AI‑generated drafts feel like a shortcut, but the shortcut often bypasses the mental rehearsal that steadies judgment. The first pass of an AI model supplies a polished outline, so the human reader treats it as a finished argument rather than a raw sketch to interrogate. That mental shortcut reduces the “friction” that normally forces a reviewer to ask why each claim matters, to spot gaps, and to re‑frame the story for the audience. Without that friction, the brain’s internal error‑checking loop quiets, and the final product inherits the model’s blind spots.

At a consulting firm, a senior analyst stopped manually summarizing client data after a language model began producing a one‑page executive snapshot each morning. She trusted the model’s phrasing, handed the snapshot to the partners, and noticed that the next strategy session spent most of its time debating the same three points that had already been presented. The missing step was the analyst’s habitual “what‑if” rehearsal, a mental probe that usually surfaces hidden assumptions. By outsourcing that rehearsal, the firm lost a critical filter, and the conversation stalled on surface‑level insights.

The collapse isn’t a flaw in the model; it’s a systemic loss of the cognitive buffer that protects against over‑reliance on convenience. When the buffer erodes, speed gains turn into decision inertia, because teams no longer generate the tension that sparks deeper inquiry. The remedy is to re‑introduce deliberate friction into the workflow, turning AI output into a draft that must be actively reshaped, not a finished product.