Introduction
An agent's real test starts after you publish it. Once people use the agent, real requests show where it succeeds and where it falls short. Some users get the answer they need. Others rate a response poorly, hit a failed tool call, or ask for something the agent wasn't built to handle. Each of these moments leaves evidence behind.
As the maker of a published agent, your job is to turn that evidence into improvements and to show what the agent delivers. You read production signals to spot where the agent needs attention. You trace those signals to specific sessions and causes, make targeted changes, and confirm that the changes work. You also explain the agent's value to the people who fund and depend on it.
In this module, you explore:
- How to identify where a published agent needs attention on the Monitor tab.
- How to diagnose the cause of an agent issue from session evidence.
- How to improve an agent from recurring patterns in its sessions.
- How to monitor agents with environment-level telemetry in Application Insights.
- How to build a value story that connects an agent's signals to business outcomes.
By the end of this module, you'll be able to monitor a published agent, improve it based on evidence, and communicate the value it delivers.
Note
Agent evaluations on the Evaluate tab and environment-level telemetry in Application Insights are in preview. Agent evaluation is a production-ready preview feature.