Alpha School is selling parents a $75K experiment the research never supported. The generation that cannot afford it is the one being asked to buy it.
Alpha School grew from one Austin campus to fifteen-plus nationwide. Unbound Academy. Khan Lab School. Bill Gates speculating that AI will replace human teachers within a decade. The machinery of Silicon Valley disruption, aimed at the one expense millennial parents already cannot afford — their children’s education.
Here is the math.
Alpha School charges between $40,000 and $75,000 a year depending on the campus. The median millennial household income is roughly $72,000 before taxes. One year at Alpha School consumes between 55 and 104 percent of a typical millennial family’s entire pre-tax earnings. My childcare for Eva and Ben runs $2,400 a month — $28,800 a year — and that number already keeps me at the kitchen table at night running columns. Alpha School’s tuition is one-and-a-half to two-and-a-half times what I pay someone to watch my children while I work.
For that money: two hours a day staring at an AI tutoring screen. The rest is life-skills workshops supervised by “coaches” who are not necessarily licensed teachers. That is not replacing teachers with better teaching. That is replacing teachers with cheaper labor and a screen, and billing parents the cost of a semester at a private university for the privilege.
The research has never backed the pitch.
A NBER meta-analysis shows human tutoring produces consistent learning gains. A 2026 Brookings study shows generative AI makes computer-assisted tutoring more interactive — students talk to the program, the software adjusts. All real. All worth acknowledging as incremental improvement in educational technology.
But “AI tutoring is better than a human teacher”? That claim depends on which studies you pick and which comparison you choose. A 2025 meta-analysis found “significant positive effects” of AI tutoring. Compare AI tutors directly against human tutors and the advantage vanishes. Studies from 2015, 2024, 2025, 2026 converge: AI tutors are not worse than human tutors, but they are not better. The Harvard physics study everyone cites — students said they learned faster, felt more motivated — used students who walked in already highly motivated with strong study skills, and the AI tutor had been built by the same professors who taught the course. Not a controlled trial. A proprietary tool evaluated by the people who built it.
This is the pattern: cherry-picked studies, proprietary tools, and a gap between the pitch and the evidence wide enough to drive a tuition check through.
The families being targeted are the ones who can least afford the experiment. A millennial couple earning the median — $72,000 before taxes, roughly $58,000 after — spends $24,000 to $36,000 a year on childcare for two kids under five. They carry a median student-loan balance of $35,000, with monthly payments that shifted from $92 to $270 to $550 as federal repayment plans changed underneath them. Their mortgage or rent is another $2,000 a month in a city like Philadelphia. More than half their gross income goes to housing, childcare, and debt service before they buy groceries. Someone tells them the answer to their child’s education is $75,000 a year and a screen.
I sat at my kitchen table last spring doing the math on Eva’s education. Not Alpha School. Not private school. The basic K-through-eight pipeline — whether we could afford the parish-school tuition my parents managed on a single USPS supervisor’s income. The math did not work on the first pass or the third. My parents put three kids through Catholic school on my father’s wage because the infrastructure was there: the parish subsidized tuition, the women religious staffed classrooms for almost nothing, the diocesan system was built for families like ours. That system has contracted by more than half since the 1960s. The standard of living my parents delivered on one income requires roughly 2.7 times that income in real terms for my family to replicate with two. I am flagging my position: two incomes, a rowhouse we bought with help from David’s grandmother’s estate, college-educated parents who navigated my financial-aid paperwork. The math still does not work for a school my parents absorbed on one salary. For families without those advantages, the gap is wider, and the pitch is crueler.
Anne Helen Petersen documented how millennials “fully conceptualized themselves as walking college resumes.” The AI school pitch extends that self-conception to the next generation at a price that was never part of the original calculation.
Swift’s “Anti-Hero” stages the recognition that the speaker is herself the problem — “It’s me, hi, I’m the problem, it’s me” — where the problem is structural pressure metabolized into character. That is the posture the AI school pitch exploits. Public schools are underfunded. The Pell Grant covers 25 percent of a public university instead of the 80 percent it covered when my parents were young. Schools in my city receive half the per-pupil funding of the suburban district I grew up in. The pitch tells parents the problem is that they have not found the right product — not that the pool was drained, as Heather McGhee has documented, but that they owe their child a $75,000 algorithm.
Pamela Druckerman’s French middle-class families did not solve education by buying a product. They had subsidized crèches, free maternelle from age three, universal pediatric coverage, year-long paid maternity leave. The structural policy came first. When someone tells you the answer to American education is an algorithm and a checkbook, they are selling the same lie every generation of American parents has been sold: that if your family is struggling, the problem is you, and the solution is a more expensive version of the thing that is not working.
What actually works is not a mystery. A 2024 study found that middle schools in low-income areas improved outcomes when human tutors had access to AI tutoring support. A 2026 study showed experienced teachers using AI for lesson planning critically revised the AI’s output to connect it to the broader curriculum. They used the tool as a tool — not as a replacement for themselves. Teachers are not afraid of the technology. They are afraid of being replaced by a bad implementation of it sold to administrators who never consulted them.
The AI school boom is not innovation. It is what happens when a public good is defunded and someone arrives to sell the wreckage back to the people who can least afford it — at a price most families cannot pay, to fill a gap that policy choices created, in a market that would not exist if the pool had not been drained.