Project Beautiful
David Deutsch, a theoretical physicist, in his lecture "Why Are Flowers Beautiful?" argues that beauty is not subjective preference.
Flowers had to attract insects across a huge evolutionary gap. No shared genes, no shared signals. Species-specific tricks wouldn't work. So flowers evolved toward standards any observer could recognize. A universal language, not a local one.
This matters because it turns beauty into something to discover, not something to prefer. Just like a mathematician looks for an elegant proof, an artist looks for standards that are "there to be found." Beauty is a tool for truth. Not a matter of taste.
What is Design
Design is the practice of finding those standards and applying them. Every era did this work. Swiss International Style found clarity through grids and typographic hierarchy. Art Deco found rhythm through geometry and repetition. Italian modernism found how form follows function without losing warmth.
None of these were invented out of taste. They were found through decades of trial, criticism, refinement, and remixing, the same process Deutsch describes for beauty itself.
Years of experiments, failures, and corrections, distilled into something you can recognize in seconds. Design history is this record. It's the evidence for what actually works, tested across generations of designers who each pushed the search further.
Our Problem with AI
Frontier models can describe any of these movements in detail. Ask about Bauhaus and you get an essay. Ask them to design like it, and you get the same result every time. The average of the last 20 years of the web.
Models are optimized to write and to code, not to reason like designer and pull from history coming up with new styles. They can browse references if you set the task for that, but left alone they default to the most used style. And even the best of them don't know your history. They can decompose a project and copy it, but they have zero awareness of the years of experiments that shaped why you make the choices you make.
The result is slop. Remixing patterns with no depth underneath to remix from.
What We Do
We're training a family of models on decades of design history, movements, and principles, so models can internalize what's true, not just describe it.
We think this is the only way to unlock real creative range. Frontier labs are racing to make models more agentic, more autonomous, compensating for all of the knowledge they can’t put in weights.
That's the opposite of how design actually works. Creativity is often in the moment. Creativity relies on practical knowledge and experience. Creativity heavily relies on trying dozens of options before you land on the right one. With the current approach, models will never unlock true creativity.
Models need the fundamentals inside their weights, like any real designer, before they can break the rules. That's what we're building.
Good design says: I thought about you. We're building models that can say that too.