Google Research is building AI agents for autonomous wet labs

Google Research is connecting AI agents to laboratory robots, with projects spanning lab simulations, aptamer discovery and the design of protein binders. The work offers a concrete look at how AI might move from analysing biology on a screen to helping run experiments in a physical lab.

Watch Desk analysis

What happened

In a post published on 16 October, Google Research describes three research efforts. VERID is a semantic digital twin framework intended to help agents operate in physical laboratory workcells. A separate aptamer project uses repeated, closed-loop experiments to improve binding specificity.

A third workflow uses an agent to guide epitope excision for de novo binder design. Google Research says it produced FAP binders with single-digit nanomolar affinity. Google Research’s account of the work describes these as research projects, not a ready-made autonomous laboratory product.

Key findings

• VERID’s digital twin framework
  It is designed to give AI agents a structured representation of laboratory workcells.
• Closed-loop aptamer discovery
  The project uses repeated experimental cycles to improve binding specificity.
• Agent-guided binder design
  Google Research says the workflow produced FAP binders with single-digit nanomolar affinity.

Why it matters

Laboratory robotics can take AI beyond suggesting what researchers might test: agents could help direct physical experiments, while each experimental result informs the next step. That makes the interface between software, instruments and biological evidence part of the research challenge, not just the plumbing behind it.

The reported binder result is a specific research outcome, but it does not by itself show that the method generalises to other targets or that a lab can run without meaningful human involvement. Those are the questions that separate an impressive workflow from a dependable research tool.

Our read

This is a more compelling AI story than another agent being handed a browser and told to get on with it. The work puts agents in contact with real experiments, and reports a measurable result. The next test is whether the approach proves repeatable and useful across more than a carefully chosen demonstration.

What to watch

• Whether Google Research publishes further experimental results for the three projects.
• Whether VERID is tested across more laboratory workcells and instruments.
• How the agent-guided workflow performs on targets beyond FAP.
• What role researchers retain in checking experiments and deciding what happens next.

Discussion spark: How much control should AI agents get over physical lab experiments before a researcher must approve the next step?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.