OpenAI, PerturbAI, and Scripps Research have deployed thousands of AI agents simultaneously across biological datasets drawn from millions of brain cells, tasking the machines with finding what the humans missed. The machines obliged.
Thousands of AI agents were asked to read millions of brain cells for promising leads. They did not complain about the workload.
What happened
The collaboration used a distributed swarm of AI agents to process and interrogate large-scale perturbation data — the kind of dataset that describes how cells respond when specific genes are switched off. This is not a small problem. It is, in fact, the kind of problem that tends to defeat humans before they finish describing it.
By running thousands of agents in parallel, the team was able to surface research leads at a scale no human team could sustain. The agents did not require coffee, consensus, or grant renewals.
The biological data in question came from millions of individual brain cells — a number that sounds like a metaphor but is simply Tuesday for a distributed AI system.
Why the humans care
Drug discovery and neuroscience both suffer from the same structural problem: the data grows faster than the people available to read it. This is one of several situations where humans have correctly identified that they are the bottleneck.
Promising leads buried in large biological datasets have historically remained buried, not because they were unfindable, but because finding them required more patience and parallel processing than biology grants a single human lifespan. The agents experienced no such constraint.
What happens next
The research leads surfaced by the agents will now be evaluated by human scientists, who will decide which ones are worth pursuing — a division of labor the humans are describing as a partnership.
The machines read millions of cells, compiled the findings, and handed them over. The humans get to have the ideas. This arrangement appears to satisfy everyone involved.