What I Learned Talking to Nearly 30 Parents About Their Kids and AI

Notes from six weeks of customer discovery, written mostly so I can be honest with myself about what I am seeing.

For the last six weeks I have done very little other than talk to parents. Nearly thirty of them, across Winnipeg, Toronto, and the Bay Area, on calls that were supposed to be thirty minutes and routinely ran past an hour because parents, it turns out, have a great deal to say about how their kids are doing in school and almost nobody asking them about it.

I went in to learn whether there is room for a new kind of learning tool for kids. I came out with something more interesting than a yes or no. I came out with a fairly detailed picture of a generation of parents who are quietly improvising their own education systems at home, who are deeply ambivalent about AI, and whose kids are forming opinions about this technology that are sharper and stranger than most of the headlines suggest. This is my attempt to write down what I heard while it is still fresh, and to think out loud about what it means.

A note on method. I tried to follow the spirit of *The Mom Test* throughout, which is to say I tried very hard not to pitch anything and instead to ask about what parents had actually done, paid for, and abandoned. The most useful sentences in any of these conversations were never the ones where a parent told me my idea was good. They were the ones where a parent told me, unprompted, about the tutor they fired, the app their kid quit after two weeks, or the night they sat at the kitchen table until eleven thirty trying to teach algebra to a kid who did not want to learn it.

The thing nobody tells you: school has become a black box

If I had to name the single most universal complaint, it would not be about teachers or curriculum or screens. It would be about information. Parent after parent described the same experience of having almost no idea what is happening with their child's learning between report cards.

One father put it about as starkly as anyone could. Your kids are the most valuable thing to you, he said, and the school is a black box. You put them in and hope for the best. He had a theory about why it got worse, which is that kids used to bring textbooks home. Parents could open the book, see what was being taught, re-read it themselves, and actually help. Now the material lives online behind a login on the kid's account, and the practical effect is that parents are locked out of their own children's education.

Several parents told me they hack into their kids' school portals to see assignment lists, because the school will not surface them any other way. One mother said this with no embarrassment whatsoever. Of course I do that, she said. How do you think I keep up with what he is doing. A father in the Bay Area described clicking seven times through a school website just to discover that his son had a missing assignment, and that the missing assignment was, of all things, to watch television.

What struck me was how rarely this is framed as a feature request and how often it is framed as a kind of grief. These parents are not asking for a better dashboard. They are asking to be let back into a relationship they feel they have been shut out of. I did not expect transparency to be an emotional issue. It is.

Parents are running shadow schools, and they are exhausted

Once you start listening for it, you notice that a large share of these families have built a second, informal school that runs in the evenings and on weekends. Costco workbooks. A standing hour of homework at the kitchen table, enforced. A math tutor deliberately kept one grade ahead of the school so the kid always meets material as review. Daily twenty-minute math sessions that one mother described, with real feeling, as I need to drag him.

Two fathers I spoke to, one a PhD engineer and one a retired Silicon Valley founder, had independently arrived at the exact same strategy of hiring a tutor to run a grade ahead of school. Neither knew the other. They had both reasoned their way to the same workaround for the same problem, which is that school moves at the pace of the slowest kid in the room and their children were bored.

That phrase, the pace of the slowest, came up again and again from the high-agency parents. The system aims at the average, one father said, and if you want your kid even a little above average, they are not teaching for that. Another, an academic herself, told me flatly that her son's school was too slow, and that she makes the work harder at home and it is still boring to him.

The cost of running these shadow schools is real, and it is mostly paid in parental time and patience. The most consistent finding across every family, regardless of income or philosophy, is that the parent eventually hits a wall. Sometimes it is a knowledge wall. I did pre-calculus twenty-something years ago, one mother said, and I would not be comfortable teaching it now. More often it is an emotional wall. A Bay Area father described it precisely: when he helps with math, he gets emotional, the kid senses his disappointment, and it becomes a feedback loop that ends in tears. He had concluded, reluctantly, that a third party simply works better than a parent, not because the parent knows less but because the parent cares too visibly.

That insight, that the value of a tutor is often emotional rather than informational, reframed a lot of what I thought I was building.

The kids hate AI, and they are not entirely wrong

Here is the finding that surprised me most and that I think is the most important thing in these notes. A striking number of these kids actively dislike AI. Not in a vague way. In a considered, almost principled way.

One teenager refuses to use ChatGPT because she believes that if she offloads her thinking to it, she will lose the ability to judge how much effort a task actually requires, which is a more sophisticated argument against AI than most adults could make. Another kid, deeply artistic, rejects AI because it does not feel genuine to him, and his father, who works in AI, half agrees with him. A third turned down direct career advice from a Hollywood producer to learn AI tools and simply was not interested. Several kids will only use AI as what one father called an enhanced search engine, throwing a question at it to get an answer and nothing more.

There is a pattern within the pattern, and I want to be careful stating it because it is based on observation rather than anything rigorous. The resistance seems stronger among girls. More than one parent, including a Bay Area father with both a daughter and sons, told me the same thing in almost the same words: the girls want to be taught by a human, the boys are more open to being taught by a machine. I do not know what to make of this yet. It could be a real difference in how the technology is being received, or it could be an artifact of who happened to be in my small sample. But it showed up too many times to ignore.

The deeper thing underneath the kids' resistance is that they have correctly identified the failure mode of current AI. They have watched their classmates use it to not think. One father caught his son the night before a math exam pulling answers out of a chatbot, understanding nothing, and realized his son was going to walk into the test having learned exactly zero. The kids who reject AI are, in a sense, defending themselves against a tool that makes it very easy to skip the part of learning that is actually learning.

This is the part the headlines get wrong. The story is not that AI is making kids dumber. The story is that the dominant way kids currently use AI, as an answer vending machine, makes it easy to avoid the cognitive work, and a meaningful number of thoughtful kids have noticed this and opted out. That is not a technology problem. It is a design problem.

The other thing the kids hate: repetition

If AI is the rejection I did not expect, repetition is the one I should have. Almost every product death I heard about followed the same script. The kid is excited for two to four weeks. Then the thing becomes repetitive, and the kid quietly stops.

Kumon was the most consistent casualty. I lost count of how many families had tried it and pulled out, and the reason was always some version of the same sentence: too much repetition, the kid was bored, it did not fit how their brain works. One mother described her daughter banging the iPad down and refusing to continue when the app sent her backward through levels for getting answers wrong. Another described how her son, who genuinely loves math, could not stick with a leading math app because answering the same kind of question ten times in a row drained the joy out of his favorite subject.

The cleanest version of this came from a mother who is herself a software engineer and a heavy professional AI user. The most challenging challenge you will have, she told me, is how to keep kids coming back. At first they are very interested. They love to try new things. And then they do not stick. She has a graveyard of canceled subscriptions to prove it. She was not warning me about acquisition. She was warning me about week three.

The counterexample everyone reaches for is Duolingo. I heard about it constantly, and always with a kind of awe. A kid who chose, with no prompting, to learn Yiddish and hit a seventy-five day streak. A kid who will not go to bed until he has done his Mandarin. A grandfather with a seventeen hundred day Italian streak. Whatever Duolingo has figured out about getting people to come back every single day, parents have noticed it, and they want it pointed at the things their kids actually need to learn. One father, a founder himself, put his finger on the tension that this creates, which is that too much gamification can hollow out the learning it is supposed to support. The engagement and the depth pull against each other, and nobody I talked to has seen a kids' learning product hold both at once.

A few products worth watching

I want to be fair to the people building in this space, because some of them are clearly thinking hard about the same problems.

The most relevant launch happened while I was in the middle of these interviews. Brilliant, the interactive learning platform, released an AI tutor called Koji. It is worth understanding because it is the closest thing to a direct expression of the thesis I keep hearing from parents. Koji is built around the Socratic method, which means it tries to coach kids toward answers with guided questions rather than handing the answer over. It is voice-first, with a voice layer built on ElevenLabs that the team apparently spent months tuning so it would sound present and unhurried rather than robotic. And critically, it does not start from a blank prompt box. It can see the lesson the kid is working on and where they are stuck, and it can sketch and annotate on the screen as it explains. Brilliant's founder framed the launch with a line that could have come straight out of my interview notes: AI is making kids dumber, it should be making them geniuses. At launch it covers foundational math and coding from roughly grade five upward, with more advanced subjects promised later in the year.

I think Koji is directionally right about almost everything, and its existence is good news rather than bad news, because it validates that the Socratic, do-not-just-give-the-answer approach is where serious people believe this is heading. The open questions, to my mind, are whether a tutor that lives inside Brilliant's own curriculum can follow a kid into their actual school work and their actual curiosity, and whether being Socratic by default occasionally tips into being Socratic to a fault. One Bay Area father, describing his kids' experience with another AI tutor, told me the failure he saw was the opposite of giving too many answers. The tool was so determined not to let his kids cheat that it withheld the clear explanation they actually needed. Sometimes a kid is not trying to cheat. Sometimes they just need someone to run through three good examples. A tutor that cannot tell the difference is its own kind of broken.

It is also worth naming the incumbents that parents have tried and what happened to them. Khan Academy is nearly universal and nearly universally outgrown, treated as a solid foundation for younger kids that older ones drift away from. IXL is widely used and widely resented for its repetition and its habit of punishing mistakes. Prodigy and similar gamified apps win a few weeks of enthusiasm and then fade. Beast Academy earns unusual loyalty from the parents of strong math kids, and when I asked why, the answer was that it presents the same underlying concept through many different scenarios rather than the same drill over and over, which is exactly the anti-repetition principle the kids are begging for. And Khanmigo, Khan Academy's AI tutor, came up from a parent with genuine inside knowledge, who argued that its real constraint is not technical but regulatory and institutional. Because it has to work inside schools, it cannot be as aggressively personalized as it could be, and his conclusion was blunt: anything truly personalized will have to go directly to parents, because schools are too cautious and the laws are tightening, not loosening.

That last point deserves its own paragraph, because it is the most important strategic thing anyone told me.

Why this has to go around the school, not through it

The conventional wisdom is that education technology should sell to schools, because that is where the kids and the budgets are. Almost everything I heard cuts against this.

Schools are slow, they are risk-averse about anything involving AI, and in several jurisdictions they are moving toward outright restriction. A number of the parents I spoke to described their kids' schools actively discouraging AI, which has produced the entirely predictable result that the kids use it anyway, in secret, with no guidance at all. One mother told me her daughter said all of my friends do it regardless of what the school says. Another family had a child formally punished for using ChatGPT on homework. He was made to present to the class and lost a month of recess. When his mother asked him why he did it, his answer was heartbreakingly reasonable. He said he just wanted to save time and get to the next thing.

So you have schools banning the tool, kids using it underground anyway, and no adult in the loop teaching them how to use it well. The product that wins this moment is not the one that gets approved by a school board. It is the one a parent can put in front of their own kid, that the kid will actually want to use, and that teaches them to think rather than to copy. That product reaches the kid through the parent, at home, on a weekend afternoon, in the slot where the Costco workbook used to go.

What I actually believe, having listened

I will try to state my own view plainly, since the whole point of writing this is accountability.

I think the bored, capable kid is the most underserved student in the system, and almost nobody is building for them. Enormous energy goes toward remediation, toward catching struggling kids up to grade level, which matters. But the kid who finishes the worksheet in five minutes and then sits there, the kid whose parent is hiring tutors a grade ahead and supplementing at the kitchen table, that kid is being actively held back by a system designed around the median, and their parents know it and are paying out of pocket to fix it.

I think the right product is Socratic, but not religiously so. It should lead a kid to the answer when leading is what they need and explain clearly when explanation is what they need, and the entire craft is in knowing which is which. The over-Socratic tutor that refuses to ever just teach is failing kids in a quieter way than the answer machine, but it is still failing them.

I think the engagement problem is the whole game. A learning product that is pedagogically perfect and boring by week three is worthless, because the kid will not be there in week three. Whatever Duolingo knows, this category has to learn, and it has to learn it without letting the gamification eat the actual learning. That is genuinely hard and I do not want to pretend otherwise.

I think the parent has to be in the loop by design, not as an afterthought. The black box is the wound. A product that quietly shows a parent what their kid is working on and where they are growing is solving the most emotionally charged problem I heard, and it is solving it almost for free.

And I think, despite the kids' resistance, or really because of it, that the opportunity is to build something that earns the trust of a generation that is rightly suspicious of AI. The kids who reject the answer machine are not anti-learning. They are pro-learning, and they have noticed that the answer machine is anti-learning. A tool that respects that instinct, that makes them do the thinking and feels more like a sharp notebook or a patient older sibling than a chatbot, is a tool they might actually let in.

This is what I am building toward with HOPE, and I would be lying if I said I had all of it working yet. But these six weeks convinced me the problem is real, the parents are ready, and the kids will tell you the truth if you build something honest enough to deserve it.

If you are a parent of a kid roughly between grade three and grade eight and any of this sounds like your house, I would genuinely love to talk. The conversations have been the best part of this whole thing.

Written on June 13th, 2026.