Grilling session

matt-pocock's inversion of AI planning: instead of asking the model for a plan (which it eagerly "poofs" out), a tiny skill makes it interview you relentlessly — one question at a time, each with a recommended answer — until human and AI share what Fred Brooks (The Design of Design) calls a design concept. Sessions run 22-100 questions; the transcript, not a document, is the asset. "I didn't need a plan, I needed to be on the same wavelength as the AI."

Downstream effect: the PRD written afterward doesn't need reading — alignment already happened, so reviewing it only tests the model's summarization. Negative decisions land in the PRD's out-of-scope section. Also works on meeting transcripts: feed a domain-expert conversation in and grill through the assumptions.

Boundaries: this is a human-in-the-loop task by definition (human-in-loop-vs-afk) — "you can't Ralph-loop a grilling session." For big decisions, add more humans (pair/mob with the AI as relentless third questioner). Pocock's meta-advice on tools like Spec Kit/Taskmaster: own your planning stack so you can fix it when it breaks.

The same instinct as dhh's "under-specify and iterate" but front-loaded: alignment first, then leave the loop.

Source: report