The AI Reckoning / Appendix D

AI Integration Checklist

Check the problem, operating boundaries, and evidence before and after adding AI.

Download printable PDFOne page · Updated October 20, 2026

Before building

State the business problem, intended users, and why AI is useful. Identify a simpler baseline for comparison.
Define acceptable outcomes, unacceptable errors, and who owns the decision to release.
Identify data sources, permissions, retention, and information that must not reach the model.
Map integrations and dependencies. Agree on tool permissions and human decision points.

Test representative tasks, edge cases, failure conditions, and hostile inputs. Evaluate correctness as well as format.
Review generated code, dependencies, authorization, and data handling. Resolve consequential findings.
Version relevant prompts, models, retrieval, and supported settings. Plan cost limits, monitoring, fallback, and rollback.
Practice recovery and handover. Name who responds when something goes wrong.

Monitor outcomes, errors, cost, latency, and user reports. Review unexpected behavior rather than simply retrying it.
Rerun evaluations before consequential changes. Recheck permissions and data sources as the system evolves.

Decision: proceed / remediate / stop. Evidence, open risks, owner, and review date: ….

Use the printable worksheet to keep your notes and evidence with your team.