learning-opportunities

Use short retrieval-based exercises after the AI completes work: predict behavior before seeing answers, sketch your own approach before comparing implementations, and explain components back in your own words. Active recall builds stronger understanding than rereading code.

Offer exercises after creating new files or modules, database schema changes, architectural decisions, refactors, or implementing unfamiliar patterns. Avoid offering them during urgent debugging, quick fixes, or when the user signals time pressure.

The fluency illusion occurs when clean, readable AI-generated code makes developers overestimate their understanding of it. The polished output masks gaps in mental models, which is why active retrieval exercises are needed to verify real comprehension.

Limit exercises to two per session and stop offering if the user declines once. This respects the spacing effect from learning science, where distributed practice beats cramming, and keeps exercises from feeling like tests.

Dynamic testing research shows that errors followed by clear corrective feedback enhance retention more than error-free learning. Being direct about what is incorrect, then exploring the gap, produces higher-value learning than ambiguous softening.