What three years of failed experiments taught us.

Lumin Labs has shipped a lot of things that no one will ever see. Prototypes that worked but solved a problem nobody had. Models that were technically impressive and practically useless. Interfaces so clever they were unusable. We kept almost none of them — and we would not trade any of them.

The value of a failed experiment is not the artefact, it is the narrowing. Each one removes a wrong answer from the search space and sharpens the question for the next attempt. Over three years that adds up to a kind of institutional intuition that is impossible to acquire any other way.

Cheap to try, cheap to kill

The trick is making experiments small enough that killing one costs nothing emotionally. When a prototype takes a week, abandoning it is a Tuesday. When it takes six months, abandoning it is a tragedy — and teams will keep a doomed project alive far too long to avoid that feeling.

So we optimise for the speed of the loop, not the polish of any single attempt. The goal of an experiment is to produce a decision, and a decision arrives just as well from a scrappy demo as from a finished one. Most of the time, it arrives faster.