The two checks that would have caught our worst mistakes.
Written down so you can run them on your own agent’s work this afternoon.
An AI agent's report — any AI you hand a task to, from ChatGPT to Claude — that the job is done is not evidence the job is done. We learned that the expensive way — this is the method we built from it, with the receipts.
Already have our free checklist? It tells you what to check. This is what you actually run — the full prompts, filled-in examples included.
What $49 gets you
Why an agent's own account of its work is never proof, how to write a spec a stranger could check, and the two-agent loop that catches what self-review can't. Every rule taught through the real failure that created it.
What agents do when context is missing (fill the gap with plausible fiction), the brief that prevents it, and how to decide when a cheap, fast AI model is fine and when the task deserves the expensive one.
The spec-writer, the verifier brief, and a complete worked pair — a thin brief that caused a real failure next to the full brief that would have prevented it. Fill in the CAPS, paste into any AI tool. These are what you actually run.
The free one-pager, included — the modules are the why behind every line of it.