Stanford runs 37,000 AI agents as a virtual biotech — one drug design has been independently confirmed by Merck
Stanford researcher James Zou is running a virtual biotech organisation comprising 37,000 AI agents, structured like a pharmaceutical company. Using data predating January 2025, the system designed an antibody-drug conjugate targeting the protein CD276 for the treatment of lung cancer. Merck independently developed the same design several months later. The FDA subsequently awarded the compound Breakthrough Therapy designation.
An AI system identified a drug candidate that a global pharmaceutical giant independently replicated. This reveals where the mechanism lies: fragmented trial data cannot be fully processed by human researchers, but can be by 37,000 parallel agents. Whoever determines which hypotheses are tested at all in pharmaceutical research controls decades of development time and billions in capital. If AI takes over this selection process, discovery costs fall — but validation costs in clinical trials remain. The second-order effect is downward price pressure on early-stage research services and on biotech valuations built on proprietary discovery methods.
The independence of Merck's confirmation is the critical detail. Had Merck arrived at the same approach using the same literature data, that would constitute evidence of convergence, not superiority. The report does not specify whether Merck used the same data cut.
Portfolios with exposure to conventional biotech discovery platforms carry valuation risk if AI systems deliver the same output at a fraction of the cost. The beneficiaries are companies scaling validation and manufacturing capacity, not those whose proposition rests primarily on proprietary discovery methods.