I run Tutorela, a U.S. tutoring marketplace that sells human tutoring, with AI for students between sessions. I’m saying that up front because this piece argues for that model, and because we learned the lesson behind it the hard way.
Alpha School has become the national test case for AI-first education, and it’s now expanding into public schools in Texas and South Carolina. As districts decide how much of the school day to hand to software, they should hear what happened when we gave students the best learning tool we could build.
We built it. They didn’t come.
For 15 years we’ve run Limud Naim, a tutoring marketplace in Israel whose tutors have taught more than 500,000 private lessons. To fill the gap between those lessons, we built a math learning and practice app with more than 10,000 exercises. Every problem came with a full solution in text and video, and we made the whole thing a game, with progress and rewards, so practice would feel less like homework. A year ago we added an AI assistant that could walk a student through any step, any time, for free.
By any product standard it was the best a student could get. Only about 10% of students used it, and most of those who started didn’t stick around. The exercises were excellent and the AI was patient and clear, but none of that mattered, because the students weren’t showing up.
We’re not the only ones.
A two-year randomized trial in 18 Tennessee middle schools, published this year by the National Bureau of Economic Research, found the same thing with Khanmigo, Khan Academy’s AI tutor. Nearly every student tried it, but the median student messaged it on only a third of the days they practiced, and in only 17% of the practice sessions where they made a mistake. Most messages were one- or two-word answers or clicks on suggested prompts. The math gains matched what Khan Academy practice delivers without any AI. Khan Academy itself says only about 15% of students with access to Khanmigo engage with it. “It may be that human attention is the key ingredient,” the researchers wrote.
Then we added one person.
At Limud Naim we didn’t change the app. We changed who was in charge of it. Each student’s tutor began managing their learning: picking the practice that fit where the student was, assigning it, and following up at the next lesson.
Everything changed. More than half of the students who got assignments from their tutor completed them, up from about 10% who used the app on their own. It was the same app with the same exercises. The only new ingredient was a human who knew the student and expected the work to get done.
That result is why we built Tutorela as a hybrid from day one.
Motivation comes from people.
This is the part AI-first education gets wrong. The bet behind an AI-run school day is that if content is personalized enough and feedback is instant enough, students will learn. Our experience and the Khanmigo data both say content was never the bottleneck. Students already have more free, high-quality explanations than any generation in history. What they’re missing is someone who notices whether they did the work and cares how it went.
A child practices fractions on a Tuesday night because their tutor, whom they’ll see on Thursday, asked them to. AI can generate the perfect worksheet, but it can’t make a 12-year-old feel that someone is counting on them.
Where AI actually belongs
None of this makes me an AI skeptic. I believe AI makes tutoring better, as long as the tutor sets the direction. Then AI makes the time between lessons count: it answers the question at 9 p.m., explains a step a third time without frustration, and offers more practice than any human could grade. The tutor gets a better-prepared student, and the student gets help exactly when they’re stuck.
Put AI in charge and remove the human, and the system loses its engine. You keep the explanations and lose the reason to show up.
What schools and parents should take from this
For school leaders weighing AI-heavy models, the question isn’t whether the software is good. It usually is, and ours was too. The question is who will hold each student accountable for using it. If the answer is “the software,” expect what we and the Tennessee researchers saw: wide access, thin engagement, and students who quietly stop.
For parents, the advice is simple. Use AI tools, but pair them with a person, whether a tutor, a teacher, or a parent who checks in, who sets the goals and follows up.
The future of learning isn’t AI instead of teachers. It’s a teacher in charge, with AI for everything in between.
The real test is not whether a company can succeed in a market once. It is whether it can keep adapting as consumers and technology evolve. It is no longer enough for leaders to see where the future is headed. The best leaders build organizations capable of riding on the opportunities and challenges.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
This story was originally featured on Fortune.com
.png)
1 hour ago
2



English (US) ·