PerturbAI and OpenAI, with collaborators from Scripps Research, UC San Diego, and HHMI, disclosed a distributed AI reasoning framework for interpreting large experimental datasets and generating testable hypotheses. In a bioRxiv preprint, they described thousands of parallel agents that analyze PerturbAI’s whole-brain atlas and integrate literature context into a “reasoning atlas.” The system reportedly recovered prior conclusions, predicted held-out perturbation effects, and identified relationships across perturbation conditions, with analysis completed in minutes. In blinded comparisons against single-agent outputs, human experts preferred the reasoning-agents’ interpretations 75% of the time. The collaboration is now expanding toward joint drug-target discovery, aiming to connect causal biological data with automated hypothesis generation. The emphasis on distributed agents and literature retrieval signals a shift from purely predictive models toward reasoning workflows that can scale scientific synthesis.