Google Research is pioneering a new approach to bridge the gap between AI predictions and real-world biological experiments. In an October 16, 2026 blog post, the team introduced the concept of autonomous wet labs, where AI agents design, execute, and analyze experiments using robotic systems. The goal is to create AI that learns from actual lab conditions rather than relying solely on simulations.
The project centers around VERID, a framework described as a semantic digital twin of the lab. VERID enables agents to inspect the lab's state, run simulations, and execute protocols safely on physical hardware. This allows for closed-loop cycles where agents design candidates, test them, analyze results, and refine subsequent batches.
One notable result involved aptamers, short DNA strands designed to bind target molecules. Initially, the agents achieved a 0% success rate, but by the third iteration, they hit a 42.1% success rate. Each round tested around 1,000 candidates at a cost of $1 to $3 per candidate. The team also successfully designed proteins with strong binding affinities for cancer-related proteins, demonstrating the potential of AI-driven biological research.
The research addresses a longstanding issue in computational biology: predictions made on computers often fail in real lab settings due to variables like measurement errors and contamination. The work aligns with Google's broader AI-for-science initiatives, including the Virtual Biology Initiative, which received nearly $1.8 billion in funding on October 7, 2026. This initiative aims to expand biological datasets and build virtual cell models.