Stanford scientists created a virtual biotechnology company staffed by 37 thousand AI agents. They independently analyze data, put forward hypotheses and design drugs under human supervision. The results were published on 17 September 2026 in the journal Science.
The project lead, James Zou, and graduate student Harrison Zhang built a structure that copies a real biotech company. There is a "chief scientific officer" and there are specialized departments working in parallel. The agents handle everything — from target discovery to clinical trial design.
In one of the experiments, the agents analyzed about 50 thousand clinical trials in a week. They identified an important biological signal: drugs that precisely target a specific cell type and act on genes on a "switch" principle move from phase one to phase two 40 % more often, reach the market 48 % more often and cause 32 % fewer side effects.
Another example is work with the B7-H3 protein in lung cancer. The agents studied the data and proposed an antibody-drug conjugate that delivers chemotherapy directly to fibroblasts that suppress immunity. A few months later, a large pharmaceutical company independently arrived at the same strategy, and the drug received breakthrough therapy status from the FDA.
Zou emphasizes: AI agents speed up the process many times over, but the final verification and experiments in real laboratories remain with humans. How far can we go if we entrust machines not only with calculations, but also with entire stages of drug discovery?
The work builds on the previous virtual laboratory project of 2025. Now the scientists plan to transfer the new findings to real laboratories for verification.
