Researchers at Stanford University have developed a biomedical artificial intelligence agent that integrates more than 150 bioinformatics tools, 59 curated databases, and more than 100 software packages into a single research environment.
Called Biomni, the cloud-based platform can plan and perform computational tasks ranging from sequence analysis to experimental protocol design. Scientists submit a request through a chat interface, review the plan Biomni produces, and monitor or redirect its work as it proceeds, according to a Stanford Institute for Human-Centered Artificial Intelligence overview.
The project addresses a familiar laboratory problem: researchers often move among separate databases, analysis packages, spreadsheets, and custom scripts to complete one study. Biomni’s developers designed the agent to identify the appropriate resources for a task, write code, execute an analysis, and record the steps in the resulting workflow.
Testing the agent on realistic research tasks
The developers evaluated Biomni on biomedical question-and-answer benchmarks and eight research scenarios that were not used during development. The results are described in a bioRxiv preprint, which had not completed peer review at the time of the Stanford report.
In one case, the agent received 458 spreadsheet files containing continuous glucose-monitoring and body-temperature data from 30 participants. Biomni generated and executed a 10-step analysis, inferred meal events from glucose spikes, and examined temperature patterns before and after meals. Other demonstrations involved genomic sequence data and the development of wet-lab protocols.
Stanford reported that more than 15,000 scientists used the open-source platform to run approximately 100,000 workflows during its first nine months. The developers also acknowledged important limitations: Biomni did not cover every biomedical field and performed less reliably on tasks requiring nuanced clinical judgment, novel experimental reasoning, or deep biological synthesis.
What an AI agent changes for lab operations
For laboratory managers, an integrated agent raises questions that extend beyond computational performance. A lab evaluating this type of platform would need to establish which datasets staff can upload, how users verify cited sources and generated code, and what records must accompany an AI-assisted analysis. Managers would also need to determine whether the platform’s logs meet institutional requirements for traceability, retention, and access control.
The ability to document a computational workflow could support reproducibility, but documentation alone does not validate the scientific choices within it. Labs would still need qualified personnel to review data transformations, software versions, statistical methods, assumptions, and proposed experimental conditions. Those expectations align with broader guidance on preparing laboratory data for AI and defining responsibility for AI-assisted work.
The original codebase remains open source, although the public platform migrated to Phylo, a company spun out of the Stanford AI Lab. That transition gives managers another consideration: whether to operate open-source components internally or use a hosted service, with different implications for technical support, cybersecurity, cost, and data governance.
Biomni demonstrates how AI agents could consolidate fragmented digital workflows. Its usefulness in a regulated or quality-controlled laboratory, however, will depend on local validation, human review, and a clear record of how each output was produced.
This article was created with the assistance of Generative AI and has undergone editorial review before publishing.
