Worker Runs
Review AI worker execution history.
Worker runs show what happened when an AI worker executed.
Open Worker Runs
Open AI Workers, then select the worker. The worker detail page includes a Runs section under the worker overview.
The worker overview shows:
- worker status
- output mode
- template key, when the worker was created from a template
- context sources included with each run
- callable tool categories
- read-only worker instructions
- recent run history
Read the Runs Section
Each run card shows:
- run status
- created time
- linked task, when the run was tied to a task
- output preview, when the worker produced one
- error message, when the run failed
Select the task link to review the business record the worker acted on.
Use the linked task or record as the source of truth. The run preview is helpful for scanning, but the full record shows whether the worker had enough context and whether the output belongs in the real workflow.
Statuses to Watch
Common worker run states include pending, queued, running, completed, and failed. The worker detail page refreshes while recent runs are still pending, queued, or running.
Use statuses this way:
- Pending/queued: the run has been requested but has not finished.
- Running: the worker is actively executing.
- Completed: the worker finished and produced output or completed its action.
- Failed: the worker could not complete the requested work.
Review Output
For task-comment workers, the run output preview may show the generated
task_comment. Read the output before assuming the work is ready to use.
Check that the output:
- matches the worker instructions
- uses the right task or record context
- stays within the worker's allowed tool categories
- does not claim work that was not completed
- gives a useful next step when the worker could not finish
If the output will influence a customer response, invoice, project update, or support decision, have a person review the source record before acting on it.
Review Worker Configuration While Debugging
When a run looks wrong, review the same detail page before editing:
- Context tells you which record data the worker can see.
- Callable tool categories tells you what the worker is allowed to do.
- Instructions tell you the task the worker is trying to follow.
If the worker needs different behavior, select Edit and update its instructions, context sources, or tool categories.
Failed Runs
If a run fails, check permissions, tool access, connected data, provider availability, and whether the worker instructions asked for something outside its allowed scope.
Troubleshooting order:
- Read the error message on the run card.
- Open the linked task or source record, if present.
- Confirm the record still exists and the worker can access it.
- Check that the worker has the needed context source.
- Check that the allowed tool categories include the action it was asked to perform.
- Edit the worker instructions if they are too broad or ask for unsupported work.
Do not repeatedly rerun a failed worker without changing the cause or reviewing side effects. If the worker writes comments or creates records, repeated runs can create duplicate output.
Run Review Checklist
For each important run, confirm:
- the status is completed or the error is understood
- the linked source record is correct
- the output matches the worker instructions
- the worker used only appropriate context
- no duplicate comment or record was created
- the next human action is clear