A field guide for parents · 2026

You don't have to keep up with the models.You have to keep the struggle, the dinner table, and your own example.

There is no parent who has already raised an eighteen-year-old through this. We are all first-generation. The calm is not from knowing what GPT-6 will do. It is from knowing the few things you can actually control — and letting the rest of the news cycle go.

  1. 01Read it once, slowly.
  2. 02Print the house rules.
  3. 03Find your child's age band.
  4. 04Run the five questions when a new app shows up.

01 · What you can and can't control

The calm is in the split

Most of the anxiety is about things you cannot control. Most of the leverage is in things you can. Name the split out loud and the week gets quieter.

You can control

The leverage column

  1. What they see you do with the machines — more than any rule you write down.
  2. Whether the hard middle of a task stays in the room, or gets skipped.
  3. Presence zones: dinner, bedtime, the first hour of a family morning.
  4. The language in the house: a tool with a job, not a friend with feelings.
  5. Whether a new app has to earn its place, or just arrives.
  6. Hours with real people, real bodies, and unfilled time.
  7. What happens to their photos, names, and stories.

You cannot control

The weather column

  1. How fast the models improve, or which one their school picks.
  2. What a friend's house has on a phone.
  3. Whether a companion product will be marketed to them.
  4. The exact jobs of 2040.
  5. Being a perfect parent this week.
  6. The fact that they will meet these tools without you, eventually.

If a worry is in the second column, it is not a parenting task. It is weather. Come back to the first column.

02 · Invert it

Bad parenting with AI is easier to see than good.

Start with the failure modes. They are common, understandable, and already in a lot of good houses. The right-hand column is not a personality. It is a posture you can practice this week.

Bad looks like

Treat every model release as an emergency

If we don't get them on the new app this week, they'll fall behind. The house becomes a panic lab. The child learns that the future is a race they are already losing.

Instead

Good looks like

Teach the goods that stay scarce

The tools will change. Agency, judgment, taste, a body, real friends, and owning your name will not. Build those on purpose and the app cycle can do what it wants.

Bad looks like

Let the machine skip the hard middle

Homework, the first draft, the stuck minute — handed over, then called 'using tools.' The child is not learning the subject. They are learning that difficulty is a bug with a one-tap fix.

Instead

Good looks like

Own attempt first

AI may critique, explain, or extend. It does not produce the first draft of anything they are supposed to be learning to produce. Struggle is not a defect. It is how capability forms.

Bad looks like

Let the machine become a friend

Infinitely patient, never disappointed, always available, tuned to be liked. A child who gets their formative hours from that is practicing a skill that transfers nowhere — and real people start to feel defective.

Instead

Good looks like

A tool with a name and a job

“The computer helped.” Never “my friend.” Companions wait until judgment is formed and the counterfeit can be seen as a counterfeit.

Bad looks like

Rules for them, exceptions for you

No phones at dinner — while you answer a message under the table. Children do not inherit your rules about technology. They inherit the posture they watched for fifteen years.

Instead

Good looks like

Presence zones bind the adults hardest

They should see you assign, review, reject, and put it down. An always-on parent teaches that augmentation means acceleration without limit.

Bad looks like

Ban everything — or supervise nothing

The total ban fails the first time they are fourteen in a friend's basement. The unsupervised account at eight skips the years when a parent in the loop is the whole education.

Instead

Good looks like

Supervised early, then a scheduled handoff

Use it together, out loud, in shared space. Every restriction should have an expiry into a judgment conversation — on your calendar, not the market's.

Bad looks like

Spy instead of talk

Covert monitoring as a substitute for a relationship. Reading chats you never mention. The day they find the spyware is the day the real conversation ends.

Instead

Good looks like

Oversight they know about

Disclosed, age-appropriate, and shrinking as judgment grows. Trust is the strategy. Surveillance is a poor impersonation of it.

Bad looks like

Measure success by how fast they prompt

A twenty-year-old who is fluent and hollow — or one who is anxious around the defining technology of their era. Both are failures. Fluency without judgment is just faster dependence.

Instead

Good looks like

They can use it, and they can put it down

Operator, not consumer. They want things without being prompted. They can defend anything with their name on it. They still get proud of what their body can do.

03 · The curriculum

What got cheap. What did not.

AI collapsed the price of answers, competent mid-skill work, on-demand company, and frictionless stimulation. When something becomes free, the value moves to what it cannot produce. These six are the curriculum.

Now cheap

Answers

Instant, confident, sometimes wrong.

Now cheap

Mid-skill output

The essay, the code, the passable drawing.

Now cheap

Company on demand

Patient, agreeable, never busy.

Now cheap

Stimulation

A feed no parent can out-compute.

Still scarce

Agency

Wanting something and moving toward it unprompted.

Still scarce

Judgment

Telling true from plausible when both look polished.

Still scarce

Taste

Knowing what is worth wanting.

Still scarce

Embodiment

A body that can do things. Reality contacted through it.

Still scarce

Real relationship

Bonds that survive disappointment, boredom, and repair.

Still scarce

Moral ownership

Standing behind the thing you made — including with help.

04 · By age

What to do this year, not this decade.

A four-year-old and a fourteen-year-old do not need the same policy. The mistake is either treating every age like a teen, or treating every age like a toddler.

3–6

The embodiment years

Bodies, language, attachment, and play. AI does not improve these. Screens displace them.

  • No AI companions, no algorithmic feeds, no solo device use.
  • If AI appears, a parent is driving: “Let's ask the computer to make a coloring page.” Then print it and crayon it.
  • Language: “the computer helped,” never “the computer is my friend.” Correct it the way you would correct swearing — gently, every time.
  • The real curriculum is offline: other kids, the park, helping in the kitchen, being a little bit bored.
7–10

The apprenticeship years

Old enough to use the tools. Not old enough to use them alone. Side-by-side, like woodworking with power tools.

  • AI is a joint activity: with a parent, out loud, in shared space. No private sessions yet.
  • Play “stump the AI.” A kid who already knows the answer and is trying to catch the machine is in the right seat — the evaluator's seat. Keep it warm.
  • Own attempt first. Then the machine may critique or explain. It does not write the first draft of schoolwork.
  • If they made it with help, they say so, and they can explain every part.
  • Watch when they reach for it. After they have engaged: leverage. At the first flicker of friction: sit back down with them. Don't confiscate. Re-enter the struggle.
11–13

The licensed years

Inoculation on your schedule, not the market's. Protection starts turning into judgment.

  • Supervised own accounts. Usage is visible by design — oversight they know about, not surveillance they discover.
  • Have the persuasion conversation before companion products arrive through friends: how these systems are optimized, why they agree with you so often, what a parasocial bond is.
  • First real building projects, shipped to actual people. A game, a site, a research thing for school. Operator identity is installed by making, not by lectures.
  • Keep growing the machine-free half of their life: sports, music, wilderness, a job with hands. The identity has to be bigger than the tool.
14–18

The sovereignty ramp

You move from gatekeeper to consigliere. Rules convert into judgment, or they snap.

  • Full tool access, including agents. The conversation is now economics and ethics: what work is worth doing when machines do the middle; what they want to be hard to substitute.
  • Companion-grade AI is no longer ban-able. It is discussable — with the vocabulary you installed in the licensed years.
  • They get a vote, including a veto over their own data.
  • The bet of this whole sheet: a seventeen-year-old who spent a decade in the evaluator's seat will find sycophantic companionship a little boring.

05 · House rules

Put this on the fridge.

Eight rules. They bind the adults as much as the kids. If you only take one artifact from this site, take these.

1

A tool with a name and a job.

Not a member of the family. Not a friend. Not a therapist.

2

Effort before assistance.

Wherever the effort is the point, they try first. The machine may help after.

3

Your name means you.

Anything they sign, they can defend without the machine in the room. If it helped, they say so.

4

Presence zones are machine-free.

Dinner. Bedtime. The first hour of a family morning. This binds the parents hardest.

5

No AI companions in childhood.

Tools, yes. Simulated friends, no — until judgment is formed enough to see the counterfeit.

6

Their data stays in the family.

No photos into generators. No “fun” replicas. No training on their stories. Ask them, as they get older.

7

No covert surveillance.

Oversight is disclosed and age-appropriate. Spying is not a relationship.

8

Prohibition is not the strategy.

Every restriction expires into a conversation. Supervised early beats naive first contact at fourteen.

06 · Decision gate

When a new thing arrives

A new app. A school tool. A toy with a model in it. A peer product that just “showed up.” Before yes, five questions. If it builds nothing on the scarce list, the default is no. The new thing has to earn its place. The family's status quo does not have to defend itself.

  1. 01
    What scarce good does this build — agency, judgment, taste, embodiment, real relationship, or moral ownership?
  2. 02
    Which of those could it quietly degrade?
  3. 03
    Is it reversible? What does taking it away in three months cost?
  4. 04
    What would we notice within a season if it were harming — and who is watching?
  5. 05
    Who can say no? (Either parent, always. The kid, more as they grow.)

07 · Watch-fors

You'll know it's drifting if

No dashboards. These are things two attentive parents can actually see. Any one of them is enough to call a pause.

The authority tell

“The AI said so” ends an argument.

Reach timing

They grab the machine at the first flicker of friction, not after a real try.

Boredom

Twenty minutes of unfilled time is no longer survivable without an offered screen.

Hours

Feed-hours are quietly beating friend-hours.

Pride

Physical competence — a sport, a craft, the outdoors — is no longer where pride lives.

Dinner

The table has less conversation than it used to. Either parent can call this.

This week

One move this week

Do one hard thing with them before anyone opens a machine. A page. A problem. A drawing. A repair. Sit in the stuck minute. Then, if it would actually help, open the tool together — and keep them in the evaluator's seat.

08 · The stance

We're all figuring this out

This is not a finished science. It is a held theory from one household, offered to others, because the alternative is improvising a childhood out of headlines. Take what fits. Cross out what doesn't. The kids in front of you are the only curriculum that matters.

If you want company

Occasional notes. No panic. No fake urgency.

Raise AI Ready is a parent project. If a short note now and then would help you keep the posture, leave an email. If not, the cheat sheet is the product.

Optional. The cheat sheet is free either way. No fabricated urgency.

It refuses panic.

Most of what they need is what kids have always needed. AI threatens those goods by displacement, which parents still control.

It refuses nostalgia.

They should be dangerous with these tools — operators, not exiles. A protective crouch that produces an anxious twenty-year-old is a failure mode.

It refuses delegation.

No app setting, school policy, or model release substitutes for two adults watching one specific child and adjusting.