We often mistake ease for competence. However, Robert Bjork’s framework of "Desirable Difficulties" suggests the opposite: learning is most durable when the learner has to struggle to retrieve and organize information.

When you write code or draft a strategy from scratch, the struggle is the point. The difficulty forces you to resolve logic, structure arguments, and debug your own thinking.

Cognitive Load Theory, developed by John Sweller, explains the mechanism behind this. Sweller distinguishes between "extraneous load" (distractions, bad formatting) which hurts learning, and "germane load" (the effort of constructing patterns) which drives it.

The manual grind of creating a draft generates high germane load. It makes the brain build the mental models experts rely on.

The problem with GenAI is that when we read a fluent, perfect draft, we remove this friction entirely. We experience a "fluency illusion." Because the answer appears instantly and reads smoothly, our brain assumes we have mastered the content.

But we haven’t.

We have simply skipped the processing that leads to mastery.

How can we reintroduce desired difficulty with an equally specific friction that forces the brain to engage with the logic, even if the hands aren’t doing the typing?