TL;DR: A type of AI school compresses core academics into a two-hour morning block and devotes the afternoon entirely to projects. The logic is to remove friction and accelerate learning (the operator claims 2.6× growth, pending independent verification). This time structure looks a lot like real-task education, but the philosophies run in opposite directions: it believes in removing friction; I believe capability comes from friction. This piece presses one question: when AI optimizes away the struggle in academic learning, where does a child have left to practice facing difficulty?

There is a model of schooling that has generated a lot of discussion in education circles lately. It uses AI tutors to compress a full day of core academic learning into two hours.

It is called “2 Hour Learning.” In the morning, children work one-on-one with AI, advancing through math, language arts, and science at their own mastery pace. According to the operator’s claims, students progress 2.6 times faster than their peers, a figure that is currently self-reported and awaiting independent verification. With academics finished in two hours, the entire rest of the day is freed up for projects, public speaking, entrepreneurship, and outdoor activities. Teachers are no longer called teachers; they are called guides, and their role shifts from instruction to motivation management.

The first time I encountered this model, I had a complicated feeling. Its time structure maps surprisingly closely onto the ratio I use when guiding children through real tasks: compress the knowledge-acquisition phase, and give the time back to real project work. But looking more carefully, I found that our underlying philosophies are, in fact, opposite.

One Believes in Removing Friction. The Other Believes in It.

The logic of this AI school: inefficiency is the enemy of learning.

One teacher with thirty students always levels the pace, fast learners get pulled back, slower ones fall behind. One-on-one AI can fix that. It removes repetition and waiting, letting each child reach mastery by the shortest path. Time saved gets redirected to more meaningful work. This logic is elegant, and it genuinely targets the biggest waste in traditional classrooms.

But my instinct, built from actually working alongside children on learning tasks, runs the other direction: the source of capability is usually friction.

In “Four Kids, One Summer, One Website That Had to Go Live”, I wrote about this directly. The most important growth those four children experienced happened during the two weeks they sat frozen in front of a blank page, unable to figure out what the site was actually supposed to do. Those two weeks of anxiety, chaos, and starting over looked, by any efficiency metric, like total waste. But cognitive construction happened precisely there. If a tool had smoothly fed them the answers on the first day, they would have produced a faster website, and not developed the capacity to think a vague problem into clarity.

Learning Lives in the Struggle

This is not just personal intuition. Learning science has a concept for it: productive struggle.

The idea is that genuine understanding tends to grow in the stretches where you are stuck, confused, and have to find your own way through. When the answer comes too easily, that neural pathway never gets trained. I touched on the same point in “When Assignments Can Be Generated, Why Real Tasks Matter More”: what AI does best is eliminate struggle. It delivers a reasonable-looking answer in seconds and spares you all the pain of being stuck. In many contexts, that is a good thing. In learning, it removes the very site where learning happens.

So the question I most want to press about “2 Hour Learning” is not whether it is fast. It is this: if AI has optimized away every struggle in academic learning, where does a child have left to practice facing difficulty?

A Child Who Spent the Morning Without Friction, Can They Tolerate the Afternoon?

The model’s answer would be: the afternoon projects are exactly where children practice facing difficulty.

That answer is not wrong. But I want to press one layer deeper. A child who spent the entire morning in instant feedback, frictionless, shortest-path-to-mastery learning, when they walk into a real project in the afternoon, one with no standard answers, where they might stay stuck for days and might not finish, will they hold? Will they tolerate that?

The capacity to endure difficulty requires practice. And the way to practice it is to go through the cycle repeatedly: get stuck, feel awful, push through anyway. If a child has never had the chance to run that cycle in academic learning, because AI smoothed the road every time they were about to hit resistance, then the frustration they meet in afternoon projects will feel unfamiliar, something to flee from, not an old acquaintance they know how to get through.

I am not saying AI tutors should not be used. I am saying that treating all friction as an enemy to be eliminated may eliminate learning along with it.

The Real Question Is Which Kind of Friction

This piece is not a case against AI-personalized learning. It has real value: the parts that are genuinely inefficient, repetitive practice, basic mastery, AI makes faster, and the time freed should go toward harder work.

What requires thought is telling two kinds of friction apart. One kind is waste: one teacher talking to thirty listeners, fast students waiting for slow ones, copying things out by hand over and over. AI should optimize that away. The other kind is learning itself: the mental burn of thinking a vague problem into clarity, the conflict of collaborating with others, the slow grind of finding coherence in chaos. That kind of friction cannot be optimized away, because it is the site where capability is formed.

AI making learning faster is a remarkable thing. But we have to watch carefully that in optimizing away wasted friction, we do not also strip out the friction that learning requires.

In the overview piece “From Flip to Climb”, I argued that education means designing a path with enough real challenge and enough support. AI can make the support system very strong. But the challenges on that path cannot all be paved flat. What children have to climb over will always be a real slope, it will not be quick, and it will not be smooth.