I went looking for a rule about children and AI. What I found was a harder question.
My kids are four and two. I use AI, I want them to understand it, and I have already made a mistake trying to bring it into their lives. This isn’t a parenting formula. It is the boundary I am trying to draw at home.
The question changed
I started with access: when should a child be allowed to use AI, and with how much supervision?
Access matters, but it isn’t the whole decision. AI can do more than show a child something. It can write, explain, choose, encourage, and finish. It can remove tedious work, or remove the practice through which a skill is formed.
So the question I am taking home is:
Has my child learned this skill yet, or would AI be doing the learning for them?
That gives me two lists for one real task:
- What they have already learned and could use help doing faster.
- What they are still learning and need a chance to work through themselves.
This is more useful to me than trying to settle whether AI is simply good or bad for children. It also leaves the uncomfortable part visible: what will actually keep the second list from being handed over too?
Being the interface wasn’t enough
In an earlier episode, I talked about making an AI song about my son’s day. He never touched the tool. I supplied the idea, ran the prompt, and handed him a finished song.
At the time, being the parent in the middle felt responsible. But I had protected his access to the tool without preserving any of the making. He didn’t choose the words, struggle with a first version, or help decide what the song should become.
The result was lovely. The missing middle was the problem.
That experience changed the line for me. Supervision can help with what a child sees. It does not automatically protect the practice, frustration, judgment, and pride that may matter more than the finished object.
The examples didn’t give me a simple answer
I looked at school AI guidance in China, a young student’s reported business experiment, and a private school near me that places AI inside a very different school day. I expected the examples to tell me who was being reckless and who had found the answer. They didn’t.
They did make the tradeoff clearer. A child can learn to use a powerful tool while still needing time to form the skills, judgment, and sense of self that tell them when not to hand something over.
The unresolved example for me is companionship. An always-available AI can be patient, interested, and difficult for a tired parent to compete with. I don’t yet know where the right boundary is. Pretending I do would be less useful than naming the question.
Try the boundary on one task
The two-list worksheet turns this question into one small family experiment. Pick a real task, separate what has already been learned from what is still being learned, and write down what AI may and may not do this time.
For the broader version—not only for children—the journey worksheet helps separate the result you want, the experience you want to keep, and the friction you would gladly remove.
Corrections and source notes
- The national Chinese guidance and Beijing’s local teaching requirement are separate. The eight-class-period minimum is documented for Beijing from autumn 2025; I should not have described it as a nationwide eight-hour rule. See the national guidance in translation and Beijing policy.
- Alpha Santa Monica currently describes a K–9 program with $65,000 tuition, not K–8. Its performance statements are the school’s own claims.
- In the reporting about Peng Gaodi, 30,000 yuan is the school program’s funding ceiling. It is not evidence that he received that amount or earned it as business revenue; the ByteDance recognition was a token grant.
- The writing story comes from Phaedrus 274c–275b: Thamus speaks inside a story told by Socrates. Translation choices vary.
The earlier companion, five things I regret using AI for, has the full story behind the song and the question that led here.