agent-retrospective

To analyze repeated AI agent failures and fix noisy routing, you examine operational evidence to identify root causes and propose minimal updates to instructions, skills, or validation gates.

The best way to identify the root cause of recurring agent performance issues is to differentiate between one-off mistakes and durable routing or instruction problems using evidence-backed failure analysis.

You propose changes to agent instructions and skills by analyzing failure patterns from actual evidence and suggesting the smallest possible instruction, skill, or validation update to resolve the issue.

Yes, failure analysis can be used to update agent runbooks and validation gates by proposing evidence-backed changes that address durable routing problems and providing a validation plan for the updates.

Yes, agent performance optimization requires a validation plan for proposed instruction changes to ensure the minimal updates effectively resolve the failure patterns and to assess any residual risk.