Stanford University researchers have unveiled Biomni, an advanced AI-powered research agent designed to function as a full-fledged biomedical co-scientist. Published in the journal Science, the system moves beyond generative chatbot capabilities by executing complete research workflows. Led by senior author Jure Leskovec, professor of computer science at Stanford School of Engineering, and developed by lead researcher Kexin Huang, Biomni automates the mechanical components of scientific discovery, allowing researchers to input problems in natural language and receive structured analytical outputs. Biomni is specifically engineered for biomedical sciences, drawing its training corpus from the breadth of full-text papers, code, and datasets publicly available on bioRxiv. The platform integrates 150 specialized biomedical tools, 105 software packages, and 59 databases covering all 25 subdomains defined by bioRxiv, from genetics to neurology. Rather than merely generating text, the agent reads scientific literature, formulates hypotheses, selects appropriate datasets and computational tools, writes executable code, interprets results, and proposes subsequent experimental phases. Crucially, it maintains full traceability, providing complete citations and workflow logs to ensure scientific rigor and reproducibility. The system significantly compresses timelines traditionally required for scientific ideation and data processing. In a documented trial, a researcher uploaded over 450 files containing continuous glucose monitoring, dietary, and activity data. Biomni cleaned, unified, and analyzed the information in forty minutes, identifying plausible correlations between food intake and body temperature, a process estimated to require sixty or more hours of human labor. Stanford researchers emphasize that the bottleneck in modern biomedical innovation is not a lack of intelligence, but the labor-intensive mechanics of literature review, data homogenization, and coding. Biomni addresses this friction, enabling rapid hypothesis generation and iterative experimentation. Despite its autonomy, the development team stresses that Biomni operates strictly within a human-in-the-loop framework. The AI handles data synthesis, pattern recognition, and computational legwork, while scientists retain authority over strategic direction, experimental design, and critical judgment. Huang, who founded a startup to commercialize the technology, noted that the tool augments rather than replaces human expertise, positioning scientists as directors of AI-assisted inquiry. A prototype version of Biomni is already deployed across more than ten thousand academic and industrial laboratories, establishing it as the most widely adopted AI co-scientist system in biomedicine. The research involved collaborators from Stanford Medicine, the University of Washington, the Arc Institute, Genentech, Princeton University, and the University of San Francisco. Development was supported by grants from the National Science Foundation, Stanford Data Science Applications, the Wu Tsai Neurosciences Institute, the Stanford Institute for Human-Centered AI, the Chan Zuckerberg Initiative, and corporate partners including Amazon, Genentech, GSK, Hitachi, and SAP. By offloading computational overhead to a traceable, domain-specialized agent, Biomni aims to accelerate therapeutic discovery and expand the scale of biomedical innovation.