How I use premortems with Claude and Codex

The premise is simple and honest: default LLM review prompts are too optimistic, so the author replaced them with premortem-style prompts that force the model to imagine a specific failure before it happens. The technique is not new — premortems come from Gary Klein's 1990s research on decision-making — but applying them to AI-assisted code review is a practical adaptation most teams have not tried. The article walks through the actual prompts used and the kinds of failures the approach surfaced that a generic review pass missed. It is a short read, maybe twelve minutes, and the takeaway is directly portable to any Claude or Codex workflow without a new tool or dependency. No product to install, no pricing tier to worry about. The reservation is that this is one person's workflow piece, not a controlled study, so treat it as a starting template rather than a proven methodology. -> Best for: solo founder or technical PM who wants more adversarial signal from AI code review