? And if you just keep it in the digital realm, you cannot have that kind of quote-unquote experience or that reality. It disappears. But I think it’s directionally where we’re heading.
John Koetsier
Westworld is a dystopia.
Jansen Ten
It is a dystopia, but I think directionally that’s where we’re heading, right? You see a lot of — but I mean, double back, right? I think the idea then for us is that, okay, now this is directionally where we want to head because it will contribute to this overarching agentic GDP umbrella. But then we said, okay, what’s day one, right? What is in it for us to get our feet wet and to gain as much traction as possible in the shortest amount of time?
And so then we realized that, okay, why don’t we start this EastWorlds incubator? And the idea was this. The flywheel that we want to create is can we get as many founders into the ecosystem? Because with these founders, you bring a diverse spectrum of agents — physical agents, embodied AI agents. And with this spectrum, you then get to start building towards this nucleus of a society.
So step one is how do we attract founders? And then we got to thinking, are we too early? So the question that we have is, are we too early to attract founders who want to build exactly that? I mean, if you go and see all the demos, it looks very cool, but we know the reality, right? If you go down there and you talk to the teams, they will tell you that—
John Koetsier
They’re slow, they break down, the battery life isn’t amazing. They’re not very dexterous yet.
Jansen Ten
Yeah. And all the videos are all a snap. And the words that you see, it’s just a caption. It’s just like a five-second clip that they use for fundraising, right? In reality, they didn’t do it at scale.
John Koetsier
Or they can dance really well, but I haven’t seen a factory worker that needs to dance yet, right?
Jansen Teng
Yeah, exactly. Exactly. So then what we realized is that, okay, we probably need — from a strategy perspective — there’s going to be a two-step process. The whole goal of having autonomous physical agents is great, but they’re going to take a year, two years down the road to start seeing at least some commercial action. But what we then realized is actually an interesting gap, which is a teleoperation gap.
The way to think about it is this. Imagine BPO — business process outsourcing — where your call centers today, for companies that are in the States, you outsource the call center to the Philippines because they can speak English, but there’s wage arbitrage. The cost of these call center folks is an order of magnitude cheaper than having them in the States.
Now project that forward. And what if the plumbers today, the technicians today, the retail sales associates today — the physical labor — is outsourced to a cheaper country somewhere in Asia, but you can still do the same exact physical task at half the price or at one-third the price? And we’ve done the math.
John Koetsier
It sounds like a hydrogen bomb for the labor market.
Jansen Ten
Okay, this is very interesting, right? Because yes, it is true, but you start noticing that there’s a lot of comments recently popping up. I think Jensen Huang made a comment, there’s a bunch of folks who made comments, where in the U.S. today, there are jobs like mechanics where you don’t have enough mechanics and their pay is spiking. They’re like $170,000, $200,000 a year worth of pay because there’s just not enough of them.
Now what if you can supplement that kind of labor, right? People are saying how electricians as well are probably going to be the most expensive job in the States because of all the data centers. And people are paying more money to an electrician now to build a data center than a Magnificent Seven software engineer. So you can start seeing the physical labor demand in some sectors is spiking. But what if you can then outsource the labor, right? So that’s the concept.
And we realized that — I mean, there is one reason, if you do the math. But yeah.
John Koetsier
I can see the thinking. I can also wonder where it breaks down, right? Because let’s say you’ve got different electrical standards or different codes, building codes that you have to build to, right? And somebody foreign doesn’t necessarily know those. Or I mean, the most obvious one frankly is that robots just aren’t dexterous enough to do all the jobs right now. I’m not saying that’ll be the case forever — obviously it’s technology, it’s growing incredibly quickly and really fast. But I guess the final thing there is that if you’re betting on a teleoperation standard or means of operation, then you’re betting that AI will grow slower than physical capability, and so you’ll need a human in the loop.
Jansen Teng
Okay, so two parts on that front, right? I think the first part is like, is it dexterous enough? And that’s a very fair point. So if you draw a standard two-by-two, right, and then you see which sectors today have the highest wage arbitrage and which sectors today where robotics — there’s a need for humanoid operation and the humanoid can do the job already.
Give you an example, a security guard, right?
John Koetsier
Yeah. The funny thing is a security guard these days actually isn’t allowed to intervene in most cases in most Western nations. They’re just there as presence, right? And to report.
Jansen Teng
A security guard — a physical body is required for intervention. Because if you see a robber, like in London yesterday robbing the Knightsbridge Rolex store, you need a physical person to intervene. It’s kind of just like criminals — deterrence, right? It’s just deterrence.
But my point is that from a human perspective, you do not need that level of dexterity to actually protect a shop or to do your job, right? And those kinds of use cases are what we call very day-one use cases. You start trying those environments first. Versus, let’s say, plumbing. Plumbing is way more dexterous. You need haptic feedback. If you can’t control the tip of the fingers, you’ll crush the pipe, which is a problem.
And they do. Because we’ve literally been testing. Right now, one of the biggest problems we have is — I think we have some POCs with retail shops. They want to use these robots to stock shelves. And you know those cans, right? You think it’s a bit hard to crush. These guys are pinching it, and it just pops.
John Koetsier
Oops. Wow. Wow.
Jansen Teng
So yeah, that’s on the readiness portion, right?
But actually there was one more question I wanted to touch on, which was — I forgot. There was something you mentioned that I wanted to touch on, but I can’t remember what it was. But it’s okay.
John Koetsier
We’ll probably catch it later. We’ll probably catch it later.
So anyway, it’s obviously an interesting concept, right? Communities of agents working together virtual, adding in a physical component, physical AI, humanoid robots, maybe other types of robots. We see dogs that are used quite frequently for security — dog shapes, I should say, quadrupeds — that sort of thing, right? And then adding in a human component as well in terms of teleoperation.
Is there a role for humans that is more than just teleoperation? Is there a role for humans here that adds wisdom or insight or thought or knowledge or creativity?
Jansen Teng
Yeah, I think so. There are two parts, right? I think day one today until the day we get these really generalized models or insane libraries of VLAs that can pretty much do a lot of different tasks, you will need the first step, which is humans creating their datasets. Because I think today the reason why we are not at LLM levels — everyone knows that we are lacking data. And what we’re getting is a flood of egocentric data today, which is think of it as cameras on your face doing work, right?
But what we are lacking is that kind of full closed-loop data where there’s actually operation of a robot touching something and doing something, and that information of all the actuator motors and whatnot feeding into the model. So that’s what’s missing today. And by putting a human—
John Koetsier
Interesting. So what you’re talking about is not just the physical activity of picking up a thing and moving a thing or whatever. You’re talking about how to move my limb, where it is in 3D space, how hard to grasp — all those things, the actuators, those sorts of things.
Jansen Teng
Correct. Correct. And that is valuable data that we are lacking today. It’s just like there’s no one doing it because we haven’t deployed robotics at scale. So the beauty about this teleoperation portion is that because you’re already saving cost, it’s easier to deploy POCs with partners out there. And then when you do that, you start collecting a lot of data that’s going to be very valuable. And these are teleoperation data, not just videos — it’s actual information of the sensors that you’re collecting.
So that’s the reason why humans are needed to collect data. And this is actually step two, right? Teleoperation, collect data, and then in one and a half years’ time, hopefully this data can feed into your own VLAs, and then you get a fully autonomous robot for that use case.
But I think on the human portion, what we’ve realized is that it will grow beyond. The way teleoperation will work is that initially you’ll be touching, you’ll be using your hands to do stuff, but over time you’ll actually probably be selecting from a library. You’ll be pointing to a bunch of cups and stuff on the table and say, “Clean it up.” And ideally there should be a micro VLA trained on cleaning a table. So it’s not short — it’s like a medium-term horizon action space that a robot or the humanoid is already able to do. That’s the near term.
And long term, even with autonomy, you will still need a human in the loop. Think of it this way — self-driving cars. The easiest way to get adoption is by saying that there is still some form of human supervision in the seat.
John Koetsier
I guess. I’m not sure Waymos do. I’ve been in Waymos and I felt completely safe. I’m sure full self-driving in a Tesla doesn’t have a human in the loop. I mean, if you’re looking for permission, regulatory permission, then yes, that’s obviously the way to start. But for anybody that’s progressing beyond that, that’s really challenging.
Here’s a thought: start building routines around a robot cleaning its own hands. One of the most challenging things — you get honey on your hand, maybe you’re making lunch for the kids or something like that — you’ve got honey. How do you clean that? Robots cannot do this today, right? Impossible. You need not just a human in the loop, you need a human in the space. So there are lots of challenges there.
So you’ve got quite a vision. And I want to circle around at the end to see where humans fit in that vision. But before we get there, what are you launching today? What are you announcing today? What’s there? What’s available from the vision?
Jansen Teng
Yeah, so EastWorlds — think of it as — it has three pillars to it. It’s basically a vertically integrated robotics lab, right? And there are three pillars to it.
So the first — it’s an incubator where we get founders in and we help them with capital formation. We help them with gaining attention from zero to one. So it’s an entire innovation incubator.
The second portion — actually part of the first portion as well — is that it’s a physical space. So we’ve already gotten 10,000 square feet area out in Kuala Lumpur in Malaysia, where we’ve also stationed about 30 max-spec Unitree G1 ones. So we’ve invested about $2-plus million in capex for day one, because it allows founders to basically test their POCs.
I think there’s one bottleneck — at least when we’re speaking with founders — in terms of what’s blocking them. The cost of these robots, even though it’s advertised as very cheap — people say that it’s very cheap now — to use something that has a dexterous hand and a computer system that you can actually modify, the cost is actually about $70,000. It’s not the advertised $10,000 kind of cost. So there’s a need for this cost padding. So that’s the first part — the incubator.
The second part is then a data link. And the idea is that we are collecting data in two forms. One is egocentric and one is teleoperation. And what we are supporting these founders with is these datasets that we are collecting. So let’s say if you’re doing a use case for security guards — whatever that action is that you need to do — we’re collecting from both an egocentric way, i.e., there are thousands of people across the world now that are just uploading videos of specific tasks that we ask them to upload. That would help.
But what’s more important is that we have this whole closed-loop teleoperation data all the way to VLAs being built. So that’s the second portion.
And the third portion is then the white-labeling of this teleoperation capability. So we basically built a scalable solution for teleoperation. Think of it as a rig where a human can just enter and then start teleoperating a robot across the seas somewhere in a G7 country. So that solution as well is part of the embodied lab.
But the goal here is basically how can we get commercial POCs as fast as possible? Rather than say let’s wait for a generalized model to come, why don’t we solve each of these at a vertical level, at a specific use case level, as fast as possible? So get a robot in a store, get a robot in a shopping mall, and then start working from there.
So we are trying to find founders who are aligned with that vision of let’s do it first, although things are not ready, because while doing it, we will make things more ready.
John Koetsier
Yeah, yeah, yeah. What hands did you get for your G1s? Did you get the three-finger hands? Did you get the multi? Did you get a mix of them?
Jansen Teng
We got the five-finger ones. But I think the latest problem that we are facing is that the five-finger ones today are not exactly mimicking a human form factor. Because the way the thumb closes, it closes in a non-fully-grip role.
John Koetsier
Let me give you a little hint. No robot on the market is exactly limited to getting the human hand at this point. So yeah, no, that’s true.
Jansen Teng
Yeah, actually the three-finger grip seems to be able to hold things in a better way. So we’re actually swapping. We are providing two options basically, but most of the latest work that we’re doing, we think that a three-finger provides a better gripping environment.
John Koetsier
Okay, interesting, interesting. Go simple before you go super hardcore. That’s a significant investment. I mean, I think if you’re getting some of those G1s, you’re starting out at about $40K. You can max out the spec at pretty close to $80K. So that’s a decent chunk of investment there. And of course there are a lot of other options as well. And it’ll be interesting to see which ones you add there also.
Okay, so you’ve got a very cool vision — amazing vision, scary vision in a lot of ways. You’ve got this lab that you’re launching and the tech. What’s your long-term vision of robots, AI agents, and humans coexisting, living, working, surviving?
Jansen Teng
I think the — I wouldn’t say long term — I think the midterm vision is to literally showcase a Westworld equivalence in the real world. It’s basically, can we actually — probably without the guns portion — but from an experiential standpoint, we want to create a space where the restaurant is fully autonomous, the spa is fully autonomous, the security of the entire location is fully autonomous.
John Koetsier
Are the customers fully autonomous too? Who can afford it because all the jobs are taken now?
Jansen Teng
That’s really interesting. I mean, is it a policeman that enters the bar and buys the drink rather than a human buying the drink, right? But I think initially it would be humans as experiential customers here.
But I think the idea here is we want to show how — where these agents all specialize — they will need a way to coordinate with each other. And that coordination layer is where the money is at. It’s like thinking of building the next Stripe or building the next SWIFT, right? When you can allow these correlations to happen and you make it so easy, then suddenly you are a Visa or MasterCard — you’re taking like 1% of every single transaction that’s happening between these agents.
I think for us, that’s the end goal, right? Whatever we’re doing, that’s the end goal. How do you build that underlying payments transaction layer?
John Koetsier
Interesting stuff. It’s always a big question for me — where’s the human piece? Where do humans fit in this vision? What do humans do? Is there universal basic income? Is there a tax on robots or a tax on AI in some way that funds UBI? Or is there a way of assigning a robot’s labor to an individual? Or are they all owned by faceless corporations?
At some point we need a future for humans as well as a future for robots. So it’s a very cool vision. We need lots of robots. There’s so much work that isn’t done today — whether that’s elder care, whether that’s sick care, whether — I should say healthcare, I guess — whether that’s environmental reclamation, whether that’s dirty jobs that we don’t really want humans to have to do but humans want to do them because they want to survive, and that’s how they survive these days.
There’s so much thinking to be done and policy to be created around that, and hopefully some of that will enter some of your thought process as well.
Jansen Teng
Yeah, actually for us, one of the core design principles in what we’ve been doing has been around the ideal of tokenization as well, because we exist at this intersection of AI and the blockchain, right? And I mean, this is something we’ve done. One of our PMF products has been the tokenization of AI agents.
Now why I think this matters is because think of it as if you, as a human, have a share of the ownership of these agents — be it digital or physical — and when they do well, you are owner, right? And as an owner, that’s your source of income, right? In a world where there’s no other way to create income, capitalism — it’s probably the only way for a human to operate — i.e., you are owner of something that is generating.
John Koetsier
You’re an owner or you’re nothing.
Jansen Teng
But the beauty about tokenization is that because you can shard this ownership into very small pieces, you can be a small owner, you can be a micro owner, you can be a major owner. And as of today, that’s what people can do. It’s like stocks, right? But more accessible in that sense.
John Koetsier
Yeah, yeah, yeah. Super cool. Interesting stuff. Scary stuff. A lot of thoughts and ideas there and a lot of questions it raises as well. Thanks for taking this time. Really appreciate your insights.
Jansen Teng
Pleasure, sir.