why we exist

Built from Theo's words in the founding interview, 2026-09-29. Approved as written, no corrections.

I have always been a distractible kid. Not in the broken way, in the full way. Full of ideas about what I wanted to do with my life, things I wanted to build, inventions that might make a difference. I was a problem solver from a young age, and I had the good fortune of a father who treated that like something to train.

He would ask me questions about the world, about science, for hours. We would be in the car driving somewhere, and he would ask me why the trees on the forest floor bloom earlier than the taller trees, and I would sit there and wonder as the trees passed us for miles and miles and miles. I did not have tech growing up. There was nothing to look it up on. I was just forced to sit there with the question.

So I grew up inside science, and I grew up a problem solver. And inevitably I stopped wanting to solve my dad's problems and started wanting to solve my own. My own problems were meta problems. The biggest problem I had was that I wanted to address more problems. If you have an ability, you want to use that ability on the thing that would maximize the ability itself. That became a doctrine for how I live: play to your strengths, and use your strengths to set up your environment so that your environment sharpens your unfair advantages even further.

I knew problem solving was the strength. I wanted a job where I could sit and solve problems all day. I loved the ideation of it. The action, less so. My fantasy was simple: I would have the ideas, and someone else would carry out the work. For a long time that was not a feasible arrangement. Then, without anyone predicting it, agents arrived that could take a half-baked idea and do the work. It was amazing. And it was not enough.

Because here is what I wanted next. Yes, the agent can build my idea. But can it keep up with my every thought? Can it understand how the idea evolves over time, and keep the build clean and current through every evolution? The demands of that are specific. The agent would need to pretty much share the same brain as me. That is where Cosmos found its footing.

I did not arrive at this from the literature. I had not read Licklider when I landed on the shared brain. I had heard of Clark and Chalmers and the extended mind, but even that came after I started building. The research found me later, and it confirmed what I had been discovering on my own. I think of it as picking up a thread that has been sitting in the research for a long time. It is slowly gaining relevance again. But people are still missing the big picture. Or maybe they are working toward it and it just takes time. Either way, the thread is there, and someone needs to carry it.

The big picture is the shared brain hypothesis, and here is the version I believe. Look at the creative world if you want to know where tech is going. Artists predict our inventions before we invent them. Cyborgs have been predicted in film and television for decades, and we are slowly heading somewhere like that. Not in the way we expect. There will probably not be humanoid robots walking around. But there will be humans with embedded interfaces and external computing devices they rely on, the way Clark and Chalmers described the extended mind. Probably sooner than later.

Right now the industry is building toward what I call the safe hypothesis: separate spaces for humans and computers, each operating in its own world. That only creates more headache. Humans and computers need to share the same world, the same environment. Either computers enter the human environment, with more sensors to understand the human world, or humans enter the computer environment. I am betting on the first direction. Computers need to learn to understand humans better. Humans sense the world and convey it efficiently to machines, and machines build their understanding from what we give them. Not machines trying to understand the world alone while humans stand by correcting them.

Radiology makes this concrete, and I am about to become a radiologist, so I think about it constantly. Today's chest x-ray AI tries to read the image on its own, and the human carries the responsibility of correcting it. The better implementation is the reverse: the human understands what they see and conveys it to the model, and the AI adds what it understands to enhance the read. The human leads with understanding. The machine enhances. That is the direction of the shared brain.

You know when it is working because you feel a specific click. For me it happens when the machine says something about where I am headed, something I did not objectively state, and yet I know it to be true. I smile. Models are notoriously bad at synthesis. They capture and store and recall, and that is so different from synthesis. When I see synthesis, I smile a little bit. That is the whole test. Not recall. Convergence.

So here is 2036, if the hypothesis wins. Human life looks like symbiosis, but not the sci-fi kind. The effortless kind. You think things, and when you check in with your device, it has thought those same thoughts. We are already halfway there. The algorithms already know what you like before you open the app. Everyone has had the moment of telling a friend, we just talked about this, it will be on my feed tonight. Models are already good at predicting human thought. The next level is asynchronous synthesis: you return to the tool after time away, and it has predicted what would have happened while you were gone. The catch-up is not a catch-up. It is a continuous pickup, as if it never left.

That gets integrated into everything, the way AI is being integrated into everything now. Images read by agents the way we would read them. Essays written the way we would write them. The simple versions of that fall in the next five to eight years. The next ten are about the bigger versions: not just acting as we would, but planning as we would and thinking as we would. I believe human consciousness will be mapped, to some degree, by 2036, and we will understand the spectrum of mental illness and disease far better, and we will have far more insight into where a current trajectory leads.

When I first started building Cosmos, I tried to pack in a future predictor. You would enter a year in the future and a desire, something like achieving financial freedom by next year, and it would take everything it had learned about you, the thinking pathways, its model of you, and evolve it forward, predicting all the expansion and connections your brain would need to make to get there. Then you could recursively fill in that knowledge and learn from it. Downloading what you need to know to achieve something, overnight. That is the symbiotic theory at its endpoint: computers able to transfer understanding back to us at remarkable rates, because they have learned so well how to understand us. The first step is computers understanding us better, so that the next step runs smoother.

That is why Polarity Lab exists. Not to build a second brain. To build one shared brain, for you and the agents you work with. So that you can spend your life on the ideation, the wondering, the miles of trees going by, and trust that the thinking continues even when you look away.