OpenAI scales agents for biology

OpenAI says it worked with PerturbAI and Scripps Research to use thousands of AI agents to analyze experimental data from millions of mouse brain cells. The agents examined gene disruptions, looked for related findings in scientific literature, and connected the results into a map of possible biological explanations. The analysis took 32 minutes. OpenAI estimates that comparable work by people, done one gene disruption at a time, would take months to a year. The team’s report describes the output as promising leads for scientists to investigate, not proven explanations or treatments.

The practical point isn’t that AI has solved a disease. It’s that a large team of agents can search many parts of a complex dataset and connect findings that might be easy to overlook. In this example, the agents raised questions about differences between two disease-linked genes and a possible connection between cholesterol regulation and neurodegeneration. Scientists still need to test those ideas. For a business, the broader lesson is that AI may help teams make more use of information they already have, especially when it’s spread across large datasets and research sources. But the result is a set of leads for people to judge, not a decision to hand over to software. This is the kind of thing we build into an AI command centre, where people can see how AI tasks fit together and where human review remains necessary.