UndoBench: Separating Task Competence from Recovery Capability in Tool-Using AI Agents
UndoBench shows tool agents hit 83.54% task success but only 46.72% conditional recovery, with retries duplicating effects.
2 stories, newest first · one card per story
UndoBench shows tool agents hit 83.54% task success but only 46.72% conditional recovery, with retries duplicating effects.
Re-derivability, not extra stages, decides how staged LLM agents recover after an upstream fault.
Today 245 readers · 261 views · 66 reads · all time 199K views