The Hearth Thesis
1An older species governing a newer one
Children today are not just quicker on a screen than their parents. In the digital world they are, genuinely, more capable. They learn a new tool by touching it. They see what a thing is for before an adult has finished reading its description. They carry none of the fear that makes an adult hesitate at an unfamiliar interface. And that world now contains the most powerful learning resource ever built: the accumulated knowledge of the species, searchable, and, since 2023, able to talk back.
Yet a ten-year-old’s access to that world is decided, avenue by avenue, by an adult who moves through it more slowly, understands it less, and has to research, vet and personally experience each avenue before allowing it. Exaggerating only a little: a newer, more capable species is being governed, in the one realm where it clearly outruns us, by an older one. If that arrangement were proposed for any other domain, we would call it absurd. In this one we call it parenting.
I don’t say this to blame parents. I am one. The point is the opposite: we have built a world in which the good outcome for a child depends on the parent being exceptional, and most parents, however loving, cannot be exceptional about everything. That is a design failure, not a parenting failure. It can be fixed by design.
2“Screen time” is a word for not looking
Watch what happens when a family talks about the digital world. The whole of it, the tutorials and the games, the making and the scrolling, the friend on a video call and the stranger in a comment thread, collapses into two words: screen time. It is a derogatory blanket term, and it is derogatory precisely because it is a blanket. It exists so that a parent who cannot tell a redstone tutorial from a brainrot loop can still say something about both. What they say is: less.
The blanket produces, mostly, two kinds of children.
The first is shut out. Their parents, unable to see past the blanket, keep the whole world at arm’s length. These children lose, entirely, the chance to learn at the speed the world now allows. They will meet the tools late, and meet them as consumers.
The second is swallowed. Their parents cannot hold a line, or cannot be present enough to hold one, and these children fall into the most addictive loops the internet has ever engineered, loops built by teams whose only metric is the next minute of attention.
There is a rare third kind: the child who finds the right loop. Who spends the hour building rather than scrolling, who comes out of a session knowing something. Look closely at that child and you will almost always find a parent who could afford attention. Who had the time, the literacy and the patience to look past the blanket and say yes to the right thing. That is selection bias, not virtue. We have made the good outcome depend on the parent being unusual.
Both of the common outcomes are wrong. The third should be the default. And it cannot require every parent to be super-conscious and deliberate, because that is not a plan, it is a wish.
We have made the good outcome depend on the parent being exceptional. That is a design failure, not a parenting failure.
3The parent is the bottleneck, and the child knows it
Here is the shape of a child’s exploration today. Everything goes through the parent. Before a child can try a new realm, the parent must first understand it, research it, ideally experience it, and then decide. The parent’s time is the rate limit. The parent’s biases are the filter. The parent’s literacy is the ceiling.
Which means the biggest blocker to a child’s learning is, structurally, the person who loves them most. A parent who never used a certain tool, who read one scary headline, who cannot see what a game teaches, will say no, and will not know what they said no to.
Children learn this early. They learn that bringing the cool thing to a parent gets it shot down. So they stop bringing it. They stop partnering, stop discussing, and start routing around. The parent ends up with less visibility, not more, and the child ends up exploring alone. The gate, meant to protect the relationship, is what quietly ends the conversation.
4Gates: the only tool a slow decider has
Every product built for this problem is a gate. Minutes per day. Lists of allowed apps. Age ratings. A dashboard of what the child did, delivered to the parent after the fact. Every child-facing product, in turn, builds its own gates, limits and blockers to stop the child from using it “too much.” Nowhere is there nuance, because nuance would require judgment, and judgment was the thing the parent could not afford.
I want to be fair to the gate. It is not a stupid design. It is the inevitable design given the actors we had: a decider who cannot keep up, and a child who can. A gate is what you build when you cannot look. The parental-control industry is not wrong about the danger; it is stuck with the wrong tool, and it has made a business of selling the tool with fear attached.
The gate’s deepest cost is not the lost hour. It is what it teaches. A gate says: the world is dangerous and you cannot be trusted in it. The child hears that every day, and at some point believes it, or resents it, or both.
5The fourth actor
Every family in the digital world has three actors. Parents and guardians, who hold the values and carry the responsibility. Mentors and teachers, who hold the skills and the goals. And children, who hold the curiosity and, increasingly, the capability. All three are stuck, because everything must pass through the actor with the least time and the least fluency.
Large language models make a fourth actor possible. I want to be precise about what it is, because the obvious reading is wrong.
It is not a chatbot the child talks to. It is not a control panel the parent operates. It is a household intelligence, and its defining property is that at every turn it speaks for whoever isn’t in the room.
When a parent is deciding, the fourth actor carries the child’s case. It does the homework the parent never had time for: what Scratch actually is, why this particular child wants it, how it sits alongside what this family already believes. It says the thing the child cannot yet articulate: “he wants to join a modding server; that’s a workshop, not a game, and it’s also where the unmoderated chat lives. You said making beats consuming and no strangers in his messages. There’s a way in that gives him the first without the second.” It makes the parent’s yes cheap, and the parent’s no informed.
When a child is deciding, the fourth actor carries the parent’s reasons. It turns “no” into “here’s why, and here’s how.” It knows that weekday evenings are for reading in this house, and it knows that because the parent said so once, in plain words, not because a rule was configured. It shows the child what is open, how to earn what isn’t, and how to make a case for an exception: “the no was to the chat, not to the modding. Show her the mod you built, then ask again.”
Its loyalty is to the relationship, not to either side. It keeps no secrets, because it holds no confidences; it holds what the family has agreed, and argues both sides of it. A child can respect that, and a parent can trust it, which is more than can be said for either a buddy or a nanny.
We used to have this. It was called a household. The well-off never learned to cook, and never had to inspect every visitor; they had people who understood the family’s ways and applied them a thousand times a day without asking. The fourth actor is the old family butler, back, and this time affordable by everyone.
The working name is Hearth. The domestication of fire was the invention of home: a household is, literally, the holding of a hearth, the thing a family gathers around and the thing that keeps the dark out without a wall. That is what the fourth actor is for.
6What it is not
Because the category it will be filed under is full of things it must never become, some negatives are worth writing down.
Not surveillance. Nothing about the child is gathered that the child cannot see. The fourth actor tells the child what it will tell the parent, and vice versa. Discretion, not secrecy.
Not a companion. It is not there to be the child’s friend, and it should never suggest it has feelings. The friend-shaped AI for children is the most dangerous product of this decade and the fourth actor is its opposite: it exists to make the human relationships in the house work better, not to substitute one.
Not a nanny. It does not decide. Parents set the destination. Children drive. The fourth actor makes both possible; it does not take the wheel.
Not an automated stick. It never punishes on a schedule. A consequence delivered by a parent with a reason teaches accountability; one delivered by a scheduler at two in the morning teaches that the world is arbitrary.
Not a seller of fear. It will never be marketed on danger. The pitch is what a child can reach, not what they can be kept from.
7Hearth as an operating system
I keep calling it an operating system, and I owe the reader a reason. An operating system is the thing many programs share so that none of them has to solve the hard problems again. A family in the digital world has a handful of hard problems that every product for children currently re-solves badly, alone:
- Who is in this household. Parents, co-parents, a grandmother who is allowed to approve things, a tutor, the children, the devices each of them uses. Consent recorded once, by a verified adult, as a fact and not a checkbox.
- What this family cares about, in its own words. Not a settings screen. A page a parent could write in ten minutes and a child could read. “We’re fine with anything that makes something. We don’t do shorts. Weekday evenings are for books.” The fourth actor turns that into the thousand small decisions, and shows its work.
- A shared memory of what happened and why. Every decision, with a reason and a face attached, visible to the person it is about. This is the trust system and the pedagogy at once, and it is how the fourth actor learns the family rather than guessing at it.
- The fourth actor itself. The reasoning that shapes the family’s terms with the parent, explains them to the child, and extends them when something new comes along.
- Rooms and seats. A child’s room and a parent’s room that never look alike, and a seat at the table for a mentor. Two dialects of one language.
Products live above that. Each one makes the kernel know the family a little better; the kernel makes each product trustworthy on the day it launches. A product built on it does not need its own consent flow, its own idea of who the parent is, its own notion of what “too much” means. That is what “operating system” buys.
8pigg: the first product, built before I could name the kernel
This year I built a small product for my own family and a few others. It is called Piggin’, and it lives at pigg.in. It is a points bank: children earn Gems for things the family values, save them toward things they want, and spend them on things that were priced. Points, never money. Parents set the prices and the trust levels once; children log their own days and make their own requests; parents grant at the moment of handing over the ice cream. There is no language model in it.
I did not set out to build a kernel. But looking back at the decisions I made, the framework was already there, unextracted:
- The child holds the wheel. A child logs a chore in one tap, without permission. A child names a wish and it becomes a goal. A child can post a chore for yesterday, and the trust model did not change, because the child never had less trust than “sight unseen” to begin with.
- The parent sets the mix, not the moves. Every economic constant is a family setting with a sane default. Parents tune how fun and work balance; they do not approve each step.
- Nothing is hidden from the person it is about. Every Gem movement is visible to the child, with a reason and a name attached. A parent’s adjustment says who and why. The record is glass.
- Never automate the stick. Consequences exist and are manual, with a reason. No scheduler ever charges a child.
- Price the things you are willing to refuse. Screen time gets a price because at zero Gems a parent would honestly say “not today.” The park does not, because the child is going regardless, and a price you won’t enforce is a fake price.
And here is what the live families showed. Given the wheel inside a well-shaped world, children steer well. My two worked out on their own that a fifty-Gem TV show “wasn’t worth it” and put the Gems toward something they wanted more. Nobody lectured them about opportunity cost; the world was shaped so that the lesson was visible, and they read it. A parent’s role in that moment was to have set a fair price weeks earlier and then to get out of the way.
That is the fourth actor’s shape, without its brain. Every one of pigg’s decisions is a thing the fourth actor would have argued for. What pigg cannot yet do is the reasoning: warn me that fifty is too high before my daughter finds out; explain to my son why a thing costs what it costs in words he accepts; take his request for a game I have never heard of and come back with what it is, how it fits us, and a suggested price. Adding the brain is the work.
9Why now
Every cycle ends the same way, because nobody inside the loop knows the family. Judgment that belongs to the household is what breaks it.
First, a pattern that is by now older than the children living inside it. A world opens. Roblox teaches building, physics, economies and even code to 123 million people a day, over 40% of them under thirteen; by the company’s own account, most American nine-to-twelve-year-olds play it. Then the loop takes over: strangers, chat, the attention machinery, and by 2026 a hundred and forty-eight consolidated federal lawsuits and half a dozen state attorneys general. And the answer, every time, is a gate. Roblox’s answer this year is a face scan before a child can talk to anyone. YouTube’s answer to Shorts was YouTube Kids. Every cycle ends at the gate, because nobody inside the loop knows the family. A fourth actor is what breaks the cycle: judgment that belongs to the household, not to the platform.
Three further things are true in 2026 that were not true three years ago.
Children are already there. Common Sense Media’s 2026 census found that 86% of American 9-to-17-year-olds use AI, and nearly a quarter use it every day. This is not a future to prepare for. The most capable generation has already adopted the most capable tool.
The tool is gated at thirteen. ChatGPT permits no account under 13 and requires parental permission to 17; in August 2026 it launched a separate teen mode with parental controls and age prediction. The frontier labs are, reasonably, building gates, because a gate is what you build when you cannot know the family. Below thirteen the answer is simply no. A nine-year-old in the world’s most powerful classroom is a trespasser.
Regulators are forcing a family layer to exist. The amended U.S. COPPA rule took full effect in April 2026. The UK’s Online Safety Act children’s codes have been enforceable since July 2025. Texas’s App Store Accountability Act is in force and Utah’s follows in 2027, both requiring app stores to know a user’s age and route minors’ consent through a parent. India’s DPDP Act sets the bar at eighteen. Every platform will soon know whether a user is a child and who their parent is. The only open question is whether the family layer that sits on top of that fact is a gate, or something with judgment.
And underneath all three: language models can now hold what a family cares about in plain language, and reason about a situation nobody wrote a rule for. That is the capability the gate never had. It is the reason the fourth actor is possible now and was not possible before.
10Products that could stand on the kernel
These are sketches, not a roadmap. They exist to show that “operating system” is a real claim and not a mood.
An economy of effort and reward. pigg, as it exists, with the fourth actor added: prices argued before they are set, reasons given in the child’s language, requests that arrive with the homework done.
Asking for a new world. A child wants to try something the family has never considered: a language, a game, a tool, a community. Today that request dies at the parent’s research cost. Here it arrives at the parent with what the thing is, what it teaches, what it risks, and how it fits what this family already said it cares about. The parent’s job shrinks to a decision. The child’s job grows to making a case.
A guided way into AI. The nine-year-old trespasser gets a door. An assistant that operates inside the family’s own terms, opens the world rather than fencing it, and shows the parent everything without asking the parent to read everything. This is the product that answers the question “how do we not deprive them,” and it is the one no lab will build, because no lab knows the family.
A seat for mentors. A tutor, a coach, a grandmother with standing. Goals they set flow into the family’s terms; progress flows back; the parent stops being the relay between the people helping their child.
The fabric. As platforms are made to expose a child’s age and a parent’s consent, the kernel becomes the thing that answers on the family’s behalf, with judgment, across the products the child actually uses. This is the furthest sketch and the one that would justify the word “infrastructure.”
11What has to be true
I would rather write the hard parts down than have a reader find them.
Loyalty has to be designed, not promised. “Speaks for whoever isn’t in the room” is a contract with edges: what is said to whom, what is never inferred, what the child is always shown. Getting this wrong produces a nicer-sounding surveillance product. It is the central design problem and deserves to be treated as one.
Parents will not write essays. The family’s terms have to be drawn out of a parent in ten minutes, from choices and examples, the way pigg’s starter plan draws out prices, and then kept in prose the parent can read back and edit.
Children will test it. Of course they will; that is what capable people do to systems. The kernel’s answer is the one pigg already gives: bounds, not surveillance. A price you can’t lie your way past. A cap that holds regardless of the story.
The under-13 question is real. An AI that reasons about a child under thirteen sits under COPPA, DPDP, and the children’s codes, and under the shadow of the companion-AI lawsuits. The posture that makes this survivable is the one pigg already has: the parent is the controller, the child’s data is minimal and visible, nothing leaves the household to a third party, and the model acts on the family’s instructions, not its own.
It must not quantify childhood. pigg learned to price only the things a parent would honestly refuse. The fourth actor must learn the same restraint at a larger scale: not every hour needs a value, and a child’s love of the park is not a transaction.
The gate will tempt us. Every incentive in the category pulls toward selling parents control and fear. The test of the fourth actor is whether it makes more of the world reachable, measured in things a child got to do that they would otherwise have been refused.
12The principles
- The child drives. The parent sets the destination.
- The fourth actor speaks for whoever isn’t in the room, and argues both sides.
- Nothing about a person is hidden from that person.
- Never automate a consequence.
- Never sell fear. Measure the world opened, not the world closed.
- Ask the parent for values, never for vigilance.
- Build it so a parent who is not exceptional gets the outcome that used to require one.
13Closing
We will not get a more capable generation by governing it with a slower one. We will get it by giving every family the one thing the fortunate have always had: a household that runs on their values, faster than they can. The children are already in the room. The question is whether the adults join them, or keep standing at the gate.
The argument in nine cards
The third one has a parent who could afford attention. We made the good outcome depend on the parent being exceptional.
To the parent, the child's case, homework done. To the child, the parent's reasons, and a way through. Loyal to the relationship, not to a side.
Sources for §9
- Roblox, Q2 2026 results (123M daily active users); under-13 share and the “most 9-to-12-year-olds” figure are Roblox’s own reported numbers as cited in County of Los Angeles v. Roblox, February 2026. https://lacounty.gov/2026/02/19/la-county-sues-roblox-for-unfair-and-deceptive-business-practices-that-endanger-and-exploit-children/
- Roblox child-safety multidistrict litigation, MDL 3166 (148 consolidated federal cases as of August 2026). https://mdlupdate.com/mdl/3166-roblox-child-safety/
- Common Sense Media, The Common Sense Census: AI Use by Tweens and Teens (June 2026): 86% of 9–17s use AI; nearly a quarter daily. https://www.commonsensemedia.org/press-releases/common-sense-media-releases-inaugural-annual-study-on-ai-use-by-tweens-and-teens
- Common Sense Media, Nearly 3 in 4 Teens Have Used AI Companions (July 2025). https://www.commonsensemedia.org/press-releases/nearly-3-in-4-teens-have-used-ai-companions-new-national-survey-finds
- OpenAI, Introducing ChatGPT for Teens (August 2026) and Our approach to age prediction. https://openai.com/index/chatgpt-for-teens/ · https://openai.com/index/our-approach-to-age-prediction/
- FTC, amended COPPA Rule, Federal Register, 22 April 2025; compliance date 22 April 2026. https://www.federalregister.gov/documents/2025/04/22/2025-05904/childrens-online-privacy-protection-rule
- Ofcom, Age checks to protect children online, Protection of Children Codes enforceable from 25 July 2025. https://www.ofcom.org.uk/online-safety/protecting-children/age-checks-to-protect-children-online
- Future of Privacy Forum, App Store Accountability Act comparison chart (TX/UT/LA), June 2026. https://fpf.org/wp-content/uploads/2026/06/FPF-Legislation-TX-UT-LA-App-Store-Accountability-Act-Comparison-Chart.pdf