Pogl researches foundation world models that make imagined worlds real.
A video model makes the scene you watch, and it plays out the same way whatever you do. A world model takes the current state and your action and produces what happens next, so the world you end up in depends on what you did in it.
For as long as we have built worlds, a world was whatever someone wrote down. The way water flows, the angle light bends, the order things collapse in: every rule had to be authored by hand, and a world existed only as far as it was written. That is why worlds cost studios and years, and why almost every world that could exist never does.
Pogl builds both halves of the alternative: a model that learns how a world behaves instead of being told, and the place where people use it.
Pogl Unworld is one model that learns the dynamics of a world. No engine, no assets, no genre fixed in advance: it takes input and generates the next moment. It has to hold together when you look away, respond to things nobody scripted, and run fast enough that you forget it is being generated.
Pogl Play is where anyone makes an experience with AI and steps into it with other people. What gets created and experienced there teaches the world model, and in time the model will run inside the platform itself. Research that never leaves the lab is easy to believe in and hard to prove, and we would rather be judged by whether people actually want to be inside what we make.
We are still early, and building in stealth. If you want to join us, write to founders@pogl.com. Now is a good time.
Backed by Primer and TheVentures.
Pogl © 2026