§ Free download
Production AI Agent Checklist
38 yes-or-no checks to run before an AI agent touches real users, real money or a real codebase, plus three Claude Code subagent files you can copy.
- ✓ Scope: what the agent does, and what it must never do
- ✓ Tools and permissions: read and write access, approvals for destructive actions
- ✓ Untrusted input: prompt injection from pages, tickets and files
- ✓ Cost and runaway loops: turn caps, spend limits, alerts
- ✓ Context and memory, testing and evals, logging and review
- ✓ Failure and rollback: a one-step off switch and safe retries
By Amaresh Adak, who tests AI coding tools and writes up what actually happens.