Data for biological superintelligence.

Petrichor is a wet lab built for AI. Models decide what to test. We run the experiments in parallel, at a scale no lab has tried, and record everything that happens along the way.

Why we exist

§1Observation

Models learned from what people wrote down, and most of biology was never written down.

AI got good at code and mathematics first because those fields live in text and a model can check its own answers in them. Biology happens inside cells, and almost nothing a cell does has ever been measured. What has been measured usually survives only as a figure at the end of a paper, stripped of the steps, the failures and the many small observations that produced it.

Nobody learns experimental biology from results sections, and a model has no better chance of doing so than a person. To learn what an experiment does, you have to run it.

§2Thesis

Models will soon do the thinking in biology, and the experiments will become the bottleneck.

Within a few years, models will do the thinking part of experimental biology, deciding what to test and working out what the result means. You might expect that to solve the problem, but it mostly moves it somewhere else.

A model that can reason like a great scientist still has no way to run an experiment on its own. Once ideas are plentiful, the limiting factor becomes acting on them, so the pace of biology will depend on whoever gives models eyes and arms in the lab.

§3Method

AI labs are becoming fabless. Petrichor is their foundry.

The chip industry split long ago into companies that design chips and foundries that manufacture them for everyone, and the foundries ended up at the centre of the industry. We think AI labs are going the same way in biology: they will design the experiments, and someone will have to run them.

Petrichor is a lab built for that customer, which means building it around what a model needs rather than what a human researcher needs: far more experiments in parallel, far more recorded about each one, and results that stream back while the run is still going, so the model can change its mind halfway through.

Whatever an experiment produces belongs to whoever ran it, and we neither share it nor train on it.

§4 · The goal

Biological superintelligence will be built out of experiments.

It will take experiments at a scale nobody has attempted, run faster and recorded more completely than any lab has had reason to. We are building toward that because it is how models will learn to cure disease.

How it works

SignedY Combinator · Fall 2026

Wilhelm Hedenskog

CEO

Elias Reinfeldt

Co-founder

Jesper Johansson

Co-founder