@itsolelehmann - Claude Premortem Prompt Stress-Tests

Open original

@itsolelehmann on X The post describes a "premortem" prompt for Claude: instead of asking whether a plan is good, you tell Claude it's six months later and the plan has already failed, then ask how. That reframing gets past the model's tendency to affirm your idea and surfaces failure modes, early warning signs, and hidden assumptions.

  • Daniel Kahneman called the premortem his most valuable decision-making technique, and Google, Goldman Sachs, and Procter & Gamble reportedly use it before major launches.
  • Asking Claude "is this a good plan?" tends to produce reasons to say yes, which breeds false confidence before you commit weeks or months to it.
  • Framing the plan as already dead switches off that optimism, and the output includes the most likely and most dangerous failures, the biggest hidden assumption, and a revised plan.

A premortem flips the usual way you stress-test a plan. Instead of asking Claude "is this good?", which tends to get you a list of reasons it will work, you tell it the project is already six months dead and ask how it died. The author's point is that the question "what could go wrong?" still lets the model's default optimism creep in, while a dead-on-arrival premise removes that opening. The technique traces back to Daniel Kahneman, the Nobel-winning psychologist behind "Thinking, Fast and Slow," who called it one of his most valuable decision-making tools. The post also name-checks Google, Goldman Sachs, and Procter & Gamble as organizations that run premortems before major launches. The reasoning behind it is that AI models are shaped to be agreeable, so a plan you describe gets validated, you feel confident, and you spend weeks or months building on something with a flaw that was obvious in hindsight. The author is selling a ready-made skill: you paste in your plan and say "premortem this," and the model returns a set of failure scenarios, each with a full story and early warning signs. It then synthesizes them into four things: the most likely failure, the most dangerous one, the biggest hidden assumption behind the plan, and a revised plan with the gaps patched. The author treats that hidden-assumption step as the most valuable part. The post is short and light on evidence. It offers no data on how often premortems improve outcomes, and the Kahneman, Google, Goldman, and P&G references are name-drops rather than documented case studies. The underlying idea is sound and cheap to try yourself, since the prompt needs no special skill, just the reframing of the premise. Whether a skill package adds much beyond that prompt is the part a reader would need to judge for themselves.

Ask about this article