'Who Broke the System?' AgentLocate Pinpoints the Failing Agent and Step in Multi-Agent Runs

arXiv·medium signal

AgentLocate combines an LLM judge with multi-perspective verification from independent evaluators, aggregating their assessments via a confidence-aware strategy and adaptively fine-tuning on that feedback to improve attribution. The authors report it outperforms existing failure-localization methods at identifying both the responsible agent and the failure step while staying efficient in tokens and runtime. Directly addresses the debugging problem that grows with every agent you add to a team.