AI tutoring with Khanmigo in a two-year school experiment
bryan0
60 points
50 comments
October 06, 2026
Related Discussions
Found 5 related stories in 89.2ms across 8,581 title embeddings via pgvector HNSW
- Schools are experimenting with AI with little evidence or policy to guide them billybuckwheat · 18 pts · September 28, 2026 · 60% similar
- Building a real-time AI tutor for 5-year-olds catalinvoss · 70 pts · July 09, 2026 · 57% similar
- MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training pbui · 126 pts · August 27, 2026 · 56% similar
- CS240 AI Cheating Retrospective ArchAndStarch · 103 pts · September 30, 2026 · 55% similar
- No AI until high school:NYC schools announce major classroom technology overhaul thunderbong · 20 pts · September 02, 2026 · 55% similar
Discussion Highlights (14 comments)
firejake308
Conclusion: "The binding constraint appears to be engagement: realizing the promise of AI tutoring will require getting students to use it, not just giving them access."
nerevarthelame
My casual interpretation is that most kids didn't want to use the AI tutor. The authors estimate that time spent using the AI tutor is as effective as time spent studying without AI assistance , which is around one fourth as effective as studying with a real tutor. The real tutors are obviously quite expensive, comparatively. So, if some kids do prefer studying with this AI tutor, that's as good as studying without it. Whichever approach gets them to put the study time in.
zhivota
I have found this with my kids as well. I have an AI writing tutor that they use but only one of the two actually likes engaging with it. The other gets easily disengaged. Some kids will be able to learn at a record pace with AI, others will need to be brought along in the usual ways via good teaching that fosters engagement.
rambojohnson
blah blah blah.
parsimo2010
A very honest take. They tried something, it showed a very small improvement, and they acknowledged it. Acknowledging that the cause appears to be a lack of engagement agrees with my own experience. You can give students great tools but that doesn’t make them interested in math. Khanmigo might be a good tutor but it won’t help unless you’re chaining kids to a desk and not letting them leave until they finish some quota of problems.
bakul
> The binding constraint appears to be engagement: realizing the promise of AI tutoring will require getting students to use it, not just giving them access. This seems backward. Engaging students should be the primary goal, not AI tutoring. Great teachers engage students. Perhaps AI can somehow aid teachers in becoming better? [Though, I doubt it]
duhhhhh1212
https://www.pangram.com/history/2e681dd5-5fbb-4eb8-8562-973b... This is page 4-11 The abstract is also AI generated. Good god people.
ilaksh
Remedial students, already checked out,barely used it. How about a more assertive and active AI tutor and some consequences for not doing work? That is one thing about a human tutor. They are physically right there versus an AI that is waiting for the student to do something. So for remedial students try a system that has timers for inactivity and real life consequences for ignoring the system. Need to do a study with students that are average or above average.
iamarobot
As a high schooler, the amount of AI forced down our throats is on another level. We have assignments with Claude Prose, tutors that hallucinate, and even gemini in the apps we use to access our assignments. Even my friend's brother(a 4th grader) uses AI to generate pictures for his slideshow about "Penguins". Education is definitely facing a turn, but is it for the worse? Link: https://workspaceupdates.googleblog.com/2026/08/gemini-in-go...
ChaitanyaSai
We saw this when it was released, and it's not because of any amazing prescience. https://blog.comini.in/p/why-khanmigo-will-fail It's just quite obvious when you run a school (we do). It's because AI is focused on making the mechanics easy. But as long as we focus on measurement (how much are they learning) and start from there, the near-term focus is inevitable on the surface mechanics of learning (can they decode or encode when reading, do their tables etc) Meaning > Motivation > Mechanics > Measurement The natural order of learning is when we are allowed to discover, explore, find meaning. This motivates us to focus on getting better, mastery, which is where mechanics should come in. And finally, because it is helpful to know where we are generally, we measure some outcomes. But schooling systems have completely inverted this! More here if anyone's interested: https://blog.comini.in/p/schooling-has-a-meaning-crisis-para...
thelastgallon
Related: Why Sal Khan’t: On Learning by Making but Teaching by Telling https://news.ycombinator.com/item?id=49409862 Sal Khan: The amazing AI super tutor for students and teachers [video] | Hacker News https://news.ycombinator.com/item?id=35791433 New York City Bans A.I. in Elementary and Middle Schools https://www.nytimes.com/2026/09/01/nyregion/ai-ban-schools-n... Tutoring company tells parents to save their money and ‘use AI instead’ https://www.afr.com/policy/health-and-education/tutoring-com...
usernametaken29
> Messages that students did send were mostly bare answers This has been my observation in university as well. LLMs can be valuable in digesting proofs into smaller, workable pieces (e.g researching the origin of a particular linear algebra property). This however requires solid fundamentals and the ability to dissect proofs. In spite of this people with neither usually reach straight to the solution out of sheer desperation (myself included).
tinybear1
Related: there is another learning tool being developed by Khan Academy called “Khanductor” meant to satisfy the need for better assessments. I feel like this a more achievable goal than the lofty one set out by Khanmigo to replace a human tutor and I wish Sal Khan and his team the best in their pursuit of finding viable use of popularized AI learning assistane. It just feels like it’s an uphill battle. There are already great tools for students to use like “LearnLM” models to assess gaps in knowledge. In my opinion this experiment just goes to show that any learning tool, AI or otherwise, will be best utilized best by driven students, ignored by the apathetic, and used by unmotivated as a means to cheat and shortcut their way through school. https://youtube.com/watch?v=rBsiKjVI7tc&pp=0gcJCY8Co7VqN5tD&...
kraig911
Yeah I tried Khanmigo with my daughters and it just didn't stick. Tried making my own that just would piece meal out the answer and strive to gamify it with kudos. I feel weird now to think it but some problems are people just trying to shoe horn AI into something and hope it works out. My kids just didn't want to use their brain once something just spat out the answers. Even if it output all the work they'd just keep pressing for the answer and go next. It really showed up in test scores where it wasn't working when their homework was usually A- in algebra. What has helped me when using AI is it gives me a place much faster to be able to start. I can easily take a pic or draw on a tablet and get it classified in a way so I can start helping her. Now the issue is meeting her in the middle she prefers apps and tech not face to face help. So I made our own thing where it only doles out clues and I get a report back that summarizes the session and how the AI thinks it went. I then take it and give her a reward if she does it week to week. Still though this is a perplexing problem. Feel free to check out what I made (trying to monetize something anything nowadays seems shilling) Cluestep.com My wife thinks it could make money but there's a ton of competition out there. Would love your input.