Stanford's Zou puts agentic AI to work across biomed research

James Zou, associate professor of biomedical data science at Stanford Medicine, told his university's news office that AI agents now run through "almost every stage of our research — everything from refining research project ideas to reviewing manuscripts before we submit to a journal."

The distinction from a chatbot, in his framing: "If you think of a chatbot as akin to an AI brain that can think and talk, agents are like adding the arms and legs." Agents, he says, can "use and create their own tools, annotate their own datasets, generate their own hypotheses or make new discoveries."

His group's showcase projects include a virtual laboratory that assigns agents to run research operations, a virtual biotech company that "investigates potential drug targets and other areas of drug development," and Paper2Agent, which converts individual scientific manuscripts into agents that can be queried and can interact with other paper agents. A separate experiment tried to "revive" Albert Einstein and Richard Feynman as AI agents and had them debate scientific questions.

Zou is explicit about what the systems cannot yet do. "Agents can't work in the physical world yet, so it's crucial that the agentic scientific process includes validating agents' findings in a real lab," he says, adding that human expertise and oversight remain critical when testing hypotheses, drug candidates or other AI outputs for viability.