AI in education gets discussed in two extremes, a tool that will personalize learning for every student, or a shortcut that will hollow out genuine learning. Khan Academy's Khanmigo, now used by millions of students across more than 100 countries, gives us real, publicly reported usage data to check both claims against, and the picture is more specific than either extreme.
The growth numbers are genuinely dramatic
In the 2024–25 academic year, Khanmigo usage grew from 40,000 to 700,000 K-12 students, with projections to surpass one million in 2025–26. By late 2025, more than five million students across 110 countries had used the tool, reaching classrooms not just in the US but in India, Brazil, and the Philippines. That's a scale of adoption that puts Khanmigo well ahead of most AI education tools in terms of measured, sustained use rather than pilot-stage experimentation.
How it's actually designed to work
Khanmigo is built around guided questioning rather than supplying direct answers, a deliberate design choice meant to encourage critical thinking and persistence rather than simply providing solutions. For teachers, it generates lesson plans, rubrics, grouping strategies, and progress summaries, reducing administrative burden in a way that parallels the "AI drafts, human decides" pattern we've found scaling successfully across other regulated, human-judgment-dependent fields, including the ambient documentation tools covered in our healthcare AI piece.
A real result worth taking seriously
One documented case is genuinely striking: at Enid High School, educators began using Khanmigo for geometry, and after a single semester, there were no students failing that class, a result the school found compelling enough to expand Khanmigo across all of its math classes. That's a real, specific, publicly reported outcome, not a marketing claim without a source behind it.
The engagement gap that complicates the growth story
Here's the finding that keeps the overall picture honest: despite the scale of access, only 15% of students who have Khanmigo available actually use it, a gap significant enough that Khan Academy is actively working on a redesign in response. Reported friction points include technical challenges typing equations or interpreting math prompts, and some students found the chatbot's conversational style too "chatty" for what they needed in the moment.
What the gap between access and use actually tells us
This is a genuinely important distinction that access numbers alone obscure: "700,000 students have access to Khanmigo" and "700,000 students are meaningfully learning from Khanmigo" are different claims, and the 15% usage figure is the more informative one for understanding real impact. It's a pattern worth watching across AI tool adoption broadly. Access and provisioning scale much faster than genuine, sustained engagement, and reporting only the access number without the usage number tells an incomplete story.
Why full personalized-curriculum tools still face this same structural challenge
Tools promising to fully personalize a student's entire learning path face a harder adoption path than a supplementary tool like Khanmigo, for reasons that are more organizational than technical: a fully individualized curriculum is a significant departure from how schools are structured to operate, including standardized testing requirements and existing teacher workflows built around whole-class instruction.
The honest state of play
Khanmigo's real numbers support a more specific and more useful claim than either "AI is transforming education" or "AI in classrooms is a failure": genuine, documented positive outcomes exist at the classroom level, adoption at scale is real and growing fast, and the gap between granting access and achieving sustained engagement remains a serious, actively unsolved problem even for the best-resourced tool in the category.