ProtoPilot paper shows self-evolving agents for wet-lab protocols

An arXiv paper introduces ProtoPilot, a self-evolving multi-agent system for automated biological protocol generation and execution. The system spans protocol design, SOP expansion, SDK-compliant code generation, device-level checks, and feedback-guided revision. The authors evaluated it on 294 synthetic-biology and molecular-biology tasks derived from 98 protocols, reporting a 90.2% Top@3 expert-preference rate, an 89.5% protocol-to-code gate pass rate, and an 88.24% Opentrons pass rate.

Key details: Submitted June 30, 2026 to arXiv, The paper introduces ProtoPilot for autonomous wet-lab automation, The benchmark covers 294 biology tasks from 98 protocols, The system generates protocols, SOPs, SDK-compliant code, and revisions from feedback, The authors report an 88.24% Opentrons pass rate.

Why it matters: Wet-lab automation is where agent mistakes become physical errors, so protocol-to-code validation is a serious test for AI-for-science claims.