AI writes code now. Should your child still learn to code?

On this page

1. The belief behind the question
2. What happened when experts used AI
3. The child who never learned to check
4. What coding teaches once AI types the code
5. Run the test yourself
6. Where these studies fall short

This is one of the first questions parents ask me. Some ask it out of worry. A few ask it to test me before they pay: if a chatbot builds a game in ten seconds, why would a ten-year-old spend a term learning to build one?

It deserves a straight answer, with evidence you check yourself.

The belief behind the question

The belief goes like this. Coding is typing instructions into a computer. Machines now type the instructions. So coding goes the way of typewriter repair, and children should spend their time on something with a future.

Every step sounds reasonable. Then you look at what happens when people who already code use these tools.

What happened when experts used AI

In early 2025 a research group called METR ran a randomised trial. Sixteen experienced open-source programmers worked on 246 real tasks from their own projects. A coin decided which tasks allowed AI tools and which didn’t.

With AI, the programmers took 19 percent longer.

The second finding matters more for your child. Before starting, the programmers expected AI to make them 24 percent faster. After finishing, slower on the clock, they still believed AI had sped them up by about 20 percent. They were wrong about their own work, and they were experts.

A separate team at New York University tested GitHub Copilot, one of the best-known coding assistants, on security tasks. About 40 percent of the time, the code it wrote had vulnerabilities, holes an attacker would use.

Put the two studies side by side. AI produces code fast. The code is often wrong in ways a beginner won’t spot, and even skilled people misjudge how much the tool helps them.

Somebody has to decide whether the machine’s answer is right. The person who decides needs to understand the code.

The child who never learned to check

Now picture your child at 22, at a first job, with an AI tool producing pages of code a minute. Two kinds of young people sit at those desks.

The first kind built programs as children. They broke them, read the error messages, guessed why, tested the guess and fixed it, hundreds of times. When the AI hands them code, they read it, run it, find the fault and ask a sharper question.

The second kind learned to prompt. They copy whatever comes back. When it fails, they ask again and hope.

My bet: employers pay the first kind to supervise the machine, and the second kind ends up competing with the machine on price.

This goes beyond jobs. Your child already lives inside software. The feed decides what they watch next. The game decides when to show a purchase button. A child who has written even a small program knows somebody wrote those rules and chose them. A child who never has thinks the phone simply knows them.

What coding teaches once AI types the code

Strip away the typing and look at what a child practises in a coding class.

• Breaking a big goal into small steps a machine follows
• Guessing why something broke, testing the guess, fixing it
• Reading an error message instead of panicking at it
• Explaining their reasoning out loud to someone else
• Sticking with a problem for 20 minutes when the first three ideas fail

None of these go away when AI writes code. Each one becomes the skill needed to use AI well. The Raspberry Pi Foundation, the charity behind the Code Club network, argues in its 2025 position paper for the same reason: young people still need to read, change, test and explain code, because people take responsibility for what the tools produce.

Kenya agrees, at least on paper. In 2022 the Kenya Institute of Curriculum Development approved a coding curriculum for primary and secondary schools.

Run the test yourself

Try an experiment this weekend. Open any free AI chatbot and type: “Write a simple game in Scratch where a cat chases a mouse.” Show your child the answer. Then ask your child to make the cat move faster.

Watch what happens. Most children who have never coded stare at the answer, then ask the chatbot again. A child who has built a Scratch game looks for the number which controls the speed.

The gap you see in those five minutes is the gap this whole question rests on.

Where these studies fall short

The METR study used 16 programmers, a small group, on mature codebases. AI tools improve every few months, and a study run next year might show something else. The NYU study is several years old, and the tools have changed since. Read them, read the Raspberry Pi paper, and read the people who disagree.

Then decide from what you see in your own child. Do they want to know how things work, or are they content to press buttons? Both answers tell you something.

My view, after years of data science work and now a master’s in artificial intelligence: the children who understand code will direct these tools. I’d rather your child be one of them.