A May article in The New York Times described the growing pains of Khanmigo, Khan Academy’s AI tutoring tool developed in collaboration with OpenAI. The ambition is enormous: build something that could approximate the tone, guidance, and nuance of a great human tutor—maybe even something like putting Sal Khan in every student’s pocket.
That is a beautiful vision to some. It is also a very hard one to make real.
What AI can and can’t do for math teachers and tutors
The challenge with Khanmigo seems to be in finding the “Goldilocks zone”: enough context to make the AI genuinely useful, but not so much that the tool becomes prohibitively expensive or unwieldy. Meanwhile, there is another use of AI in math education that is less ambitious. And it is ready right now.
What if AI’s most immediate role were not to potentially replace teachers and tutors but to help them do the parts of teaching that are most valuable and least scalable? As a former physics teacher, now a parent, and—by extension—a reluctant math tutor, I ask this as someone who is as skeptical of AI in the classroom as anyone.
The reality of math instruction today
My son remembers most of what he reads but very little of what he hears. That matters more in math than we may realize or admit.
The classic math classroom tends to include a lot of spoken explanation, accompanied by numbers and symbols on a board. Often, the words that explain the thinking are never written down. That may be workable for students when problems are short and procedural, but as the math becomes more complex, the words that connect one step to the next become essential.
By Algebra 2, my son did not just need to see the steps; he needed written explanations of why those steps made sense. His teacher discouraged students from relying on the textbook, however, because, as she put it, “The textbook doesn’t always do things the way I think they should be done.” I understand that instinct. Many teachers have preferred methods. But the result was that the one source of durable, written explanation was pushed aside.
Add to that the reality of teaching today. Like many high school math teachers, my son’s teacher was responsible for a staggering number of students. Personalized written feedback on every mistake, every partially correct solution, every almost-right line of algebra is simply not realistic under those conditions.
So the start of the school year, I began spending evenings and weekends working through problems with my son. I helped him reconstruct what he had missed in class. I translated procedures into written explanations. I found similar problems. I checked his work. I explained the same idea in three different ways.
The hours mounted. The Herculean effort took a toll on both of us.
In January, we tried formal tutoring. The tutor was kind and capable, but the format had the same fundamental problem as his classroom: the feedback was mostly spoken. For my son, those words evaporated almost as soon as he heard them.
We turned a corner when we started using math-focused AI.
Tips for using math-focused AI
The first—and most important—thing my husband and I did when we made the switch to using math-focused AI as a tutoring support tool was ensure he and I own the account. My son does not have direct access. This is not a tool we handed to a 15-year-old so he could outsource his homework. It is a tool that helps us help him. And in that role, it has been remarkable.
I can upload a photo of a page of handwritten work and ask the AI to review it. It can identify which steps are correct, where the reasoning goes wrong, and how to repair the solution. It can write out the explanation my son needs, in words, next to the algebra. For him, that is the holy grail: detailed written feedback on his own work.
I can also ask the AI tool to generate new problems like the ones he missed. Not vaguely similar problems. Not whatever I can find after twenty minutes of searching online. Problems at the same level, using the same concept, with worked solutions included. That means my son gets more practice exactly where he needs it.
Embracing the middle ground
This experience has changed how I think about AI in math education. Not because I now believe every student should have an AI tutor in their pocket. I do not. If I handed this tool directly to my son, would he always have the self-control to use it for learning rather than shortcutting? I wish I could say yes, but he is 15. The adult-in-the-loop model matters.
AI does not have to replace teachers—nor should it. It does not have to replace tutors, either. AI does not—and should not—have to become the classroom authority. Its most powerful near-term use may be much more practical: helping adults give students the kind of detailed, personalized written feedback we already know is valuable but rarely have time to provide.
Math teachers can use AI to draft explanations for common errors. Math tutors can use it to create targeted practice sets. Parents can use it to better understand what their children are struggling with. And kids can rely on it to receive more written feedback, more quickly. None of us have to pretend AI has somehow solved the deeper human work of teaching.
The revolution does not need to be an AI tutor for every student. It can be an AI assistant for every teacher, tutor, and parent trying to help a student learn instead.
