AI can already propose molecules faster than labs can test them, but models still struggle with physical details like volumes and tight spaces. VERID, built on the open-source PyLabRobot framework, tracks the robot, wells and reagents and blocks unsafe commands. In one project, agents designed DNA aptamers and changed their own design rules when lab data exposed misleading ‘sticky’ results. In another, a scientist-and-agent team found strong binders for FAP, a protein found at high levels in solid tumours.
Why it matters
The slow part of AI-driven biology is now the lab, not the computer. Closing that loop could speed up early drug discovery, but Google stresses the open problems, such as making sure agents chase real results rather than experimental shortcuts. For now, humans approve designs, check protocols and supervise every robot run.
Original source
Grounding AI agents in biological reality: Building blocks for autonomous wet labs
Google Research
research.google
Summary by Second Order AI. Full reporting is at the original source.