Researchers at Stanford Medicine have created a virtual biotechnology company that operates entirely through artificial intelligence (AI). This company, developed by associate professor of biomedical data science James Zou and graduate student Harrison Zhang, has no human employees, no physical lab space, and no traditional payroll. Instead, it uses 37,000 AI "scientist agents" to perform tasks typically handled by human researchers, such as identifying molecular targets for drugs and designing clinical trials. The structure of the company mimics that of a real biotech firm, with specialized divisions led by a chief science officer agent. The virtual biotech has already made a significant discovery: it identified a biological signal that can predict which drug candidates are more likely to succeed in clinical trials. This finding was published in the journal Science on September 17, 2026, with Zou as the senior author and Zhang as the lead author. The AI agents analyzed over 50,000 clinical trials in less than a week, a task that would have taken human researchers years. The agents developed two scoring systems—one to evaluate how specifically a drug targets a certain cell type, and another to measure bimodality, a property that indicates whether a gene's activity is more like a light switch (on-off) or a dimmer. Trials with high scores in both categories had better outcomes, with drugs targeting switch-like genes showing greater chances of progressing through clinical trials and reaching the market. To test whether an all-AI company could design a new drug with real-world potential, Zou and his team directed the AI agents to focus on a protein called B7-H3, which has been of interest to lung cancer researchers. The AI found that B7-H3 was highly expressed in fibroblasts—cells found in connective tissue and often near tumor cells. Using this information, the AI scientists designed an antibody-drug conjugate, a type of therapy that targets cells with high levels of B7-H3 and delivers a toxic chemotherapy payload directly to them. This design was proposed using data available before January 2025. Months later, in August 2025, a private pharmaceutical company independently developed the same strategy against B7-H3. That therapy received a Food and Drug Administration breakthrough therapy designation, which helps accelerate the development and approval of promising drugs. While the virtual biotech has made impressive strides, Zou emphasized that human researchers will still be needed to conduct physical experiments and validate the AI's findings. The next step is to transition the new discoveries from the virtual company into real-world laboratories and test how many of them hold up in practice. This hybrid approach, combining AI-driven research with human experimentation, may represent a new frontier in drug development, offering the potential to speed up the discovery of effective treatments while maintaining scientific rigor.