Google Research says it is developing AI agents that can plan, run and refine wet-lab experiments with human scientific oversight. In a research post published October 16, 2026, the company presented three projects aimed at connecting computational design with physical laboratory measurements.

The first project, called the Verified Infrastructure for Discovery Laboratories (VERID), is a semantic digital twin of a laboratory workcell. Google says it tracks equipment geometry, well volumes, reagents and pipette-tip inventory, while allowing agents to simulate protocols, block physically invalid commands and record synchronized video and execution logs. In reported case studies, the system detected mixing problems, recovered from a plate-reader failure and identified a tip-wall carryover effect during dye experiments.

A second project used AI agents to design, test and analyze DNA aptamers over three cycles involving myoglobin and alpha-thrombin. Google reports that confirmed on-target hit rates reached 33.5% for myoglobin and 42.1% for alpha-thrombin after the agents adjusted their design strategies in response to laboratory results, including nonspecific binding and folding problems.

The third workflow focused on de novo protein binders for fibroblast activation protein alpha. Google says excising the local binding region during structure prediction accelerated each design cycle by more than eightfold, while a human-agent process nearly doubled the in-silico hit rate compared with a scientist working alone at equal compute. The company also reports that top designs produced single-digit nanomolar binders in wet-lab testing. Google says its next step is to combine VERID’s traceable automation with closed-loop protein discovery.