The "cognitive debt" framing is slightly mislocated. The debt isn't from using AI — it's from confusing editorial fluency with generative fluency.

When you write, you discover what you think in the act of arranging words. That's why writing feels hard — it's thinking, not the output of thinking. When you prompt an AI and refine its output, you're doing editorial work, which builds different muscles. You get better at recognizing good prose without getting better at producing it.

That's the real debt: the growing gap between your ability to evaluate writing and your ability to generate it. Same thing happens when programmers who only use AI-assisted code generation start losing the ability to reason about systems from scratch.