Why AI Tutoring Works Best When It's Honest About Its Limits

There is a comfortable story going around about AI in education: that it can replace tutors, accelerate learning, and personalise everything. The reality is more nuanced — and far more interesting.

After working with students from late primary through early university, I have come to believe that AI tutoring is genuinely useful, but only when paired with something it cannot provide on its own: a human who notices what the student is actually struggling with.

What AI does well

Generating practice problems on demand. Re-explaining a concept three different ways in three minutes. Spotting arithmetic slips before the student does. These are real, measurable gains, and any honest educator should welcome them.

What AI quietly misses

AI does not see when a student is performing understanding rather than holding it. It does not catch the small hesitations that signal a missing prerequisite from two years ago. It does not know when to push and when to pause. Those judgements still belong to the teacher in the room.

The best lesson I taught last term was the one I almost cancelled — because the student arrived tired and I switched plans on the spot. No model would have made that call.

How I use AI in my own tutoring

• As a problem generator — fast, varied, and instantly adjustable in difficulty.

• As a second explainer — for when my analogy did not land and the student needs another angle.

• As a check on my own marking — useful for catching things I miss when I am tired.

• Never as the lesson itself.

The point

AI is a tool that rewards teachers who already think carefully about pedagogy. It punishes those who use it as a substitute for that thinking. The students who do best are the ones whose teachers stay in the loop — not the ones whose teachers step out of it.