Stanford Medicine scientists have created a new kind of laboratory—one where virtual scientists, powered by artificial intelligence, collaborate to solve complex biological problems. Led by James Zou, PhD, associate professor of biomedical data science, the project introduces a team of AI agents that work together like seasoned researchers in a real lab setting.

“Good science happens when we have deep, interdisciplinary collaborations where people from different backgrounds work together,” Zou said. “Often that’s one of the main bottlenecks and challenging parts of research.” Inspired by recent advances in language model-based AI agents, Zou and his team developed a system that emulates human scientific collaboration.

These agents do more than answer questions. They retrieve data, operate software tools, and communicate in human language. The system represents a form of agentic AI, where different AI components work in tandem to solve problems.

Zou said he designed the system to imitate how top scientists think—asking questions, testing hypotheses, and providing expert feedback. “The virtual lab could help expedite the development of solutions for a variety of problems,” he added.

To demonstrate the lab’s potential, the researchers tasked the AI team with designing a new vaccine strategy for SARS-CoV-2, the virus that causes COVID-19. Within a few days, the AI agents proposed using nanobodies—small, simplified fragments of antibodies—as a more promising solution.

“From the beginning of their meetings the AI scientists decided that nanobodies would be a more promising strategy than antibodies — and they provided explanations,” Zou said. Their reasoning centered on nanobodies' smaller size, which improves computational modeling and protein design.

The AI-generated nanobody designs were then synthesized by John Pak, PhD, and his team at Chan Zuckerberg Biohub. The results were promising: the nanobodies were experimentally stable and bound tightly to a SARS-CoV-2 variant—more effectively than previously lab-designed antibodies. They also avoided off-target binding and even showed strong interaction with the original Wuhan strain of the virus, suggesting broader vaccine potential.

Now, the researchers are feeding experimental data back into the AI lab to further refine the molecular designs.

A paper detailing the study’s findings was published on July 29 in Nature. The senior authors are James Zou, PhD, and John Pak, PhD, a scientist at Chan Zuckerberg Biohub. Kyle Swanson, a computer science graduate student at Stanford University, is the lead author.

The virtual lab mirrors a traditional scientific process. A human researcher presents a problem to the AI principal investigator (PI), who then assembles, or creates, a team of specialized agents. For the COVID-19 project, these included an immunology agent, a computational biology agent, and a machine learning agent. One agent always acts as a critic, offering counterpoints and flagging potential issues.

The agents meet regularly in virtual sessions—some in groups to generate ideas, others one-on-one with the AI PI to discuss ideas. Unlike human meetings, these sessions last just seconds or minutes and can occur in parallel. Additionally, AI agents don’t require rest or breaks, allowing them to conduct multiple research discussions continuously and sustain rapid progress around the clock.

“By the time I’ve had my morning coffee, they’ve already had hundreds of research discussions,” Zou said during the RAISE Health Symposium where he presented the work.

The agents also make tool requests, which the researchers fulfill. For instance, they integrated AlphaFold, a protein-structure prediction tool, to support the agents’ work.

The system runs independently, with minimal human oversight. Beyond the initial prompt, the primary constraint placed on the AI lab is budget-related, preventing the agents from pursuing extravagant or impractical ideas that can't be tested in a real-world laboratory.

“I don’t want to tell the AI scientists exactly how they should do their work. That really limits their creativity. I want them to come up with new solutions and ideas that are beyond what I would think about,” Zou said. Still, each interaction is recorded, allowing researchers to monitor progress and intervene if needed.

Following the success of the COVID-19 project, Zou’s team is now applying the virtual lab to other scientific challenges. They’ve also developed new agents that act as advanced data analysts, capable of reevaluating previously published biomedical papers.

“The datasets that we collect in biology and medicine are very complex, and we’re just scratching the surface when we analyze those data,” Zou said. “Often the AI agents are able to come up with new findings beyond what the previous human researchers published on. I think that’s really exciting.”

The research was supported by the Knight-Hennessy Scholarship, the Stanford Bio-X Fellowship, and institutions including Stanford’s Human Centered AI Institute and the Department of Biomedical Data Science.

Q: What is Stanford’s virtual lab?
A: It is a team of AI agents designed to simulate a real research lab, led by an AI principal investigator and specialized scientific agents.

Q: How does the virtual lab generate research ideas?
A: A human provides a research prompt, and the AI PI assembles a team of agents to explore, critique, and solve the problem collaboratively.

Q: What breakthrough did the lab achieve with COVID-19?
A: The AI lab proposed a nanobody-based vaccine design, which was later validated by real-world experiments.

Q: How do the AI agents operate?
A: They meet in virtual sessions, use scientific tools like AlphaFold, and can conduct hundreds of discussions in minutes.

Q: What are the future applications?
A: The researchers plan to apply the virtual lab to analyze existing datasets, revisit past studies, and address new scientific questions.

Stanford’s AI-powered virtual lab marks a significant shift in how scientific research can be conducted. By enabling intelligent, autonomous agents to simulate expert-level collaboration, this system dramatically accelerates the pace of discovery—turning what would normally take weeks or months into a matter of days or even hours.

The technology doesn’t just replicate human workflows—it reimagines them. These AI scientists can think critically, debate ideas, and propose novel hypotheses, all while operating at machine speed and scale. The success of their nanobody vaccine design underscores their ability to deliver actionable, real-world results—not just simulations or theoretical outputs.

This approach also introduces a new paradigm for democratizing access to research. With the right infrastructure, teams across the world could leverage similar AI labs to tackle urgent scientific questions, even without large, resource-heavy physical facilities.

As AI agents become more capable, collaborative, and independent, they could shift the role of human scientists—from direct problem-solvers to strategic guides and evaluators—freeing researchers to focus on ethics, creativity, and big-picture thinking.

By bridging AI reasoning with experimental science, Stanford’s virtual lab doesn’t just point to faster research—it signals a future where the boundaries between human and machine-led discovery begin to blur.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.