OpenAI says it has put thousands of AI agents to work on one of biology's largest datasets - and finished in minutes what it estimates would have taken human analysts months to a year. In a post published on October 21, 2026, the company describes a collaboration with PerturbAI and Scripps Research that applied a technique it calls distributed biological reasoning to a whole-brain CRISPR atlas: existing experimental data showing how individual gene disruptions affect different cell types across millions of mouse brain cells. The agents, powered by a model OpenAI calls GPT-Rosalind, completed their analysis in 32 minutes and surfaced candidate research directions for diseases including Alzheimer's and Parkinson's.
Divide the work, deepen the reasoning
The team split the analysis among thousands of agents. Each one took responsibility for a single gene disruption: it examined how cells responded, searched the scientific literature for relevant findings, and proposed an explanation for what it saw. The individual reports were then merged into what OpenAI calls a reasoning atlas - a linked structure connecting genes, cellular effects, relevant publications and candidate biological mechanisms - which the company says is now available for researchers to explore.
Why not hand the whole dataset to one agent? OpenAI says it tested exactly that while holding the model and compute budget fixed. As the workload grew, the single agent read progressively less of the literature, citing only 11 references at the largest workload, while the distributed agents collectively cited more than 700. In blinded comparisons, human experts preferred the distributed agents' reports 75% of the time. OpenAI also reports that the agents reproduced the conclusions of earlier human analyses while adding new leads, and that their ratings of novelty and scientific value aligned with judgments from blinded human reviewers.
The reasoning atlas is available now, and OpenAI frames it as an exploration tool rather than an answer key: scientists can compare related genes, investigate unexpected responses, and identify promising leads to pursue with their own expertise. New discoveries, the post argues, often come from connecting familiar ideas across fields - and AI can scale those connections by linking fresh experimental data with knowledge scattered across disciplines.
The leads the agents surfaced
Two examples anchor the announcement. The first concerns TSC1 and TSC2, two genes that both regulate mTOR, a protein central to cell growth, metabolism and aging. Disrupting either gene produced similar effects in several groups of neurons - but in the cerebellum, which coordinates movement, Tsc2 disruption altered the activity of dozens of genes involved in neuronal communication, while Tsc1 disruption produced little change. That divergence matters because in people with tuberous sclerosis, cerebellar abnormalities are strongly linked to TSC2 mutations. Medicines that reduce mTOR activity already treat certain brain tumors and seizures, but the learning and behavioral difficulties of the disease remain poorly addressed - and the agents' finding points at a specific group of neurons and cellular changes where existing drugs might be tested.
The second example is DCTN1, a gene whose mutations cause Perry syndrome, a neurodegenerative disease with Parkinson-like symptoms. DCTN1 helps move cargo inside neurons. In the atlas, disrupting it increased the expression of genes involved in making cholesterol - and the agents connected that observation to research suggesting that cells produce more cholesterol when its transport breaks down. The resulting hypothesis: DCTN1 disruption might alter cholesterol distribution within neurons, and that change might contribute to neurodegeneration - raising the follow-up question of whether restoring normal distribution could be protective. OpenAI says human experts judged this hypothesis novel, consequential and worth following up.
What the scientists say
"We build cell atlases so that other scientists can make discoveries we haven't imagined yet," said Sarah Teichmann, Chair in Stem Cell Medicine at the University of Cambridge and founding co-chair of the Human Cell Atlas, quoted in OpenAI's post. "AI could help more researchers explore these resources and turn unexpected observations into new questions to test." Tomasz Nowakowski, associate professor of neurological surgery at UCSF, added: "Sometimes the clues are already there, scattered across different fields - we just haven't connected them."
The context to keep
These are leads, not results. Everything in the announcement comes from OpenAI's own account of the collaboration, and the company is explicit that the agents' role is to prioritize experiments: each proposed mechanism still has to be tested in a lab. The comparison to "months to a year" of human analysis is the research team's estimate, not a measurement. And an analysis of existing mouse-brain data is a long way from a treatment. What the work demonstrates is narrower and still significant: that large experimental datasets hold findings their creators never had time to look for, and that swarms of focused agents - each reading the literature on one small question - can surface them at a speed and breadth a single analyst, human or machine, does not match.