CodeIt guide
Your kid used AI to write their code. Now what?
AI wrote your kid's project. The learning isn't lost, but it hasn't happened yet. What to ask, what to try this week, and both sides of vibe coding.
Short answer: the learning is not lost, but it has not happened yet either. It happens in what your child does next, changing the code, watching it break, and working out why, and you can start that this week with three questions, on any platform, without knowing how to code yourself.
In practice the tool is usually ChatGPT, Claude or whatever their friends are using, and which one matters far less than what happens after the code arrives. This page covers why kids reach for AI (it is rational, not lazy), what is genuinely lost when code arrives finished, why banning it fails, and where the "should kids learn vibe coding" argument actually stands. We build a product in this space and say so plainly near the end.
Why kids reach for AI
Do the arithmetic from a 10-year-old's side. They want a working snake game. Route one is six hours of typing, punctuation errors and frustration. Route two is thirty seconds. No child picks route one on principle.
It is also not new. Kids have always optimised for the artifact rather than the process, copying tutorial code without reading it, remixing Scratch projects and swapping the sprites, dropping free models into a Roblox game. AI is that instinct with a much better tool attached.
Two things follow. "Lazy" is the wrong diagnosis, and an expensive one: a kid told the shortcut is laziness stops telling you they used it, and you lose the conversation, which is the only tool that works here. And "make the thing" and "learn to make the thing" are different goals, only the first is your child's. The job is not to talk them out of it, but to attach the learning to it.
What is actually lost when the code arrives finished
Not the typing. Typing is transcription, and nobody ever learned to program by being good at transcription.
What is lost is the debugging loop, which is where learning actually lives:
predict → run → be surprised → form a theory → change one thing → test it
That loop is the whole mechanism. It is how a kid builds a working mental model of what a computer does, and it only runs when something behaves unexpectedly. Code that arrives finished skips every step.
Reading finished code feels like learning and mostly is not, the fluency illusion. You recognise the material, so you rate your understanding highly, right up until you have to produce it. Watching someone play piano is not practice.
| What AI takes over | Does it matter? |
|---|---|
| Typing the characters | Barely. That was never the skill. |
| Remembering exact syntax | A little. Professionals look it up too. |
| Choosing what to build | Yes. Choosing is where the interest lives. |
| Breaking a problem into steps | Yes, a lot. This is the part that transfers to maths, writing and everything else. |
| The moment it does not work | Most of all. The surprise is the learning signal. |
| Staying with a stuck problem | Most of all. This trait predicts who keeps going. |
Why "just ban it" fails
- It is unenforceable. There is a phone, a friend's laptop, and school-issued tools.
- It moves the behaviour underground. What you lose is not the AI use. It is the conversation, the only thing that actually changes how they use it.
- It is already inside the software. Search, documentation, editors, and the learning platforms themselves. CodeAI, Code.org, which rebranded in June 2026, moved toward AI literacy rather than away from it. The industry will not help you hold this line.
- Directing an AI is a real skill. It will be assumed of them the way search was assumed of you. A ban practises the wrong thing.
A ban converts a teachable moment into a compliance problem, and those you lose slowly.
One real exception. If the work is graded and the teacher's rule is no AI, that is an honesty question, not a pedagogy one. Handle it separately, and do not let it contaminate how you talk about AI in their own projects.
Vending machine or starting draft?
This is the distinction that matters, and it has nothing to do with whether AI was used.
| AI as a vending machine | AI as a starting draft |
|---|---|
| Prompt, paste, done | Prompt, read, modify, break, extend |
| Cannot say what any line does | Can point at the line that does the thing |
| Bugs get fixed by prompting again | Bugs get fixed by changing code |
| The project is finished on arrival | Arrival is the beginning |
| The work belongs to the model | The work belongs to the kid |
The one-question test. Ask for a change nobody asked the AI for: "make the ball speed up each bounce," or "make the button turn green when you win." If the answer is a new prompt, you are in vending machine mode. If they open the code and start hunting, you are in draft mode.
Which mode a kid ends up in is mostly a property of the tool, not the child. Plenty of AI tools hand back a locked result. A file you cannot see into, or a chat answer that must be copied somewhere else before it will run. Whether a kid modifies the output is largely decided by whether modifying it is easy.
Four things to try this week
Each takes about ten minutes and works with any tool, including a plain chatbot. You do not need to understand the code.
1. Change one thing, but predict first. Point at a number. "What happens if we make this 5 instead of 20?" Make them answer out loud before running it. The prediction is the exercise; running it is just marking.
2. Break it on purpose. Save first, then delete a line and ask what will stop working. Then restore it. Kids who have never deliberately broken code are afraid of it, and fear is what stops them experimenting.
3. Ask about one line, but ask the better question. Not "what does this do," which they answer by reading it back to you. Ask "what would go wrong if this line wasn't there?" That needs a mental model, not recall.
4. Add one thing the AI did not. A scoreboard, a second colour, a sound, a restart button. This is the real assessment. If they can add a feature, they understood it. If not, they did not, and now you both know, without an argument.
Bonus: keep the prompt. Have them write down what they asked for before reading what came back, then compare. What did the AI add, skip, or quietly change? That is specification practice, the half of the skill most likely to still matter in ten years.
A short script for the conversation
Do not open with "did you write this yourself?" It is an accusation, and it buys you either a lie or a fight.
You: That looks good. Did you use AI to build it? Them: Yeah. You: Makes sense, it is faster. Show me the part you would change first. You: What do you think breaks if you delete that line? Them: I don't know. You: Want to find out? Save it, then delete it and see. You: Now add one thing it doesn't do yet. Then it is actually yours.
If they get defensive: "I am not saying you cheated. I am asking whether you can change it, because that is the part worth anything."
What this communicates without lecturing: using AI is fine, and not being able to change what it made is the problem. That framing works because it is true, and because it gives them somewhere to go.
Should kids learn "vibe coding"? Both sides
The term came into wide use in 2025. It means roughly: describe what you want in ordinary language, accept the generated code largely without reading it, and iterate by prompting rather than editing.
| The case for | The case against |
|---|---|
| It is increasingly how software actually gets built; pretending otherwise trains kids for a world that is leaving | You cannot evaluate output you cannot read, and evaluation is the skill everyone agrees survives |
| The durable skills shift to specification and judgment, saying precisely what you want, and telling whether you got it | Debugging generated code needs more understanding, not less; when it breaks, the kid is stranded in code they never wrote |
| It keeps motivation alive, and motivation is the scarcest resource in kids' coding, most quitting happens before any concept lands | The fluency illusion is strong; kids badly over-rate what they know, and the correction arrives late and hard |
| It lowers the barrier for kids who would never have survived setup and syntax, including kids with no parent who codes | Models produce plausible, subtly wrong code, and a beginner has no way to detect it. Confidence without accuracy is worse than neither |
| Professionals already work this way, and pretending they do not is dishonest with older teens | The workflow is a moving target. Betting a child's decade of learning on the 2026 way of prompting is a genuine gamble |
Where the argument actually is. Both sides are mostly arguing about sequence, not permission. Very few serious people think a 10-year-old should never touch AI, and very few think reading code is now optional. The disagreement is over how much foundation to install first and how much can be picked up alongside.
That is not settled, and anyone selling you certainty about it is selling. The part both camps agree on, and therefore the part worth practising this year: a kid who can modify generated code is better off than one who cannot, whichever side turns out to be right.
What this looks like at different ages
| Age | What AI use tends to look like | Your job |
|---|---|---|
| 5–8 | Asks for a whole finished thing; cannot read most of the code | Sit with them. Treat the output like a picture book you read together. Nothing more is expected. |
| 9–12 | Generates projects happily, then gets stuck the moment they want a change | The key window. Each session: one prediction, one deliberate break, one added feature. |
| 13–15 | Uses AI for homework and personal projects, and may not mention it | Shift from supervision to conversation. The extension test is the whole assessment. |
| 16–18 | Works something like a junior developer | Talk about honesty policies and about reading before shipping. They should be able to explain any code they hand in. |
What to look for in any tool
Ask these of any tool, including ours:
- Does the generated code stay visible and editable, or is it a locked result?
- Can the kid run a broken or half-finished version and see what happens?
- Is there save or version history, so breaking things on purpose is safe?
- Can the kid make a small change without starting a new prompt?
Yes to those keeps the debugging loop intact. No has replaced the loop with a delivery service.
Disclosure: what we built, and who it is not for
We make CodeIt (codeitlearn.com). Treat this as an ad.
CodeIt is a browser-based studio where kids build websites, games and quizzes by describing them, then change them by moving things and picking colours, and can open a tab that shows what the project is made of and links to a lesson for each idea in it. The AI-assisted project generation is built so the result stays editable instead of becoming a locked AI result. The specific mechanic this page is about. The loop is: make something, play it, change it, see what it is made of, and publish it. Built for ages 8 to 14, with parent-managed, email-verified private profiles for ages 5 to 12 and independent accounts from 13. The free plan does not expire, needs no card, and includes ten assisted project builds a month with two learner profiles under a parent view. A paid family plan is available at CA$12 a month and can be cancelled at any time.
Things we will not blur:
- Our structured lesson sequence is Python only: thirty-one beginner lessons, starting with print, then variables, strings, conditionals, loops, lists and functions. The web side is learn-by-editing-your-own-project. There is no web curriculum.
- Under-13 managed profiles cannot publish publicly. Sharing happens inside the parent-managed account.
- Ages 5 to 12 require a parent-managed profile. This is not a set-and-forget product for young children.
CodeIt is not for you if: your household rule is no AI at all, we generate code, so we are the wrong product; your kid wants Minecraft or Roblox mods (Tynker); your kid mainly wants to keep making games (CodeCombat); your teen wants a professional AI development workflow (they should be using real developer tools); you want a sequenced web-development course; or you are buying for a school, because we have no school, district, rostering or teacher-dashboard product.
Common questions
Is using AI to write code cheating? For a personal project, no, it is how the tools work now. For graded schoolwork where the teacher said not to, yes, and that is an honesty conversation rather than a technology one.
My child cannot explain any of the code. Should I worry? Not immediately. Ask them to add one small feature the AI did not build. If they can, the understanding is ahead of the explaining. If they cannot, you have found the real gap, and the four exercises above are the fix.
Will AI stop my kid from ever learning to code properly? Only if the loop never runs. Kids who only prompt and paste do not build a mental model. Kids who prompt and then modify are learning the way people have always learned from worked examples.
What age should a kid start using AI coding tools? There is no verified research answer. Practically: under about 9, use it with them rather than handing it over. From 9 to 12 it works well as a starting draft with an adult nearby. From 13 they will use it whether or not you approve, so a habit of modifying beats a rule about access.
Should I make them write code by hand first? Some fundamentals first helps, and a fully AI-first start leaves gaps. But holding a motivated kid at a blank file for months is the most reliable way to end their interest. Most families do better mixing the two than sequencing them strictly.
Last verified: 21 August 2026. Platform names, pricing and terms were checked on this date and change often, confirm on the vendor's own site before paying.
Questions parents ask first
Is using AI to write code cheating?
For a personal project, no, it is how the tools work now. For graded schoolwork where the teacher said not to, yes, and that is an honesty conversation rather than a technology one.
My child cannot explain any of the code. Should I worry?
Not immediately. Ask them to add one small feature the AI did not build. If they can, the understanding is ahead of the explaining. If they cannot, you have found the real gap, and the four exercises above are the fix.
Will AI stop my kid from ever learning to code properly?
Only if the loop never runs. Kids who only prompt and paste do not build a mental model. Kids who prompt and then modify are learning the way people have always learned from worked examples.
What age should a kid start using AI coding tools?
There is no verified research answer. Practically: under about 9, use it with them rather than handing it over. From 9 to 12 it works well as a starting draft with an adult nearby. From 13 they will use it whether or not you approve, so a habit of modifying beats a rule about access.
Should I make them write code by hand first?
Some fundamentals first helps, and a fully AI-first start leaves gaps. But holding a motivated kid at a blank file for months is the most reliable way to end their interest. Most families do better mixing the two than sequencing them strictly.
What you can do on CodeIt
Last verified 2026-08-21.