ProtoPilot is a self-evolving multi-agent AI system designed to automate biological laboratory protocols by converting protocol descriptions into executable code and physical experiments. Tested on 294 synthetic and molecular biology tasks, it achieved 90.2% expert preference and 88.24% pass rate on Opentrons automation, demonstrating reliable conversion from biological intent through code generation to validated wet-lab execution with feedback-guided revision.
A proposal outlines five grand challenges for biology, including synthesizing life from chemical precursors in the lab, cryopreserving and recovering whole mice, creating an enzyme to reverse-translate peptides into nucleic acids, engineering Rubisco with improved specificity and turnover, and developing a living cell with a four-base codon system.