A small company, built around the work rather than around a single idea.

basically AI started because a few of us wanted to build cool things and see what happened. Turns out that's still basically the plan.

We don't organize around one model, one modality, or one grand theory about where AI is headed. Some weeks that means training a language model. Other weeks it means cleaning up a dataset, building a tool we needed ourselves, or chasing an idea that might not go anywhere at all. We follow whatever's interesting, not a roadmap.

What stays constant is how we work. We build things we'd actually use. We're upfront about what didn't pan out, because that's usually the most useful part. And we try to keep the gap between "fun idea" and "real thing" as short as we can, without cutting corners that matter.

We're not betting the company on one category of AI or one architecture. The areas below are where our attention is right now, not a permanent list, and not a promise about what we'll work on next.