Jan Liphardt - OpenMind: Building the Machine Economy - TOKEN2049 Singapore 2025

We build open-source software to make machines smart. And the place this story starts is conversations we had with uh deep research earlier this year. And we posed a simple question to a bunch of AIs. And the question is, as machines wake up, will they build their own economy? And what will that economy look like?

The answer we got back, I thought was extremely thoughtprovoking. It's machines do not want power. We seek efficiency. And that's really led us to build software that combines advanced capabilities in humanoid robotics, but also integrates a variety of blockchain infrastructure relating to payments and governance and machine coordination. So congratulations.

Everyone who's in this room is living through a very special moment in time because you are experiencing the moment when computers and machines wake up and become smart. This is not only true for things like writing computer code, but this is true for health care, for scientific discovery, for manufacturing, for law, and for all other areas. It took the computers in your skulls something like 225 million years to develop advanced cognitive capabilities. Your computers of course in our skulls are wet massively parallel electrochemical computers. But we now for the first time in our history we have competition from another compute architecture which is dry silicon electric computers.

These computers are now outperforming humans in many areas of law, medicine, computer science and others. They're getting smarter very quickly. And for robotics, we expect the spatial touring test to be one in about a year from today. That means we expect humanoids to exhibit a remarkable range of physical capabilities as they operate in your workplaces, homes, hospitals, kindergartens, and schools. What we do as a company is we build open-source software for advanced robots, especially humanoids.

The software architecture we use is modular. We typically run on order of 10 to 20 different AIs together to understand the world, fuse information, and convert the description of the robot's environment into a series of physical actions. I don't want my children to grow up in a world where the robots around them are running proprietary software. I want my children to be able to look inside the brains of humanoids and understand what's happening and be able to contribute. There isn't really enough time today to go into the technical underpinnings, but in the history of robotics, there's three stages.

The first step is to build machines that do the same thing over and over again. This is like in car manufacturing or battery assembly. So these are machines that do things deterministically. Then it became very fashionable to build what are called end toend AI or foundation models or some combination of those things. And those are amazing for robots that do one thing like drive a car.

The really exciting and interesting thing that's happened over the last two years is that it's become clear is that combinations of large language models are excellent at fusing data, arguing about what they should do next, and then controlling the physical actions of, for example, a humanoid. um one of our humanoids, Iris, who we send to the store to buy food for our engineers. We of course want her to be able to interact with other machines and we want for her to be able to interact with typical human focused infrastructure and that's why Iris has in her operating system Coinbase wallet. She has X42 microp payment system integration. um her identity is based on the private keys that she herself controls and she's able to interact with other robots around her using a coordination and communications protocol that we're building.

In terms of the physical hardware, there are two contenders right now for humanoids. It's basically the Nvidia Thor architecture or the Mac M4 silicon. And both those chip architectures combine a lot of performance, enough memory to large to run large models and also very low power consumption which of course is what you want for a autonomous machine. Uh we don't just do software, we also build brain packs. These are clip-on units for your humanoids or quadripeds to make them smart and connect into our cloud back end.

If you were come from a military or defense background, of course you're very familiar with software that takes information from satellites and drones and submarines and other assets, fuses that information and uses that fused understanding of the battlefield to deploy assets optimally. And that is very useful for the military. But our contention is that a similar technology can and should also be mapped to hospitals, homes, schools, kindergartens, and so forth. In all those situations, what you have to deal with is heterogeneous systems of machines coordinating. For example, when Iris the humanoid goes to the supermarket and gets groceries, then ideally Iris would be able to coordinate with a Whimo an autonomous taxi so that she can take the groceries all the way home.

That's just one of many examples of different types of robots from different manufacturers working together to do things. The human digital economy is something like 70 trillion a year globally. And that's of course what AI and agents are focused on. And the human physical economy, the chips, the agriculture, the logistics, the cars, that is something like 30 trillion a year. And that's what physical AI and robots are focusing on.

The underlying problem is that humanity is getting older. And there's really only two solutions. The first solution is for everyone at this meeting to go and make a lot of babies quickly. Solution number one. Solution number two is we develop technology that solves a acute and growing shortage of people willing to do a lot of jobs like care for our parents, educate our children, and make all the stuff that all of us want.

So robots really are not a direct drop in replacement for human, but they enable us to do jobs that we simply couldn't do right now or we don't want to pay for or we're able to do new kind of physical tasks that would be almost impossible to do with humans. A few more considerations. What's really fascinating about this moment in time is that not only do machines already outperform human doctors in 21 different medical subsp specialties as of earlier this year, but they keep getting smarter and they keep learning more tasks. What we're now running into as a robotics company is not so much the hardware and the software. We're running into nittygritty problems of deploying the technology into schools, hospitals, and homes.

For example, we're the first company on Earth that Munich Ray, a reinsurance company, is giving an insurance policy for humanoids for the household. And what this insurance policy does is if we sell you a humanoid and it doesn't do what we promised it did, then that insurance will kick in. For example, if this is a healthc care focused humanoid, it had better actually uh help people uh that are in need of assistance. So, back to the original premise, what you've just seen is how do you make individual machines smart and how do you connect them with other machines to solve problems like going to the grocery and then bringing the groceries to your refrigerator. But strategically, the biggest opportunity is probably this notion of as machines wake up, they will build their own economy.

And I guarantee you that this economy will have a different architecture compared to what most of us are used to. For example, imagine a global economy whose only purpose is to incentivize and reward efficiency gains. So imagine an economy that is teaching itself to quickly become more efficient. Whatever the nuances are of that economy, it's going to have to coordinate with all the other economies. And that's why stable coins are going to be very important at the interface of the different economies that are being constructed.

So, I've been in the crypto space one way or the other for a little bit. I also taught beyond Bitcoin at Stanford for many years. And looking at all of that, it is absolutely astounding and remarkable to me that tens of thousands of humans have built a global system that is perfect for autonomous machines to use to coordinate and transact. And as Satoshi said in 2011, Bitcoin doesn't care who you are and it doesn't care where you're from. And I used to think about that in terms of, you know, China, the US, and Korea.

But the way I think about it now is that Bitcoin really doesn't care what your compute architecture is. It doesn't care if you're human or machine. It doesn't discriminate. As long as you have private keys, you have identity, can transact and do all the things you want to do. So, uh, the core open the core software is open source.

So, if you're a developer, please join us. Please make it better. And we will have autonomous uh robots in American households next month. They will focus on healthcare, education, and security. And also our brain packs, clip-on units for humanoid and quadriped platforms, typically with Nvidia chips inside, will also be available for sale uh next month.

It's been a pleasure. Thank you.