Stanford's Biomni AI Agent Now Powers 15,000 Scientists' Research

A team from Stanford University has published a peer-reviewed paper in Science describing Biomni, an autonomous AI agent that independently plans and carries out biomedical experiments. The system, available free of charge to researchers, had already made waves across several fields of medicine before it formally cleared peer review.

What Biomni actually does

Biomni combines large language models with retrieval-augmented generation (RAG) and the ability to write and run its own code. In practice, that means the system doesn't follow a fixed script. Instead, it draws up its own plan of action for a given research question, reaches for the appropriate tools and databases, and then carries out the successive steps of the experiment itself.

Rather than waiting for step-by-step instructions, the agent receives a question in natural language and decides on its own which resources to use. Its creators compare it to a lab assistant capable of independently designing an experiment, finding the necessary data in the scientific literature, and running the analysis without supervision at every stage.

Biomni can understand a simple question like: why do these patients respond differently to the same drug. Then it gets to work and does a good chunk of the scientific labor itself - Kexin Huang, creator of Biomni, former PhD student in Jure Leskovec's lab at Stanford

If you think of an AI agent as a carpenter, a carpenter without tools can only talk. With Biomni, we're giving that carpenter a toolkit so it can actually build - Jure Leskovec, professor of computer science at Stanford

Scale of adoption and early wins

Since its early release, the system has built up a community of 15,000 researchers using it free of charge. Together, they have automated more than 100,000 research processes, from prioritizing genes in rare diseases, to repurposing existing drugs, to microbiome analysis and molecular cloning design.

The most concrete proof of effectiveness so far is the Biomni-AD variant, fine-tuned for Alzheimer's research. It won a $1 million prize awarded by the Alzheimer's Disease Data Initiative, an organization that pools data and funding for research into the disease. It's the first tangible payoff from the agent's work, rather than just a benchmark result in a paper.

Part of a broader trend

Biomni fits into a growing wave of AI tools meant to speed up scientific discovery. OpenAI's head of research, Mark Chen, recently predicted that artificial intelligence would drive Nobel Prize-level discoveries within the next two years, especially in biology and medicine. Such forecasts are often met with skepticism, but concrete examples of accelerated research are already emerging.

One example is the drug rentosertib, developed by Insilico Medicine, whose development time was cut from the typical three to four years down to 13-18 months thanks to generative AI. Biomni goes a step further than tools that merely assist with drug design, since it covers the entire research workflow, from forming a hypothesis to analyzing results.

What it means for science, and for Poland

For research labs, including Polish medical universities and institutes, tools like Biomni lower the barrier to advanced analyses that previously required a dedicated bioinformatics team. The system is open source and the interface is free to use, meaning access isn't limited to the wealthiest institutions.

At the same time, autonomous research agents raise questions about accountability for faulty conclusions or poorly designed experiments, especially when the results feed into clinical decisions. The Science publication, having gone through full peer review, is meant to address doubts about the reliability of systems that had previously existed mainly as preprints.

The creators say they plan to keep developing the tool and expand the number of supported databases and research protocols. Kexin Huang, the project's lead author, has already founded a company to commercialize the technology, suggesting that Biomni is moving quickly from academic project to market product.

Sources: Cyfrowa.rp.pl (cyfrowa.rp.pl), Stanford Report (news.stanford.edu)