failure-postmortem

To write a post-mortem for an AI agent failure, use a guided workflow to capture the incident, classify it into one of six failure patterns, and generate a publication-ready report. This structured approach turns incorrect AI outputs into actionable documentation.

During AI incident analysis, you should look for six common failure patterns: Context Degradation, Specification Drift, Sycophantic Confirmation, Tool Selection Error, Cascade Failure, and Silent Failure. Classifying incidents into these patterns helps identify systemic gaps.

To perform a 5 Whys root cause analysis on an incorrect AI output, follow a guided workflow that iteratively asks why the failure occurred. This process identifies systemic gaps and missing guardrails, which are then documented in a structured post-mortem report.

Yes, you can save AI failure post-mortem reports to a knowledge vault. After the guided workflow generates a publication-ready report using a standardized template, it provides an optional phase for verification and direct vault storage.

The best way to document AI incidents for accountability is generating a standardized post-mortem report. By applying pattern classification and root-cause analysis to the AI failure, teams produce a publishable document that supports learning and accountability.

No specific external tools are required to start diagnosing AI system failures. The post-mortem builder operates independently through a guided workflow, requiring only the details of the AI incident to classify failure patterns and generate a report.