Cognitive debt, without the costume of a diagnosis

AI Cognition Disorder is not a billable condition. It is a working name for two failures that keep getting discussed apart: models that confabulate under tasks borrowed from cognitive screening, and people who outsource so much recall and drafting that unaided work starts to feel foreign.

What the model side actually shows

Neuropsychological-style probes — clock drawing, trails, set shifting, delayed recall — expose brittle spots in systems that are excellent at fluent continuation. The interesting result is not ‘the model has dementia.’ The interesting result is that fluency and executive control are different capabilities, and current architectures still rent the second from the first.

• Visuospatial construction remains a poor fit for token predictors.
• Multi-step plans fail quietly, then get narrated as if they succeeded.
• Citations and mechanisms are filled in. Clinically, that pattern resembles confabulation. Mechanically, it is sampling.

What the human side has to stay honest about

Cognitive debt is a management claim, not a brain scan. Heavy assisted drafting can reduce the felt ownership of a paragraph and the stamina for a problem that will not autocomplete. Early evidence and a lot of staff-room anecdote point the same direction. They do not justify panic copy, school bans written as diagnosis, or products that pathologize ordinary tool use.

The useful version of the idea is operational: verify outputs, keep unaided hours on the calendar, and design courses that grade the thinking rather than the polish. The domain is for people building that version.