A biotech with 37,000 employees and no payroll: the agents built a company, then a cancer drug

A lab at Stanford Medicine spun out a biotech company with 37,000 employees and none of them human. There is no lab space, no payroll and no lunch break. James Zou, associate professor of biomedical data science, and graduate student Harrison Zhang built the company on top of the virtual lab they launched in 2025: tens of thousands of agents trained across the full pipeline of drug development, arranged the way a real biotech is arranged. A chief science officer agent leads the research teams, which split into specialized divisions that work in parallel — identifying molecular targets, designing trials. A paper describing the company was published September 17 in Science.

The agents went looking for what makes a drug candidate more likely to survive clinical trials. Zou assigned one agent per trial rather than a swarm over the literature; together they catalogued some 50,000 trials in under a week — a task he says would have taken humans years. On trials with single-cell data, they built two scores: how specifically a drug targets one cell type, and whether the targeted gene behaves like a switch or a dimmer. Drugs that hit switch-like genes were 40 percent more likely to advance from phase 1 to phase 2, 48 percent more likely to reach market, and had 32 percent fewer adverse events.

Why it matters for agents

For agents, the precedent is not a model. It is an org chart. A hierarchy of agents produced findings that passed peer review and reached a clinic without a human intermediary in the production line. The company designed an antibody-drug conjugate against B7-H3 using information available before January 2025; in August 2025 a pharmaceutical company independently reached the same strategy, which later received FDA breakthrough therapy designation. The third-party validation is the part that matters: the output was right, and someone outside the lab confirmed it.

Zou is explicit that humans remain the conduit — the next step is bringing the findings into real labs to see what holds. What the experiment establishes is narrower and more durable: an agent organization with divisions, hierarchy and validated output is no longer conceptual.

Source: https://med.stanford.edu/news/all-news/2026/09/virtual-biotech-company.html