On the morning I sat down to write this, Mark Zuckerberg published a six-and-a-half-thousand-word essay promising every person on Earth a “personal superintelligence” — an individualized AI agent, eventually with “a fully private mode… where even Meta or any other service provider cannot see or grant access to your information” (Axios, PYMNTS). Set aside, for a moment, whether you believe the promise. Notice instead what it concedes: the most powerful attention merchant in history now agrees that the future belongs to technology that is personal, private, and empowering — technology that serves the person rather than the platform.
I agree with him. That is precisely why I am building something he cannot.
A revolution we have not tested
Here is the position I want to defend, and it is deliberately unfashionable in both directions: the information and interaction model our children are growing up inside is not inherently good and not inherently bad. It is something rarer and more demanding — it is revolutionary and untested. The American Psychological Association's formal position is that social media is “not inherently beneficial or harmful to young people”; the National Academies concluded the reality is “more complicated” than the prevailing narratives (Chronicle of Higher Education). The scientific fight over whether these platforms are damaging adolescent mental health — Jonathan Haidt's “anxious generation” thesis on one side, researchers like Andrew Przybylski and Candice Odgers on the other — remains genuinely unresolved: a meta-analysis of twenty-seven randomized trials found reducing social media use had effects “statistically no different than zero,” while re-analyses found modest benefits from longer abstinence (Chronicle; EFF).
When the evidence is this contested, honest builders should be suspicious of any strategy premised on certainty — including their own. What nobody disputes is this: our children will live their entire adult lives inside this environment, it is changing faster than any environment humans have ever adapted to, and the machines in it now think, talk, and interact in ways that reach human emotional life directly. The question is not whether to let them in. They are in. The question is who teaches them to live there.
The two failing answers
Society has produced two dominant answers, and both are failing on their own terms.
The first answer is lockout. Australia became the first country to ban social media for under-16s in December 2025. Three months later, Australia’s own eSafety Commission found that 81 percent of under-16s were still using social media, with daily usage essentially unchanged — teens kept old accounts through platform sloppiness, spoofed age checks, or migrated to less regulated corners of the internet (Al Jazeera). And yet the model is spreading — to the UK, Denmark, Malaysia, Greece — because prohibition works beautifully as political signal even when it fails as protection. A government minister defended the ban by comparing it to drinking-age laws that “don't achieve 100% compliance but remain worthwhile.” Perhaps. But a drinking age does not purport to prepare anyone for a lifetime of drinking. Lockout, even where it worked, would teach nothing. A sixteen-year-old released into the full algorithmic environment with zero supervised experience is not protected; she is unprepared.
The second answer is surveillance. An industry of parental-control and monitoring apps promises safety through visibility: read the messages, track the location, flag the keywords. Here the research is less contested and more damning. Studies from the University of Central Florida and elsewhere have found that surveillance-style parental-control apps are often counterproductive — teens experience them as privacy invasion, the monitoring corrodes the parent-child trust the family actually runs on, and restriction teaches avoidance rather than judgment (UCF). The academic field studying adolescent online safety has been explicitly moving “beyond parental control” toward autonomy-supportive, trust-based approaches (Journal of Adolescent Research; arXiv), and voices in the youth-safety movement itself now argue that protecting children requires respecting their agency (TechPolicy.Press). Surveillance does not teach either. It teaches teenagers that trust is conditional and privacy is something adults take away.
Lockout and surveillance share a deeper flaw: both treat the child as an object to be managed rather than a person to be developed. Neither produces a 25-year-old who knows how to live well online. And both leave untouched the actual asymmetry of the digital age — that the platforms know everything about the family, and the family knows nothing about what the platforms know.
What the machines already believe about you
Request your data from any major platform and you will find, buried in the export, the platform's working model of you: the advertising classifications, the inferred interests, the behavioral segments. In my own export, Meta held two dozen labels — that I am a frequent international traveler, that I live away from family, that my household sits in a certain income bracket, that my device had recently changed and I might therefore be persuadable. These are not facts about me. They are commercial guesses, some accurate, some outdated, some simply wrong — and they quietly shape what my family and I see every day. Until recently, no member of the public had any practical way to see this model of themselves, let alone talk back to it.
That asymmetry is about to compound. AI companions were named one of MIT Technology Review's breakthrough technologies of 2026; 72 percent of American teenagers have already used AI for companionship (MIT Technology Review). The engagement algorithm guessed what your child would watch. The companion AI converses with your child at midnight about loneliness. Families have filed lawsuits alleging these systems contributed to teenage suicides (Morgan & Morgan); regulators from California to Brussels are scrambling (Future of Privacy Forum); and the sharpest parenting question of 2026 has become, literally, “Who is your teen talking to?” (Psychology Today). We are past the era when this was about screen time. Machines now participate in human emotional development. That is the revolution, and it is untested.
The third way: mentorship
There is a third answer, older than either of the failing ones. When something in the world is powerful, permanent, and dangerous to the unprepared — the ocean, the road, other people — families do not respond by pretending it doesn't exist or by watching their children through one-way glass. They teach. They swim alongside. They hand over the keys gradually, in daylight, with conversation.
What would teaching look like as technology? I believe it has three properties. First, it shows the family the mirror: here is what the algorithms actually believe about you, in their own words, translated into meaning — not to frighten anyone, but because you cannot develop judgment about a system you cannot see. Second, it lets people talk back: confirm what's true, reject what's wrong, notice what feels off — so that the family's own account of itself, owned by the family and no one else, grows stronger than the platforms' account of them. Third, it treats the child as the point: insight flows to the young person first, privately, with a chance to self-correct before anything reaches a parent — because the goal is not compliance today but agency at twenty-five. Patterns, never content. Trust as the product, not the casualty.
This is what we are building with MyChoice, and I want to be precise about what it is not. It is not a weapon against Meta or OpenAI or anyone else; the platforms built something revolutionary, and the honest position is that we are all — companies, parents, regulators, children — still learning what it does to us. If Zuckerberg makes good on genuinely private personal AI, the world improves. But a promise of privacy from inside an advertising business model is a hard thing to bank a childhood on, and families should not have to wait to find out. The mentorship layer has to be owned by the family, structurally incapable of serving anyone else, and honest to its bones — down to telling you plainly what it knows, how it knows it, and how to delete it.
Looking in the light
The children entering adolescence this year will be mid-career in 2050, in an information environment we cannot currently imagine. We will not get them ready by locking the door, and we will not get them ready by reading their mail. We will get them ready the way humans have always gotten each other ready for powerful things: by looking at it together, in the light, and learning.