LabAgent Uses AI Agents to Preserve and Extend Laboratory Research

Summary

Scientific research often depends on methods passed from one generation of lab members to the next, but graduation and staff turnover can leave those methods difficult to reproduce. The paper introduces LabAgent, a harness designed to support a laboratory’s continuing work by reproducing established methods and helping researchers pursue new discoveries. It uses two mechanisms: one ensures that stored skills can be executed and verified, while the other records corrective procedures and prior experience so similar errors can be corrected or avoided later. The authors apply LabAgent to drug property prediction, biomedical problem analysis, protein variant effect prediction, and statistical genetics. In each domain, LabAgent ranks first against the commercial generalist agents used for comparison. The system also accurately reproduces a figure from a published paper. The results suggest that an AI-agent framework can preserve laboratory knowledge across personnel changes while integrating that knowledge to support reasonable extensions of research.