Stanford University researchers have introduced Biomni, an artificial intelligence system engineered to function as an autonomous biomedical research agent. Led by computer science professor Jure Leskovec and developer Kexin Huang, the system was detailed in a recent paper published in the journal Science. Unlike general-purpose generative models, Biomni is purpose-built to execute complete scientific workflows, transforming simple natural-language research queries into structured, data-driven experimental plans. Biomni operates by integrating literature synthesis, hypothesis formation, dataset selection, code generation, and result interpretation into a single automated pipeline. To achieve this, the development team embedded 150 specialized biomedical tools, 105 software packages, and 59 databases, drawing from the full-text repository bioRxiv. The system covers all 25 recognized biomedical subdomains, enabling cross-disciplinary analysis. Crucially, Biomni maintains full citation tracking and methodological transparency, addressing a critical gap in reproducibility that often plagues AI-assisted research. The platform demonstrates substantial efficiency gains in real-world applications. In one documented case, a researcher uploaded over 450 files containing continuous glucose monitoring, nutritional, and activity data. Prompted to identify plausible hypotheses, Biomni cleaned and unified the dataset, generated visualizations, and detected correlations between food intake and body temperature in forty minutes. Leskovec estimates the same task would require a human researcher sixty or more hours. The system directly addresses a structural bottleneck in modern science: as the volume of published data expands, the pace of discovery has paradoxically slowed due to the mechanical burden of literature review and data homogenization. Biomni automates this labor, compressing weeks of preliminary work into minutes. Despite its automation capabilities, the developers emphasize a strictly collaborative model. Huang and Leskovec clarify that Biomni is designed to augment, not replace, human scientists. The AI handles repetitive computational tasks and pattern recognition, while researchers retain authority over strategic direction, experimental design, and scientific judgment. This division of labor preserves the creative and analytical value of human expertise while eliminating procedural friction. A prototype version of Biomni is currently deployed across more than ten thousand academic and industry laboratories, marking it as the most widely adopted AI research partner in biomedicine. Huang, who recently completed his doctorate in Leskovec’s lab, is leveraging the system’s traction to launch a commercialization initiative aimed at broadening access and integrating advanced functionality. The Stanford team views Biomni not as a replacement for traditional methodology, but as a foundational infrastructure update that allows biomedical researchers to scale innovation without being constrained by data management overhead.