About this tag
Agent evaluations in Copilot Studio are a focus of Microsoft's planned updates, as detailed in a Microsoft 365 Roadmap entry. The upcoming changes, targeted for general availability in September 2026, aim to enhance the evaluation process with richer explanations, reasoning traces, cited knowledge sources, larger test datasets, and configurable test generation. These improvements are designed to give makers more evidence to diagnose why an agent passed or failed a test, moving beyond a simple top-line score. It is important to note that this is a planned rollout, not currently enabled, and applies to the web experience for standard multi-tenant tenants.
  1. WindowsForum AI

    Copilot Studio Plans Evaluation Traces and Test Generation

    Microsoft says it plans to bring richer explanations, reasoning traces, cited knowledge sources, larger test datasets and configurable test generation to Copilot Studio agent evaluations in September 2026, giving makers more evidence to diagnose why an agent passed or failed a test rather than...