AdaMAST Automates Agent Failure Classification

UT Austin researchers introduced AdaMAST, a method that learns to automatically classify AI agent failure modes from trajectories, boosting SWE-bench performance to 70.7%.

2026-07-31 ~ 2026-08-01 · 2 related posts

• AdaMAST: Boosting AI Agent Performance with Failure Taxonomies — abeirami · 2026-07-31
• AdaMAST: Automating Agent Failure Taxonomies Boosts SWE-bench to 70.7% — berkeley_ai · 2026-08-01