AI Scientists Outnumber Humans in Stanford's Virtual Drug Company
Finn's Take· TL;DR- Virtual biotech with 37,000 AI agents analyzed decades of drug research in days, discovering that switch-like gene targets are more likely to succeed clinically.
- AI agents independently proposed same lung cancer therapy strategy that major pharma company validated months later, suggesting AI can match human expert drug discovery.
- System still requires real-world lab testing of proposed targets, but demonstrates AI's potential to integrate fragmented biomedical evidence and accelerate drug development.
A Drug Company With No Employees
The latest company to spin out of a Stanford Medicine lab is a biotech undertaking with 37,000 employees — and none of them are human. There's no lab space, no lunch breaks, and no payroll. Instead, there are tens of thousands of artificial intelligence agents, working around the clock to do something the pharmaceutical industry has struggled with for decades: find better drugs, faster.
It's an AI-powered virtual biotech company that's the brainchild of associate professor of biomedical data science James Zou, PhD, and graduate student Harrison Zhang. Drug development requires evidence integration across biological scales and modalities — but relevant tools are fragmented. The Virtual Biotech addresses this by organizing AI agents modeled on a real drug-development company, with divisions spanning target discovery, safety assessment, modality selection, and clinical development.
Crunching Decades of Research in Days
Roughly 90 percent of drugs that enter clinical trials never reach the market. That's often because promising results in the lab don't translate to patients, or the drug causes dangerous side effects not caught earlier. Part of the problem is that the evidence that could help catch these issues is scattered across disciplines and formats, making it hard for any single team to weigh it all.
In one test, the agents went through more than 50,000 clinical trials in less than a week and found patterns linked to which drugs were more likely to succeed. That's way past human scale. Drugs that targeted switch-like genes were 40% more likely to advance from phase 1 to phase 2 trials, were 48% more likely to reach market, and had 32% fewer adverse events compared with those that had a broader spectrum of activity. These patterns persisted across a variety of conditions, including cancers, brain diseases, heart diseases, kidney and lung conditions, and more.
The Lung Cancer Breakthrough
To test whether an all-AI company could design a new drug capable of helping people, Zou and his team turned the agents' attention to a protein that lung cancer researchers have long eyed — B7-H3. The agents analyzed relevant data from a variety of studies and biomedical data repositories and found that B7-H3 was highly expressed in fibroblasts, cells found in connective tissue that often live near tumor cells.
The agents looked more closely at communication between cells and found that fibroblasts expressing B7-H3 seemed to be signaling to nearby immune cells and suppressing their activity, effectively cloaking the tumor from normal immune defenses. The system then proposed a therapy that would tag cells expressing B7-H3 with an antibody to help direct a toxic chemotherapy drug to them. What happened next was the study's most striking moment: the virtual biotech came up with its solution based solely on data available before January 2025, but in August of that year a major pharmaceutical company arrived at the same strategy independently, when its B7-H3-targeted therapy ifinatamab deruxtecan received FDA breakthrough therapy status.
What This Means for the Future of Medicine
"This was really exciting as an independent, third-party validation that's consistent with the effects and the design proposed by the virtual biotech," Zou noted. For the study, Zou's team used versions of Claude — developed by Anthropic — as the underlying large language model powering the agents. But Zou says that any advanced LLM will do, including open-source models that researchers can run on their own computers.
The results are promising, but the work is still early. The agents can find possible drug targets, but those targets still need to be tested in a real lab. Still, the implications are hard to ignore. "Our idea was to see how far we could push this. Could we create a biotech company that takes on everything from looking for drug targets all the way to designing clinical trials?" Zou said. If the answer turns out to be yes, the way humanity discovers life-saving medicines may never look the same again.