Run logs help explain what happened in a specific conversation instead of guessing from the final reply. Separate knowledge, instruction, action and channel issues.
Before you start
You need access to the agent and its run logs. Start with a known case and approximate time, and handle customer data under your team’s permissions.
Steps
- 1
Open run logs
From AI agents, open the agent and its Runs or run-log view.
- 2
Find the case
Locate the conversation or time and review messages, outcome and recorded actions.
- 3
Classify the cause
Was knowledge missing, an instruction unclear, an action or integration unsuccessful, or did the event never reach the agent?
- 4
Change only what is needed
Update the relevant source, instruction or action setting rather than changing everything at once.
- 5
Retest
Repeat the case in the test panel and watch a fresh run after publishing.
Check the result
You can identify the cause and the correction, and a fresh test behaves as intended without breaking known-good cases.