What it is
An AI tutor is a personal AI mentor that does not hand over a finished answer but leads the student with guiding questions (the Socratic method), building their thinking. For teachers it generates lesson plans, rubrics and progress summaries. The flagship is Khanmigo from Khan Academy.
Where it came from
Bloom's "two sigma problem" (1984): one-on-one tutoring adds two standard deviations to results, but it does not scale with humans. GPT-4 made a personal tutor cheap in 2023 — Khan Academy launched Khanmigo and OpenAI showcased it as a flagship case.
Why it took off
Khanmigo grew from 68k to 700k+ students in a year and from 45 to 380+ US districts (2024-25), and a partnership with Microsoft gave teachers free access in 34+ languages. The AI-in-education market goes from $6.9B (2025) to $41B (2030), a CAGR of about 43%. The pain is eternal: individual attention is expensive, and AI scales it.
How to use it today
- For your own learning, use AI in tutor mode: "do not give me the answer, lead me with questions" (the prompt sets the style).
- For an edtech product: a Socratic mode plus progress checks beats an "AI answer machine".
- A teacher's chain: generate the lesson plan, differentiate it by level, auto-summarize the results.
- Measure engagement, not just access.
What to watch out for
The main problem is engagement: only about 15% of students with access to Khanmigo use it regularly. Access is not use. There is also the risk of "AI solving it for the student" — you need a design that forces thinking rather than copying.