Google’s wet-lab AI push moves agents from screens to robots
Google Research says its new lab-agent work aims to make AI test, learn and iterate in physical biology labs.
Google Research says frontier AI could move from biology ideation and analysis into running wet-lab experiments, with humans still driving the scientific agency. In a new post, the team describes three projects aimed at closing the gap between in-silico AI exploration and real-world lab work: VERID, a semantic digital twin for lab robots; closed-loop aptamer discovery; and prototypes for de novo protein design. Google says VERID tracks robot geometry, well volumes, reagent identities and pipette-tip inventory, then supports dry-run simulation, hardware safety checks and synchronized video/action logs. In aptamer work across three design-test-iteration cycles, agents designed DNA sequences, screened them against target proteins and used sequencing data for the next cycle. Google says confirmed on-target hit rates reached 33.5% for myoglobin and 42.1% for α-thrombin after agents adapted to different wet-lab failure modes.
Why it matters: For our DMV tech and science crowd, the key shift is practical: Google is framing AI agents as lab operators that can test hypotheses, learn from results and self-improve, not just write or analyze.
Sources
This post summarizes reporting by the outlets above; read them for the full story. Fact-checked against the sources before publishing. Corrections: ashik@ashik.ai
Follow this story. We’ll show an alert on Ashik.FM when it is updated.