What happens when a nine-year-old can build the thing he imagined?
His idea wasn't the problem. His ability to execute it was.
He knew what he wanted
My nine-year-old son wanted to build a Minecraft-inspired clicker game.
Although, technically, the idea was two years old.
He'd first tried building it in Scratch when he was seven. The idea was fairly simple.
Click a block. The counter goes up. Reach 100 clicks and you can spend them on an auto-clicker. Your balance resets and the game starts clicking once every second for you.
At seven, he'd had a go at making it work.
Then, like quite a lot of projects started by seven-year-olds, it got left behind.
Two years later, he came back to it.
Same basic idea. Two more years of playing games like it. And this time, a much clearer idea of how he wanted his version to behave.
He opened Scratch again and started rebuilding it.
And he got surprisingly close.
To the naked eye, it looked like it worked. Click the block, number goes up, buy the upgrade.
Except when we actually played it properly, the logic started falling apart. The reset wasn't behaving correctly and some of the numbers weren't doing what he thought they were doing.
He tried fixing it.
Then tried again.
Eventually he reached the point where he knew exactly how he wanted the game to behave but didn't know how to make Scratch do it.
Which, for a nine-year-old, seems fairly reasonable.
The interesting bit was that his idea wasn't the problem.
His ability to execute it was.
What if AI filled that gap?
I'd had a go at solving it with him.
Then I suggested something different.
Rather than asking AI to make him a game, I wanted to see whether we could use AI to extract the game that already existed in his head.
So we got ChatGPT to start asking questions.
A LOT of questions.
More than 100 in the end, mostly multiple choice.
What happens when somebody clicks the block? What happens when they reach 100? Does buying an auto-clicker reset the balance? How quickly should it click? What should unlock next?
These weren't questions about programming.
They were questions about his game.
And he could answer them.
We then ran the thinking through Claude and Gemini, looking for gaps and contradictions before anything was built.
I took over some of the later technical decisions around turning it into a progressive web app and actually delivering the thing.
But the product itself was still his.
That became Blocks.
Then we played it
This is probably the bit I wasn't expecting.
We didn't just open it, click a few buttons and declare victory.
He tested it.
He didn't know he was doing UAT. He was nine.
But that's essentially what he did.
Then my wife and I played it.
It took us about ten minutes to work through the game and reach the end. And there was a strange little sense of achievement when we did.
More importantly, he was proud.
This was an idea he'd first had when he was seven.
Two years later, after picking it back up, trying again and reaching the limits of what he currently knew how to build, we were sitting there playing it.
The thing that had existed in his head now actually existed.
And he hadn't given up to get there.
Apparently, shipping isn't the end
I thought getting the game working might be the finish line.
It wasn't.
Almost immediately he wanted to change it.
More blocks. More levels. Different progression. New ideas.
He's since gone away and designed additional blocks and started thinking about what should come next.
So I introduced another idea.
Don't try to build everything at once.
We'll improve it incrementally. Decide what matters next, build that, test it and then decide again.
Without particularly intending to, we'd moved from a Scratch project started at seven into conversations about iteration, product development and two-week sprints at nine.
Which is a slightly ridiculous conversation to be having with a nine-year-old.
But he understood the principle perfectly well.
Did AI actually teach him anything?
This is where I think the story gets more complicated.
Did he learn JavaScript?
No.
Could he now build the application himself from scratch?
No.
Did AI teach him programming?
Also no.
And I think pretending otherwise would miss something important.
What he learned was that he could take an idea, define how it should behave, answer questions about it, test the result, find problems and then improve it.
AI removed an implementation barrier.
It didn't magically give him the knowledge sitting underneath the implementation.
And I don't think those two things should be confused.
I still want him to learn Scratch
If his school gave him a Scratch programming exercise tomorrow, I wouldn't want him giving it to ChatGPT.
I'd want him to struggle with it.
Not because struggling is somehow virtuous, but because there are fundamentals I still want him to understand.
Variables. Loops. Conditions. Functions. Cause and effect.
The things underneath the thing.
Because eventually he's going to ask an AI to do something and it's going to get it wrong.
If he has no understanding of what's happening underneath, how does he know?
That question isn't particularly different for a nine-year-old using AI to make a game than it is for an experienced developer using AI to write production code.
The scale of the consequences changes.
The underlying problem doesn't.
So where should the ceiling be?
Before AI, there was a fairly hard limit on what my son could create.
It wasn't his imagination.
It was his technical ability.
In fact, Blocks is a fairly neat example of that.
The idea survived for two years.
His ability caught up with some of it.
AI bridged the rest.
And when we removed some of that limit, something interesting happened.
He didn't become less interested in creating.
He became MORE interested.
The first working version didn't satisfy him. It gave him a platform for the next idea.
As a parent, that's difficult for me to dismiss simply because he couldn't have written the code himself.
But I don't want AI replacing the understanding he still needs to develop either.
So perhaps the choice isn't between teaching children to code and letting them use AI.
Perhaps we need to do both.
Teach them enough about what's happening underneath that they can reason about what they're creating, challenge what they're given and recognise when something isn't right.
Then let the tools take their ideas further than their current technical ability would otherwise allow.
Because watching my son finally build an idea he'd first attempted when he was seven left me with a question I don't have a particularly neat answer to:
If children can now create things before they fully understand how to build them, how much understanding is enough before we let AI take them further?