There’s a familiar argument that AI is turning our brains to mush—essays prompted rather than written and engineers vibing code straight to production. I’m sympathetic. We can produce sophisticated work without knowing how, why, or sometimes even what we’re doing.
The brain-mush risk is real if we produce the work and call it a day. But for those willing to step back and study it, the same AI that avoids us thinking too hard can also offer both a teacher and a curriculum for getting better at it.
When we work with AI, we generate chat logs and drafts that record our decisions, revisions, questions, and errors. This becomes game tape we can learn from. Like a team reviewing a match and designing drills around its mistakes, we can use that record to decide what we need to practice for next time.
That record grows with every project: the product decision we reversed, the technical misunderstanding that caused a bug. Now we have a coach to replay it with—one that gives more instruction than a busy colleague could. A permissions bug that caused a headache can turn into a lesson grounded in our own codebase. If we still don’t understand what went wrong after reviewing the game tape, we can keep asking until we do. No workmate would have the patience or context to search for the answers like an agent can.
In this way, the AI that helped us complete our work can become our most valuable curriculum, if we choose to study it.
