Choose a baseline
Start with a run that produced the expected result, then select the run you need to investigate.
Regression review for AI agents
Compare AI agent runs and see what changed.
When one run succeeds and the next one fails, side-by-side comparison helps you move from “something changed” to a specific difference you can test.
Make regressions visible.
TraceProbe lets you compare runs across outcome, timing, steps, tool activity, and execution details. Use the successful run as a baseline and inspect where the failed run diverged.
- Compare successful and failed executions side by side.
- Spot changed steps, missing tool calls, retries, and timing shifts.
- Turn differences into targeted fixes and repeatable experiments.
A practical comparison workflow
Use comparison after prompt, model, tool, or orchestration changes.
Locate the first divergence
Compare the sequence and timing to find the earliest meaningful difference rather than chasing the final error.
Validate the fix
Run the experiment again and compare the new execution to confirm that the change improved behavior.
Useful for prompt and tool regressions.
Comparison is especially helpful when a prompt revision, model change, tool response, or orchestration update changes the behavior of an otherwise familiar agent workflow.
Find the difference between two runs.
Install TraceProbe and explore the built-in comparison workflow.