Google Research has unveiled a new framework for deploying AI agents in physical laboratory environments, marking a major step toward fully autonomous scientific discovery. The research addresses a critical gap in AI development: while frontier AI systems excel at in-silico design and hypothesis generation, they struggle to reason about three-dimensional geometry and physical constraints needed for real-world lab automation. The company's work focuses on bridging the gap between digital AI exploration and wet-lab experimentation, where agents can run actual experiments, learn from results, and refine their approaches in closed-loop cycles. The breakthrough centers on three interconnected research projects. The VERID framework operates as a digital twin of a laboratory workcell, tracking robotic geometry, reagent volumes, and pipette inventory in real-time. This enables AI agents to simulate liquid-handling protocols before execution, catching potential collisions or errors before they waste reagents or compromise experiments. In parallel, Google demonstrated closed-loop molecular discovery using protein-binding DNA aptamers, where agents designed candidate sequences, ran actual lab experiments, and iteratively refined their designs based on real experimental readouts over multiple design-test-analysis cycles. The implications extend beyond academic research. Google's agents have successfully discovered single-digit nanomolar protein binders for cancer targets and demonstrated the ability to identify and correct for optical and fluidic artifacts in experimental runs. By grounding AI systems in physical reality and enabling rapid iteration with actual lab feedback, the research opens pathways for AI-driven drug discovery at scale. The work also builds on the open-source PyLabRobot framework, signaling Google's intent to make autonomous lab infrastructure available to the broader research community.