I teach physics to real college students, and I build learning tools. These essays are written where those two rooms overlap: what AI actually changes about learning, what it merely exposes, and what we should build about it. New essays land here two or three times a week.
The semester the AI tools arrived, my problem sets stopped measuring anything. AI broke the proxies, not the learning. What redesign means when banning and detecting have both failed, why adaptation should stop being a luxury good, and the four pillars we're building about it.
Coding-as-literacy made sense when syntax was the gate. The gate moved. What kids actually need now: framing a problem, splitting it into pieces, judging what comes back, and the taste to know when it's wrong. Why "the AI wrote the code" stops being cheating the moment the goal changes.
The nursing student terrified of physics is the rule, not the exception, and the fear itself lowers exam scores in measurable ways. What the evidence says actually helps: retrieval practice without stakes, rebuilding from the real gap, and a course that expects you to be human.
The detectors don't work, and their false accusations land on the wrong students. The way out is older than the problem: assess the process, stage the checkpoints, say exactly where AI is allowed, and make the work worth doing with the tools turned on. From someone rebuilding her own assessments.
Time, habits, money, asking for help, starting the hard thing first: the syllabus behind the syllabus. Some kids get it taught at home and look "naturally organized." Everyone else pays for the gap in college. Why these skills are teachable, and why nobody's job is to teach them.