MGI and Shanghai AI Lab bring agents into wet labs

AI for science becomes much more interesting when a model can do more than suggest an experiment. The harder step is turning reasoning into repeatable physical execution.

What happened

MGI Tech subsidiary Genoria AI and the Shanghai Artificial Intelligence Laboratory introduced ProtoPilot and BioLab Bench, two connected systems designed to bridge AI reasoning with laboratory work.

ProtoPilot is a multi-agent framework for interpreting and operationalising biological procedures. BioLab Bench evaluates whether agents can move from textual instructions through to actions on laboratory devices.

Together, the systems target the gap between planning an experiment and actually carrying it out in a physical lab environment.

Why it matters

Many AI-for-science tools still sit at the analysis layer. They can read papers, generate hypotheses or suggest protocols, but wet-lab work remains constrained by physical devices, procedural detail and reproducibility.

Connecting agents to laboratory systems could automate more of the experimental loop, but only if execution is reliable enough to handle real instruments and biological workflows.

That makes evaluation infrastructure especially important: an agent that sounds scientifically plausible is not necessarily one that can execute a protocol safely and consistently.

The bigger picture

Autonomous science is moving from language models toward physical AI.

The long-term opportunity is a closed loop where systems can reason about an experiment, operate equipment, observe results and refine the next step. If that becomes practical, AI could change not only how researchers analyse science but how experiments themselves are run.