AWS Strands Evals detects AI agent failures

AWS has introduced Strands Evals, a new tool designed to identify and diagnose failures in AI agents. This system provides structured output, categorizing failures with associated confidence scores and detailing causal chains that link root causes to observable symptoms. Furthermore, Strands Evals offers specific recommendations for fixes, indicating whether adjustments are needed in the system prompt or tool definitions. The tool can be integrated into evaluation pipelines for automated diagnosis during every test run, enhancing the reliability and robustness of AI agent development.