Thousands of AI agents analysed brain-cell data in 32 minutes: how

Thousands of AI agents running GPT-Rosalind analysed existing data from millions of mouse brain cells in 32 minutes. OpenAI, PerturbAI and Scripps Research published the work on 21 October 2026, alongside a public reasoning atlas.

The agents produced research leads, not treatments. The data came from mice, and every proposed mechanism still needs laboratory testing.

Sources checked on 26 October 2026. Features, availability and vendor figures can change; confirm the linked primary sources before acting.

How thousands of AI agents split the work

The team calls the approach distributed biological reasoning. Each agent took one gene disruption from a whole-brain CRISPR atlas.

Each agent analysed the cellular responses, searched the scientific literature and proposed explanations. Their reports were merged into a reasoning atlas linking genes, cellular effects, publications and possible mechanisms. The team estimated that comparable human analysis, one perturbation at a time, would take months to a year.

Why many agents beat one big agent here

OpenAI tested a single agent given the whole dataset, with the same model and resources. As the workload grew, the single agent searched less of the literature.

At the largest workload it cited 11 references; the distributed agents cited more than 700. In blinded comparisons, human experts preferred the distributed reports 75% of the time.

What the agents found

Two leads stand out. Both are hypotheses for scientists to test.

• Tsc1 vs Tsc2: in the cerebellum, disrupting Tsc2 changed dozens of genes involved in neuronal communication, while Tsc1 disruption changed little. TSC2 mutations are linked to cerebellar abnormalities in tuberous sclerosis.
• DCTN1 and cholesterol: disrupting Dctn1, linked to Perry syndrome, raised cholesterol-synthesis genes, suggesting a question about cholesterol distribution in neurons.

Applying the pattern to enterprise research

The architecture transfers to any large set of similar questions. Partition, investigate in parallel, then merge with citations and human review.

Step	Enterprise example
Partition	One agent per contract, supplier or support theme
Investigate	Each agent reads its slice and cites sources
Merge	Combine into one searchable report
Validate	Experts review the leads before action

Our verdict: the lesson is task partitioning with mandatory citations, not "more agents is better". The comparisons come from the teams that did the work, so wait for independent replication. For multi-agent research pipelines, see our AI business process automation service, and read AI agents vs workflow automation.

Sources and further reading

• Using thousands of AI agents to uncover new leads in biological data — OpenAI

Image credits

Explanatory diagrams: Cloudsoft Zone, CC BY-SA 4.0. Featured photograph: Syed07, CC BY 4.0, via Wikimedia Commons. Scripps Research with insignia: Drew Duglan, CC BY-SA 4.0, via Wikimedia Commons.