Stanford’s Virtual Biotech Deploys 37,000 AI Agents for Drug Discovery
Stanford researchers built a virtual biotech company staffed by up to 37,000 AI agents that collaborate on drug discovery tasks. The system identified drug target types more likely to succeed in trials and proposed a lung cancer treatment that a major drugmaker also later pursued. It uses a virtual chief scientific officer to delegate work to specialized agents with access to databases and clinical trial data.
Stanford researchers created a virtual drug company in which as many as 37,000 AI agents fill roles resembling real biotech divisions. A virtual chief scientific officer receives a human query, then assigns work to specialist agents that consult databases and tools covering target discovery, safety, delivery, and prior trial results.
In a Science paper, the team tested this setup by asking it to extend research on genetic evidence and trial success. Agents reviewed 37,075 Phase II and III trials in about six hours, then combined tissue-expression data with a scoring method. They found drugs aimed at genes with on/off behavior and restricted to a few cell types had better odds and fewer adverse events.
This approach could speed early drug discovery and help researchers prioritize targets with stronger trial prospects, potentially benefiting patients who need new therapies. Pharmaceutical teams and regulators may use such systems to screen evidence, but AI-generated hypotheses may still require experimental and clinical validation. If widely adopted, it may shift some biotech labor toward oversight and data quality, while raising questions about transparency, accountability, and reliance on automated analysis.