Stanford puts 37,000 AI agents to work as a virtual biotech
The agents analysed almost 56,000 clinical trials in under a week, and none of their proposals has yet been tested in a laboratory.
Stanford researchers ran a drug discovery pipeline on 37,000 AI agents organised like a biotech company, with a chief science officer agent directing specialised divisions, according to a paper published in Science on Thursday. The lead author is the graduate student Harrison Zhang.
The agents annotated outcomes from 55,984 clinical trials in under a week, Stanford said. Drugs aimed at genes switched on in specific cell types were 48 percent more likely to reach market and carried 32 percent fewer adverse events than drugs with broader activity, the paper reports.
In a separate test the agents used data from before January 2025 and proposed an antibody-drug conjugate against B7-H3, a protein they found on fibroblasts near lung tumour cells. Stanford said a drugmaker independently arrived at the same strategy.
That drug is ifinatamab deruxtecan, which Daiichi Sankyo discovered and develops with Merck. The Food and Drug Administration granted it “breakthrough therapy” designation in August 2025, on the strength of a phase 2 trial in 187 patients. The trial was running well before the agents' data cutoff.
None of the proposals has been tested in a laboratory or in patients. Zou said the next step is to take the findings into real labs and see how many hold up. No independent group has reproduced the work, and the paper's authors are its only source for the figures.