# UndoBench: Separating Task Competence from Recovery Capability in Tool-Using AI Agents > Source: > Published: 2026-10-06 10:30:22+00:00 Tool-using AI agents are increasingly deployed across enterprise software systems, yet widely used benchmarks primarily evaluate nominal task completion, conflating baseline planning competence with operational fault recovery. We introduce UndoBench, a benchmark spanning 36 base workflows and 36 fau