Investigate the full record of an agentic interaction to diagnose the root cause of a quality failure. Trace each step the agent took, including tool calls and outputs, to pinpoint where things went wrong.
Before you begin
You must have a completed agentic evaluation with issues identified.
Role required: sn_aia.admin
About this task
If you are already in the flow for identifying issues, skip to step 5.
Procedure
-
Navigate to .
-
Select the automated evaluation you want to review the results of.
The evaluation details page opens, displaying the overall results and performance metrics.
-
If any issues are identified, select Review and optimize to view details.
The optimization guided setup opens for that agentic AI asset.
Each issue is associated with a specific tool or agentic AI asset.
-
Select an issue to review.
-
Select Show traces.
Each execution plan where the issue occurred can be reviewed node-by-node with the inputs and outputs of each node displayed. You can navigate between nodes by selecting them in the center section.
-
Close the traces modal.
What to do next
Select Apply optimization to look at LLM-generated suggestions for optimizations to fix the issues.