As anyone who’s met Sal will attest, this characterization is incredibly off the mark.
The author extrapolated from the product as opposed to understanding the person and how the product came to be the way it is.
(I use the flipped classroom approach extensively. But the legal profession being the eager embracers of novelty that we are </sarcasm>, I refer to the approach as "enhanced Socratic method," given that Socratic method was something we all endured in law school.)
For math and engineering courses, the basic flipped-classroom approach was used at West Point for over 150 years: In the so-called Thayer method, students study textbooks in advance, then in class, they work on problems on chalkboards. The instructor moves from chalkboard to chalkboard, coaching the students in real time. [1]
(I do much the same thing in my contract-drafting classes: In small groups, students do drafting exercises using online whiteboards. I lurk on my laptop, quietly coaching in person as needed.)
[0] https://bokcenter.harvard.edu/flipped-classrooms
[1] https://flippedlearning.org/syndicated/avoiding-flipped-lear...
This article appears to focus on the chat bot which I have not used and was not part of my experience. At the time, I had a subscription to Wolfram Alpha that was helpful if I got stuck on an intermediary step.
The high level of goal of the post which is directed at asking good questions raises a point that seems valid to me.
Here's a presentation that, kind of claims the opposite. More specifically, it claims figuring it out on your own is not as good as having a teacher tell you, and then having you use the info, and it claims to be backed up by research
For generic trope-y content like this, the voice can't be left this AI-ish. It has to be about something really interesting for this AI tone to be bearable.
I think there is a simple reason why AI tutors are not terribly successful.
AI is just not good enough yet, they are only just now capable of performing tasks, getting them to do those tasks involves a degree of careful guidance and the ability to handle frustration calmly and rationally.
Those things require a high motivation to have the task done.
They are not the properties you want in a tutor, a few students have the drive to learn in conditions that are not ideal. They are a small minority.
What you need is a AI that has a deep enough understanding of the ideas behind the metals it teaches and is engaging witty and charismatic enough to get that information across to a student.
When it has those abilities there will be a lot of parents that won't want it to have access to their kids.
I remember minimal videos (3-13 min, usually 5-8), and then exercises drilling the techniques until you understand it and don't make mistakes. If you got a perfect score (no mistakes whatsoever), you went a little faster.
That's learning by doing, or am I wrong? Or are there no exercises for things other than math?
[0] https://direct.mit.edu/books/book/5138/Teaching-MachinesThe-...
In the classrooms I see through my kids, the teacher is doing roughly what a Khan video does: walking through a worked procedure at the front of the room. If a video fails to provoke the "wait, why does that work?" moment, then a teacher following the same flow likely fails in exactly the same way as the author claims and missing the opportunity for hands'on learning. A video actually allows the learner to go back and review content if anything was unclear, and presumably any interactive AI agent can help with the feedback part even though the teacher in the video isn't.
I also wish the piece had stayed with Khanmigo, where nobody has a good theory yet on where the ai tutoring loop actually breaks and why people fail to engage it and that putting it in their face may not work, why exactly personalization at that scale hasn't produced the gains it promised.
Also irked by the title built on a pun on Khan's name, sets up an awkward framing that kind of lets me know it may not be unbiased.
> Now… how does Sal Khan want my kid to learn?
> Watch a video.
I find this a little uncharitable. I learned a lot of math at a young age from Sal Khan's early videos (back when he used to erase by switching to a black pen and scribbling across the entire screen) and, to put it in the author's terms, it served as an easy-to-digest scaffolding on top of which to build a deeper understanding. I don't think Khanacademy ever sold itself as a magic learning resource that removes the need to work out problems yourself.
I also don’t think he ever purports to be “all you need” to learn.
It reminds me of the people who criticize Duolingo because you can’t actually reach perfect fluency through their app--like wasn’t it always obvious that it was never meant to be the only education you need; wasn’t it always obvious that you need to actually _speak_ a language to learn it? And yet it’s still a great app that has helped millions of people learn languages.
In the more core experience they were super focused on not giving kids answers. I suppose built on a fear of subverting the teachers or classroom environment. But again it became a fixation that took all attention away from teaching itself.
But I figure you put something out in the world and see what happens. Yet I came back to it a year later and it was exactly the same. Nothing had changed. It baffled me then, and still baffles me... I had done enough to know the whole experience was easy to manipulate and yet I couldn't see any evidence of effort. It felt very much like "we tried nothing and it didn't work". Maybe I was missing something, I don't know.
I think there's a few different problems here:
1. I think the author is correct to highlight that underlying motivation, which is so important to teaching. We can focus on a high-minded Constructivist ideal of intrinsic motivation, but motivating students is also what grades are for, and classrooms, and learning with peers, having someone pay attention to your work, and so on. I do think that the AI chatbot is not capable of providing that. But the AI can still ENGAGE with that motivation. I have a feeling Khanmigo never did. I suspect some of their privacy controls also kept it from ever creating a good model of the learner.
2. The entire experience was very focused on answers, on solving problems. And yet it focused on that so reluctantly. The obsession with not giving away the answers had the unfortunate result in not engaging with anything but answers. It felt like it was always teasing at something it would withhold. But questions are cheap, especially rote questions, they shouldn't be treated as precious.
3. This may be unfair, but I infer the makers of Khanmigo did not build it with love and craft, which is unfortunate. I don't know enough to say why. Though there's a weird confidence to the product and to all of Khan Academy that I feel does not bring the necessary humility.
4. The chat experience was asymmetric in the wrong direction. In each turn the AI can and will spew out paragraphs of text, and the student types a couple words. Chatbots DO NOT have to work like this, they operate wonderfully with large chunks of user input. But you have to have the modality to receive it, the openness of experience to elicit that response, and the respect to make use of that response. Speech can be good, but it's a real challenge in a classroom environment (though if you are investing in Khanmigo, you can also invest in the other hardware to make it possible).
5. In some fairness, individual tutors don't represent any pedagogy at all. It's usually just someone making it up as they go. Motivation is a huge aspect, studying with someone sitting next to you, watching you study, will do wonders for keeping you focused! But also a tutor will have a natural theory of mind, hopefully the ability to see a student get the wrong answer and get inside the student's head to model their understanding and find a path to a correct understanding. That's a very challenging task, not beyond GPT-4 entirely but still quite challenging, especially if you are optimizing cost or time to response.
6. A good experience will do a TON of prework. I don't know if Khanmigo does that, if they've reshaped and expanded each educational module into a thorough playbook for the AI. That's where you have a chance to model things like wrong answers (there are actually SEVERAL books enumerating wrong answers in math), and use that to plan out responses.
If Khanmigo is not successful, I think that's on Khanmigo. Still teaching is very challenging. And maybe all of Khan Academy fails in this way: it does not seem to acknowledge the challenge of what they're trying to attempt. They seem far too confident in what they do, and not nearly creative enough.
Maybe we should focus only on having AI for the correcting/internalizing part instead? A purely deterministic grader and problem creator creates an endless stream of problems and solutions. The LLM's job is just to help you get to the point that you can assign a random problem from this problem set and the student will be able to answer it without assistance. Hopefully the teachers lecture at the beginning of class helped with understanding as well (I suspect that it would), and also provides a more broad "theory of mind" about what we are doing than the myopic LLM's "Let's get you solving this problem" does. Also, humans are social creatures, and being taught by another human being (even if it were less efficient) is likely beneficial to most people.
tldr; Normal human teacher, deterministic grading to gauge progress, and LLM assistants to help the student get to the point that they can pass the deterministic grading.
Think less science and more a conversation; establishing syntactic and semantic ground rules for the speakers grasp of a behavior relative to problem.
It's not necessarily about teaching something concrete. It's about teaching a social style in an abstract way. It's discussing a painting or piece of music, in math the symbols used to represent possible number/geometry generators.
I never just watch the videos. I rewrite what is written out. To recreate the song and dance so to speak.
Pragmatically speaking life is just performance art. Only a few specific realms of science move the needle for humanity; biological health maintenance (across all contexts; housing, food, not just medical) and our research in energy infrastructure.
Everything else is just one's relative outcome. Result of their sensory experience. Feels like this should be obvious at this point.
The studies that show brain activity goes down as we use AI. Yeah. So? Is it great to stimulate ourselves so much with ultimately arbitrary content our biology immediately tries to forget due to entropy? Heart attacks are up in <60 crowd relative to last generation. Constant interaction with flame wars and tabs v spaces (functional languages and Nix or gtfo!) computer config wars level inanity is not good Bob.
And the essay's premise is: this man wants your child to watch a video.
I am having my kids learn math with Kahn academy if we don't have anything better by that point.
"Now you add the five ... [pause] ... add the five ... add the five to the seven ... to the seven..."
Running them at 1.5 - 2x helps a bit, but it's still annoying to the point that I find it very distracting.
The other idea is that a student only advances to the next topic once they've mastered the current topic (e.g. by getting 4/5 questions right). The theory behind this is that if you don't understand a concept/topic then it is harder to understand topics that depend on that.
This does mean that students will be at different topics/stages. But because of the structure of the teaching a teacher isn't moving on to a new topic leaving students to fall further and further behind.
I grew up in a small rural town where math teachers were rare and good ones were nonexistent. I still remember watching Sal Khan's videos on calculus after struggling to get any grip on it from my frustratingly uninterested teacher who was forced into the position after our actual teacher retired. After I saw them I came back with a fire in me because for the first time he exposed me to the beauty of calculus.
This guy is completely wrong about Khan not inspiring students to care. Sometimes all students need is to see what caring looks like to ignite it in themselves. You can feel how much he cares just by watching him because the video is the product of all that caring, and he does it for millions of people.
But thank you to this blog author for being brave enough to call out Khan's failures, I am sure it'll help education much more than Khan did.
LLMs are great at this. I’m sure they’re misused constantly. But if you ask the right questions, LLMs are excellent learning aides. Especially for symbolic reasoning like mathematics and programming.
You also do not need to linearly follow the topics/videos. You can take a few quizzes from some topic, notice you don't understand something completely yet and instead pick up more trigonometry instead.
I've worked through Khan academy in a way similar to what the author describes as a good flow, but then all within the Khan Academy platform. The platform even allows for discussions and questions.
To me the problem is much more that motivation or a destination is not something that Khan Academy will give you. Then again, so won't most teachers. I much preferred Khan Academy over most of my math teachers, but occasionally I've had teachers who helped me love a subject I barely knew beforehand.
"What he has never had is pedagogical knowledge: an understanding of how people learn, what motivates them, what makes the difference between someone who pushes through difficulty and someone who types “IDK.”"
It is almost as though the author thinks that Khan Academy is an organization of 1, and that they don't employ learning scientists and have a content team and don't publish research papers in the learning science field.
The concept of Khanmigo was great. The implementation was basically useless.
It was a great attempt, but the article explains the problem:
"early OpenAI access"
This was literally the world's very first attempt at an agentic tutor, with (initially pre-release) GPT-4. That is worlds away from what's being built with fable or sol-grade models in 2026, or what will be available in 2028.
I don‘t think Khan is solely to blame for this. These entrepreneurs would probably have done equal amount of damage without Khan, siphoning funds away from proven solutions like paying teachers and providing school lunches, and into ballooning administration costs and 3rd-party tech companies. However Khan is for better or for worse the face of this movement and will justly or unjustly get that blame.
Duolingo is however less ambiguous. There is very little language learning going on on their platform. You trick people into playing their stupid game and watching their stupid ads, thinking that they are learning, when in fact they are not.
He sold his vision in "The One World Schoolhouse" which was essentially global learning via all the best videos (and one unified curriculum) you can make and teachers effectively being technicians/teaching assistants. It was the vision that utopian minded technologists craved in the early aughts.
Khan: Videos + rote exercises
Learning: Projects, complex problem-solving, exploration, argumentation, etc.
Dewey is worth reading.
It was never about replacing exercises, it was there to help you do the exercises from your actual coursework, and to serve as a refresher reference for things you might have already studied but forgot some details about.
Also the videos usually consisted of a while of time spent physically writing out and talking through examples.
IDK. Maybe some sort of RAG-on-textbook solution will let us do it.
Recall practice IS learning. Projects, exploration, and discovery-based "learning" sounds great in theory but in practice, without actual testing, really doesn't live up to the hype.
Adam Boxer is worth reading.
If you're flailing around without having memorized basic trigonometric identities, you're going to struggle severely to apply them to classical mechanics. If you can't do basic matrix multiplication, you're not going to build a graphics engine.
The article tries to damn Khan with an example about probing how neural myelination works. If you understand the foundations of electricity (by learning about capacitance and impedance) and have done some grueling foundational learning in organic chemistry, ready answers might spring to your mind. You need the former to make asking the latter fruitful or meaningful.
Khan deliberately uses the term "scaffolding" - which since we're throwing around dead education theorists harks from Vygotsky - which is precisely what a teacher provides by pantomiming elementary concepts for a child. Ultimately the child needs to internalize these basics to succeed in more complex projects with a sense of independence.
I really don't understand this article. Two things can be true at once.
"essentially global learning via all the Khan videos, in math, by him, and in other subjects, by his team."
Khan Academy is 2010-era state-of-the-art for rote learning and procedural knowledge. That's a helpful resource, but as one world schoolhouse, there is:
- No room for creativity, or much of anything similar from students (outside of the fantastic JavaScript projects).
- No room for outside innovation.
- A personality cult. Any ideas MUST be attributed to Khan.
Similar with the trend to auto-cut pauses that speakers make, just to cram in more text into a youtube video. Destroys not just the pacing but any chance for the mind to take its time.
I'd love to see studies about this.
I did the trial of Brilliant the other day. You can get 7 days free, then another 7 days added if you cancel the first trial before the end. After that you can continue to get 2 free lessons per day with ads. I think it's about $20 a month for unlimited lessons. The maths problems are more visual (which makes them quite intuitive) and it has a nicer, more polished UI than Khan Academy. It's easier to stay in the mood for doing more lessons. The main downsides compared to Khan are that the explanations are quite minimal (short descriptions, no videos), and it doesn't reach as advanced a level as Khan. I'm looking at it as more a way to do a refresher on the maths I already learned at school rather than a serious learning tool. If it were $5 a month I'd probably consider subbing for a couple of months and trying to blitz through all the material before switching back to Khan to fill in the gaps.
Playing videos at 1.5-2x fixes the problem entirely. My attention is held without any effort.
If a concept is new or difficult enough, I just slow the video back down or pause it whenever I need to think things through.
Obviously one can overdo it. And there is variance in what people require. But not giving things time, instead giving in to the dopamine-craving, that's killing your learning abilities.
1. Teacher 1 speaks at 100 words per minute.
2. Teacher 2 speaks at 50 words per minute, for the exact same content as Teacher 1.
You're not really losing anything by playing Teacher 1 at normal speed vs Teacher 2 at 2x speed. For every learner, there will be an optimal teaching speed for a given topic. This may mean listening to some audio at normal speed, some at higher speeds, and some even at slower speeds. I always customise my playback speed to the specific content I'm trying to learn.
It is the highest level of learning, but still a level of learning.
Only if your goal is to teach people how to solve that particular problem. But it’s not the best way to teach mathematics.
For my money, the best way is problem based. I heard of one math teacher who had a video playing as the students walked in. The video was a weirdly shaped bucket being slowly filled by a hose. Students came in, saw him sitting up the back, got fidgety then someone asked “how long will this take??”. At that, he shot up and said “good question”! And they spent the rest of the class trying to estimate how long it would take for the bucket to be filled. Complete with diversions into flow rate, volume, and so on. He made the students ask the question. Then got the students to try and answer it themselves, helping them work through their own ideas on the board.
That’s beautiful teaching.
"Paris is the capital of France. Paris is the capital of France" -> not spaced repetition.
"Paris is the capital of France. [one day later]. What is the capital of France? [three days later] What is the capital of France?" -> spaced repetition.
And in fact, it's this plus practice where the real value lies in learning and memorising. The only purpose of an explanation from a teacher is to help you understand the concept. Generally once you understand it once, you don't need to understand it again - unless you forget, and that's where the practice / spaced repetition comes in. That's why I prefer the teaching part to be as efficient as possible, so I can focus my limited time on the parts that will make me internalise it.
This piece was also cross-posted on the Civics of Technology blog. This piece also has a followup post that you can find at Three Questions on Questions: On Asking, Knowing and Noticing
Two pieces crossed my feed recently, both about Sal Khan and the AI tutoring revolution that wasn’t. The first was Matt Barnum’s reported piece in Chalkbeat, where Khan himself acknowledged that Khanmigo, the AI chatbot tutor he launched three years ago with world-changing ambitions, was “a non-event” for most students. “They just didn’t use it much,” Khan said. His own Chief Learning Officer, Kristen DiCerbo, put it even more plainly: “So far I am not seeing the revolution in education.”
The second was Dan Meyer’s sharp obituary on LinkedIn, titled “RIP Khanmigo & Edtech Industry Dreams of AI Tutors.” Meyer traced the whole arc: the TED talk predictions, the philanthropic subsidies, the increasingly aggressive way Khanmigo inserted itself into the student experience (because students wouldn’t seek it out voluntarily), and the steadily shrinking user projections. His conclusion was blunt: if Khanmigo died with every advantage in the world (early OpenAI access, Microsoft backing, government subsidies, Sal Khan’s Rolodex), what hope should the rest of the edtech industry place in chatbot tutors?
These are important pieces, and I’d recommend reading both. But reading them, I found myself thinking about a deeper question. Not whether the revolution failed (it clearly did) but why it was never going to work in the first place.
To explain I have to go back a bit in time, back during my days at MSU, when I was out there giving talks about technology integration and the critical role played by the teacher in this entire process. And further about the significance of students actively constructing representations of their understanding.
So in these talks I used to show a clip from the Charlie Rose show (see below). It’s an interview where Khan describes how he prepares to teach a new topic. And it’s wonderful. Here’s Khan on learning about, say, Napoleon and the French Revolution:
“I approach it from what my brain would like to see… I like to see a scaffold, I like to see a map… what is the Holy Roman Empire, like where, what is that now?”
He reads Wikipedia first, “just to get the scaffold.” He draws timelines. He copies maps and pastes them onto his digital blackboard. And then he does something genuinely important: he pushes past the surface until he hits the questions that textbooks skip. Here’s Khan on the neuron:
“A biology book will tell you okay the signal goes across because there’s a myelin sheath and I’m like yeah but how does putting a little tissue around a neuron, how does it make the signal go faster? And no biology book will tell you that answer.”
So what does he do? He ponders. He thinks it through by analogy (fiber optics, signal amplification). And then he calls up friends who are biologists or communications engineers and asks: “Does this make sense?” Sometimes they confirm his intuition. And sometimes, beautifully, they say: “You know what, we don’t know.” Khan’s response to that is perfect: “Why didn’t the book tell me that?”
I used to play this clip and then ask the audience a simple question: Look at everything Sal Khan does to learn something. He reads widely. He scaffolds. He draws. He questions. He calls friends. He argues. He makes connections across fields. He builds intuitive understanding from the ground up. And then he builds something to capture all that he had learned to share with others.
Now… how does Sal Khan want my kid to learn?
Watch a video.
The room always got it immediately.
Because nobody in that room would accept that for themselves. We all know, intuitively, that watching someone else explain something is not the same as understanding it. We would never settle for that as learners. And yet, somehow, we accept it as a solution for other people’s children. This is a version of a phenomenon that I have written about earlier: The reductive seduction of other people’s problems.
But there’s a second layer that I think is even more important, and that neither the Chalkbeat piece nor Meyer’s critique quite names. Khan’s personal learning wasn’t just active. It had a purpose. He was learning in order to make something. The video was his construction, his artifact, the thing he was building. That’s what pulled him through the hard parts, through the myelin sheath question and the calls to friends and the hours of immersion. He had a destination.
Students watching the video have no such destination. They’re receiving the product of someone else’s learning process. And when Khanmigo came along, the revolution was… a chatbot to help you receive more efficiently. Still no purpose. Still no making. Still no reason to push through difficulty. No wonder DiCerbo reported seeing more “IDK IDK” than substantive engagement. No wonder teachers at early-adopter schools found that students “didn’t really care for the bot.” Why would they? There was nothing at stake for them.
This is where John Dewey, writing over a century ago, becomes useful. Dewey argued that learning is built on four natural impulses: the impulse to inquire, to construct, to express, and to communicate. He saw these not as skills to be taught but as drives already present in every learner, drives that education should work with rather than suppress.
Go back to the Charlie Rose clip and watch Khan through this lens. He is living all four. Inquire: the relentless “why” questions, the refusal to accept surface explanations, the “why didn’t the book tell me that?” Construct: the timelines, the maps, the blackboard drawings, the scaffolds he builds for himself. Express: the video itself, Khan giving form to what he’s understood. Communicate: calling up buddies, testing his ideas against other minds, discovering together what is and isn’t known. And then building a representation of his learning, with his own unique voice and style, and sharing it with the world. Inquiry, construction, communication and expression—all in one go! Intermingled so well that it is difficult to tell them apart.
All four impulses, firing beautifully.
And none of them available to the student on the other end.
Khan’s great error, I think, was not a failure of effort or sincerity. It was a failure of educational imagination. He experienced the full richness of learning and then designed a system that offered students only the residue. He gave them the destination without the journey. And because the journey is where motivation lives, where purpose lives, students quite reasonably declined the offer. First they declined the video (or rather, passively consumed it). Then they declined the chatbot. In both cases, the diagnosis was the same: nobody had given them a reason to care.
And this is what I think the edtech world keeps getting wrong. The assumption is that if you can deliver the right content, in the right way, at the right time, learning will follow. It won’t. Not without purpose. Not without the impulse to inquire, construct, express, and communicate. Not without, in Dewey’s sense, the learner actually doing something.
Teachers know this. It is, in fact, a large part of what teachers do: take something a student is not yet interested in and create the conditions that make them interested. Not through tricks or gamification but through the design of experiences that activate those Deweyan impulses. That is the work. And it is work that no video, and no chatbot, has figured out how to do.
I’ve written recently about how evolution’s answer to an unpredictable world was not “more data” but play, and about how children are optimized not for pattern-completion but for exploration. The connection to the Khan story is direct. Khan Academy, and then Khanmigo, are both autocomplete strategies: one autocompletes explanation, the other autocompletes tutoring. Neither makes room for the exploration, the construction, the messy purposeful making that is where learning actually happens.
Khan himself seems to have arrived at something close to this realization. “I think our biggest lever is really investing in the human systems,” he told Barnum. That’s a remarkable sentence from someone who has spent nearly two decades trying to improve education by routing around the humans. Whether his benefactors in the technology industry will be as excited to invest in human systems as they have been in software that tries to replace them… that remains to be seen.
Endnote: I don’t usually bring TPACK into my blog posts. I mean, how much weight can a Venn diagram carry? But this might be the cleanest case I’ve ever seen. Khan has technology knowledge in spades. He clearly has deep content knowledge (that Charlie Rose clip is proof enough). What he has never had is pedagogical knowledge: an understanding of how people learn, what motivates them, what makes the difference between someone who pushes through difficulty and someone who types “IDK.” The circle marked P in the Venn diagram. That is where the humans live. Content and Technology mean nothing without that.