How do I diagnose failed AI collaboration, autonomous workflow, or business process executions?
Polyosync supports operations involving AI Employees, AI collaboration, autonomous workflows, and business processes.
These operations may involve multiple steps, dependencies, and external systems.
When an execution fails, the most effective approach is to identify the stage at which the failure occurred.
Step 1: Locate the execution.
Open the relevant module and find the affected execution in its available history.
Step 2: Review the overall status.
Determine whether the execution failed, completed, or remains in progress.
Step 3: Inspect the execution details.
Where available, review recorded steps, events, results, and technical error information.
Step 4: Identify the failing stage.
A failure may occur during:
- Initial input processing.
- AI Employee task execution.
- Collaboration between AI Employees.
- Dependency evaluation.
- External integration access.
- Business rule evaluation.
- Final result processing.
Step 5: Review dependencies.
If one task depends on another, determine whether the prerequisite completed successfully.
For collaborative operations, examine the relevant task or delegation result where available.
Step 6: Check permissions.
Verify that the operation had access to the resources it required.
Step 7: Review external actions.
If an external system was involved, check whether the action was rejected, timed out, or completed despite an incomplete response.
Step 8: Compare with the expected result.
Determine whether the issue was a technical failure or a valid execution that produced an unexpected business outcome.
Should I restart the entire workflow?
Not automatically.
A multi-step workflow may have completed some actions before encountering an error.
Restarting the entire operation without checking previous results could duplicate messages, records, or other external actions.
What should I send to Support?
Include the module, execution reference, approximate time, failed stage, error message, and expected outcome.
If the issue is reproducible, provide a sanitized example of the input and configuration.
A structured execution history is often more useful for diagnosis than a screenshot showing only a generic failure message.