AI Agents for Scientific Discovery

Overview

AI agents for scientific discovery are AI systems that autonomously execute multi-step scientific workflows — including literature search, hypothesis generation, experimental design, and data analysis — with minimal human intervention per step. The capability of such agents has roughly doubled every seven months as measured by the METR benchmark, tracking from GPT-2 in 2020 through Claude Opus 4.6 in 2026. Edison Scientific is among the most prominent deployed examples, having served over 2 million queries since launching its agent platform in November 2025.