You have probably done the math in your head already. Two hours of academics in the morning, the rest of the day for sports, projects, languages — and test scores that supposedly land in the top percentiles. The brochure says your kid will move twice as fast through the material because the software adapts to exactly where they are. If you can write the tuition check, you have almost certainly asked yourself the question: is an AI-first school actually better for my child, or am I buying an expensive version of a screen?
I have spent a long time watching education technology get sold, adopted, and quietly revised. So here is the answer up front, before the nuance: AI in education can compress how fast a child covers content, but the things it compresses away — friction, boredom, getting it wrong in front of someone — are often where the learning actually lives. Whether that trade is worth it depends on what you think school is for.
What you are actually being sold
The model marketed by schools like Alpha and its imitators is real, and it is more coherent than the skeptics give it credit for. A child works through adaptive software that meets them at their level. A human "guide" — not a teacher in the traditional sense — circulates, handles motivation, and keeps the day moving. Academics get done in a concentrated block. The afternoon goes to whatever the family values: entrepreneurship, athletics, music.
The promise rests on a word that sounds unambiguously good: efficiency. Why would a child sit through a forty-minute lesson pitched at the class median when the software can detect they already know two-thirds of it and skip ahead? Why would a kid who is behind in fractions be dragged forward with the group?
Put that way, the traditional classroom looks indefensible. Twenty-eight kids, one pace, most of them either bored or lost. The AI model says: we fixed that. And on the narrow thing it measures — content covered per hour — it may well have.
What the dashboard cannot see
The trouble starts when you ask what the speed is buying.
Robert Bjork, who has studied memory for decades, gave us the phrase "desirable difficulties" — the finding that the conditions which make learning feel slow and frustrating in the moment are frequently the same conditions that make it stick. Spacing practice out so you partly forget. Mixing problem types so you can't run on autopilot. Having to retrieve an answer from a blank page instead of recognizing it on a screen. Learning that feels efficient is often learning that evaporates by the following month.
This is the quiet problem with optimizing for content covered. A well-designed adaptive system is very good at producing the feeling of smooth progress — green checkmarks, a climbing mastery bar, a child who is genuinely not frustrated. But ease in the moment and durable understanding are not the same thing, and they can run in opposite directions. The dashboard shows you the first. It has no column for the second.
There is a developmental layer underneath the cognitive one. Vygotsky argued that a great deal of what children learn, they learn through other people first — talking through a problem with someone slightly ahead of them, hearing their own half-formed reasoning corrected mid-sentence. Erikson would point out that a school-age child is doing something other than acquiring facts: they are working out whether they are competent, where they stand among peers, who they are when the task is hard. A piece of software can present the next problem. It cannot be the person a child becomes more capable in front of.
We have run this experiment before
None of this is hypothetical, and the AI schools are not the first to make the pitch.
Khan Academy spent years as the great hope of personalized, self-paced learning. It is a genuinely useful resource — I have recommended it, I would recommend it again. But the dream that you could hand kids the platform and watch mastery happen at scale did not arrive the way the early enthusiasm suggested. The organization itself has been candid, in its later work with AI tutoring, about how much careful human design and supervision the thing actually requires to help rather than mislead. The tool got better. The claim that the tool could replace the harder parts of teaching got quieter.
That pattern — bold promise, real product, walked-back claim — is the one to watch for. When a school tells you its students score in the top few percent, the honest first response is not "impossible." It is: measured how, against whom, and what isn't on the test? Standardized scores are a real signal. They are also the easiest thing in the building to optimize directly, which makes them the least surprising thing to find elevated.
Where the model genuinely earns its keep
I do not want to do to this model what its boosters do to the ordinary classroom — flatten it into a straw man. There are situations where an AI-forward approach is the right call, and pretending otherwise would be dishonest.
- A specific, diagnosable gap. A kid who missed a year of math foundations during a move or an illness can recover faster with adaptive drilling than by being reabsorbed into a class that has moved on. This is the technology at its best: targeted, finite, measurable.
- A self-motivated older student who already knows how to learn and is held back by pace. For them, the afternoon freed up by a compressed morning is the actual product, and it can be a good one.
- Families who can supply the rest. If the missing social and intellectual friction gets replaced — by sports teams, debate, a music ensemble, dinner-table argument — the screen hours look very different than they would for a child whose whole day is the platform.
Notice the common thread: the model works best as one component inside a rich life that supplies everything it leaves out. It works worst as the whole thing.
Where it fails, and who should walk away
The reader this is wrong for is the parent hoping the school will do all of it — that two efficient hours and a guide will produce not just a kid who tests well but a kid who is known.
Because the parts that don't fit on a chart are not garnish. The boredom that forces a child to generate their own interest. The group project that goes badly and teaches something a frictionless task never could. The teacher who notices a kid is quiet today and asks why. These are not inefficiencies to be engineered out of the school day. For a developing person, they are frequently the curriculum.
When schools hear this, they answer that a human guide is right there in the room. And one is. But a room with an adult in it is not the same as a relationship in which a child is challenged, misunderstood, corrected, and eventually understood. You can staff the first. You mostly cannot purchase the second, and the dashboard will not tell you which one you got.
So: what are you trading? Speed and measurable scores, for friction and the slower, unmeasured work of becoming a person who can struggle in front of others. For some children, in some seasons, that trade is worth making with eyes open. For most children as a permanent arrangement, I am not convinced it is — and the people selling it have a structural reason not to dwell on the half of the ledger they can't graph.
Try this before you sign anything
This week, if you are touring one of these schools, ask to see a child do a problem they get wrong — not a demo of the platform working smoothly, but a real moment of a kid stuck. Watch what happens next. Who notices, how fast, and whether the response is to help the child sit in the difficulty or to route around it. That ninety seconds will tell you more than any score report in the folder.
Efficiency is easy to demonstrate. Ask to see them handle the mess.
