← All posts

AI Tutors Are Hot, but Teaching Is the Wrong Job for Them

Jun 17, 2026#Education AI417 words · 2 min read阅读中文原文 ↗

The short version

Overrated: personalized instruction replacing teachers. Underrated: getting teachers out from under repetitive work.

TestedEducation AI

The short version

In education, AI’s most overrated ability is personalized instruction. Its most underrated one is taking over the pile of repetitive work teachers carry. Get the direction backwards and both the budget and the expectations come to nothing.

Overrated: the personalized-learning illusion

“A custom learning path for every child” sounds unbeatable. It also assumes something false: that kids do badly because the content isn’t tailored enough.

Usually that’s not the problem. The bottleneck is motivation, the feedback loop, and whether anyone is watching to keep you going. AI can generate an unlimited supply of problems built just for you. You still won’t want to do a single one.

What you actually pay for in one-on-one tutoring is that a person is in the room. They know you slept badly last night, and they can pull you back when you’re ready to quit. AI can’t do that. Not yet, anyway.

A risk you can’t ignore: teaching it wrong is worse than not teaching it

Models get things wrong with a completely straight face. An adult using AI will usually catch it. A kid still building the basics has no way to judge right from wrong.

A wrong concept delivered confidently hardens at that age, and undoing it later costs several times the effort. In education, hallucination isn’t a “user experience issue”. It’s an incident.

Underrated: getting teachers out of repetitive work

What AI can ship in education right now, with little chance of blowing up, is the work teachers don’t want to do:

  • Writing questions and assembling papers — generate in bulk from a given topic and difficulty level, then filter by hand
  • Grading objective questions, first-pass essay feedback — the model writes a version, the teacher edits on top of it, far faster than starting from scratch
  • Backstop Q&A — a student is stuck at 10pm, AI takes the first round, and anything it answers badly gets flagged for a teacher
  • Prep material cleanup — pull scattered source material into a first draft of a handout

These have one thing in common: AI writes the draft, the teacher makes the call and signs off. The time that frees up goes to the thing AI can’t replace, which is paying attention to actual people.

The one-liner

Don’t expect AI to be the teacher. Let it be the teacher’s assistant. Put it where the work is repetitive, checkable, and backstopped by a human when it’s wrong. That’s where the value holds up and the risk stays small.

Was this useful?

If this post helped, you can buy me a coffee ☕