Why a laboratory and not a company

When I describe Artemis Labs, the first question is almost always the same: what is the product?

There is no product. The laboratory does not sell anything, and it is not going to. Artemis Labs exists to work on one problem, agentic autonomy, and to start companies when some part of that problem matures into something the world can use. The companies have products. The laboratory has a thesis.

This is an unusual structure, so I want to write down why we chose it.

The honest reason is that research and products fail in opposite ways, and I have watched both failures up close.

Research without consequence drifts. A lab that never puts its ideas in front of users can convince itself of anything. The work stays elegant, the demos stay controlled, and the gap between what is claimed and what is real grows quietly until someone tries to depend on it. A great deal of agent research is in this state right now: impressive in papers, untouched by reality.

Products without patience narrow. A company has to grow, and growth rewards whatever works this quarter. Hard questions get deferred in favor of features. If autonomy takes a decade of structural work, a product company will not do that work, because no customer is asking for year seven of it; they are asking for a fix by Friday. This is not a flaw in companies. It is what companies are for.

The structure of Artemis Labs is an attempt to hold both pressures without letting either win. The laboratory carries the long program: coordination, long horizons, self-improvement, the unglamorous architecture of trust. The companies take pieces of that program and expose them to real users, real failures, and real consequences. What the companies learn flows back into the laboratory. What the laboratory builds flows out through the companies.

Verse is the first. It builds agentic employees, and in doing so it forces the most basic question in our program: can an agent hold a job? Not perform a task. Hold a job. Everything we believe about structure, memory, and accountability gets tested by that question every day, by people who have never read our thesis and just want their work done.

There will be more companies. Each will take up a piece of the problem once it has matured. Coordination between agents is the likely next candidate, though I have learned not to predict timing in public.

People sometimes point out that this is a slower way to build a large company. They are right. It is also, I believe, the only structure that can solve the problem rather than sell fragments of it. Autonomy is not a feature that a product team will crack in a roadmap cycle. It is a body of work, and bodies of work need institutions.

So: a laboratory. Small, patient, and connected to the world through the companies it creates. I expect it to look slow from the outside for a few years. I do not expect it to look slow after that.