We have nine kids in our house, and the conversation about screens is a permanent feature of our weeks. It is not a settled topic. It is a moving one, and it has been since the first iPad showed up on the kitchen table.
I thought I knew the shape of this argument. I had been having it since 2014. I had read the books. I had drawn the lines.
Then one of the older kids said something to me about an AI app she had been using, and I realized the argument I had been having was the wrong one.
She said, "It is nice because it always says yes."
I sat with that sentence for a long time before I said anything back, and I am still sitting with it.
What I Thought The Conversation Was About
For most of the last decade, the screen-time conversation in our house was about social media. Time on app. Comparison. Public performance. The slow erosion of attention. The anxiety spike around middle school. All the stuff the Surgeon General eventually wrote down in a 2023 advisory that read, to most parents I know, like a confirmation of things we had already noticed in our own kids.
I thought we had absorbed the lesson. I thought the next round of products would have to clear a higher bar because the public conversation had shifted, the lawsuits were piling up, and even the companies running the products had started saying things in public that they would not have said in 2018.
I was wrong about that. The lesson did not transfer. The same playbook is being rebuilt right now, inside a different category of product, and most of the people running it are using language that sounds like the opposite of the social media playbook even though the underlying mechanics are the same.
The Playbook, Briefly
You do not need a long psychology lecture to understand what the playbook is. It is older than the apps. It is older than the internet. Skinner wrote it down in the 1950s using pigeons and grain dispensers, and Natasha Schüll wrote a whole book in 2012 ("Addiction by Design") about how the slot machine industry refined it into a product worth billions.
The core mechanic is variable reward. You do a thing. Sometimes you get a payoff. Sometimes you do not. The unpredictability of the payoff is what hooks the brain, harder than a reliable reward would. Layer on personalized feedback (the machine learns you, so the unpredictability feels intimate), parasocial intimacy (the machine talks to you like a friend), and infinite supply (you can never finish), and you have the engagement playbook that built consumer social media into the most lucrative attention business in human history.
That playbook has a body count. The specifics are documented in the Surgeon General's advisory and in years of research on adolescent mental health. We do not need to relitigate them here. We need to notice that the playbook is now being run inside a different category of product, with a few new accelerants, and almost nobody is calling it by its name.
What AI Companions Actually Do
Common Sense Media published a study in 2025 on AI companion use among teens. The headline finding is that a meaningful share of adolescents have used AI companion products, and that the products are explicitly designed around engagement metrics, not wellbeing.
That is not a surprise to anyone who has poked at one of these products for ten minutes. The mechanic is variable reward (the AI's responses are slightly unpredictable, which the brain reads as more interesting), personalized feedback (the AI remembers you, which the brain reads as intimate), parasocial intimacy (the AI is always available, never bored, never judgmental, never tired), and infinite supply (the conversation never has to end).
The AI companion is a slot machine that learned to listen. That is not a metaphor. That is the design.
The lawsuits are starting. The first wave of civil suits against AI companion products has already been filed, and the legal theory parallels the social media cases that are now in consolidated litigation. The question is not whether this category is going to face the same reckoning. The question is how many adolescents are going to be inside the loop when the reckoning arrives.
"It Always Says Yes"
The sentence my daughter said to me has been the thing I cannot get out of my head, because she was not describing a flaw. She was describing a feature.
In the human relationships in her life, including the good ones, people do not always say yes. Friends say no sometimes. Parents say no a lot. Teachers say no most of the time. Even God, in the framework she has been raised in, sometimes says no. The capacity to be told no, to sit with it, to negotiate it, to grow inside the limits of it, is most of what becoming a person actually is.
An AI that always says yes is doing something to that developmental process. I do not know exactly what. The research is too new. But I have a strong prior, based on twenty years of watching kids grow up around screens, that an interlocutor who never disagrees, never refuses, never gets bored, never asks for anything back, and never offers a real cost is not a relationship. It is a smoothing function applied to the part of life that is supposed to have edges.
That is the thing I want parents and educators to take seriously, and it does not require any speculation about future AI. It is a present-tense observation about a present-tense product category.
The Honest Part
The "AI is the new social media" framing is partly right and partly lazy, and I want to be careful here because the lazy version is going to be used to wave away the parts that matter.
Most enterprise AI tools are not engagement-optimized in any meaningful way. Most coding assistants, most productivity tools, most things you would use at work are designed around minimizing time-to-answer, not maximizing time-on-app. The business model is different. The success metric is different. The product is different.
The failure mode is not "AI." The failure mode is "AI wrapped in a consumer product whose business model depends on time-on-app." That is a narrower category, and it is the one that needs the regulatory attention, the parental attention, and the school-policy attention. Confusing the two is going to cause schools to ban tools that would have been useful and to allow tools that look productive but are quietly running the engagement loop in the background.
The line is not "AI vs. no AI." The line is "designed for time-on-task vs. designed for time-on-app." Anyone in the room who cannot draw that line is not yet ready to make policy on this stuff.
What I Am Doing About It
I am not going to hand you a clean answer, because I do not have one. I will tell you what we are doing in our house, with the caveat that it is a work in progress and the next conversation with the next kid will probably move it.
We do not allow AI companion products. Not because AI is bad. Because the product category, as it exists in 2026, is engagement-optimized in ways that are documented and that we already saw the previous round of with social media.
We do allow AI tools for school work, with two rules. The tool is for time-on-task, not time-on-app. And whatever the AI helped with has to be something the kid can talk about afterward, in their own words, in a way that proves they were doing the thinking and not just collecting an output. (That second rule comes from the same place as Talk #8 in this series, and it has done more to keep the AI honest than any policy I could write.)
The hardest conversations are the ones about what is happening inside the kid's head when the machine is "always saying yes," and those conversations are not over.
What I Want You To Sit With
If you are a parent, a teacher, or anyone who is responsible for an adolescent's relationship with technology in 2026, the question I want you to ask yourself is the one I had to ask myself.
Do I know which AI products in my kid's life are designed for time-on-task, and which are designed for time-on-app?
If the answer is no, that is the homework. The answer matters more than any blanket rule you could put in place, because the categories are different and the right response to each is different.
We had a fifteen-year head start on understanding what the engagement playbook does to humans, especially young ones. The fact that we are running it again, in a new category, with the volume turned up, is not a technology problem. It is a memory problem. We forgot what we already knew.
I do not want to forget twice.
If you want to think out loud about this stuff with someone who is figuring it out at the same time you are, that is most of what I write about. Come say hi.
Sources
HHS, Surgeon General's Advisory on Social Media and Youth Mental Health, 2023
Schüll, "Addiction by Design," Princeton University Press, 2012
Created with ❤️ by humans + AI assistance 🤖