For about a decade, "teach kids to code" was the most agreed-upon idea in education. Politicians said it, tech CEOs said it, school boards funded it. And the premise underneath it was right: kids who can make things with computers hold a kind of power that kids who can only consume them don't. I believed it then and I believe it now.
But the argument always had two parts, and we rarely separated them. Part one: making things with computers matters. Part two: the way in is syntax, semicolons, loops, and the patience to type them correctly. Part two was never a principle. It was a description of the bottleneck at the time. Syntax was the gate you had to pass to reach the making.
The gate moved. A ten-year-old can now describe a game in plain English and watch a working version appear. You can call that a crisis for coding education, and if your curriculum is a syntax curriculum, it is one. Or you can notice what just happened: the making, the actual point, is suddenly reachable by every kid who can think clearly about what they want. The scarce skill stopped being the typing. It became the thinking.
"But the AI wrote the code."
The objection arrives on schedule: if the machine writes the code, the kid learned nothing, and letting them use it is cheating. It is the same argument we had about calculators, and the resolution will be the same too. We still teach arithmetic. We stopped pretending long division is the summit of mathematics. The calculator didn't remove the need to understand numbers; it removed the ceiling that manual computation placed on which problems a student could touch.
Whether AI-written code is cheating depends entirely on what you claimed to be teaching. If the goal was "produce syntax," then yes, the machine defeats it, the same way a calculator defeats a long-division worksheet. But if the goal is "conceive something, get it built, and know whether it works," then the AI is not the cheat. It is the instrument. And the honest question for every coding curriculum is which of those goals it was ever really serving.
What the real literacies look like
Watch a kid build something real with an AI and you can see the actual skills, the ones with a long shelf life, surface one by one.
Framing. Saying exactly what you want turns out to be brutally hard, and it is the first thing the machine punishes you for skipping. "Make a fun game" produces junk. "Make a game where you catch falling stars, and it speeds up every ten catches" produces a game. The AI is a mirror for fuzzy thinking, and kids learn to sharpen the thought because they want the thing.
Decomposition. Big wants don't fit in one ask. The kid who wants a quiz app with scores, sounds, and a leaderboard learns, quickly and from experience, to split it into pieces small enough to build and check one at a time. That skill has a name in computer science, and computational thinking was always the defensible core of the coding movement. Syntax was just the container we shipped it in.
Evaluation. The machine returns something. Is it right? Does it do what was asked, or something adjacent that looks similar? The kid who tests, pokes at edges, and catches the case where the score goes negative is learning the single most transferable skill of the AI age: never confuse fluent output with correct output.
Debugging by description. When it's wrong, you have to say how it's wrong, precisely, in words. "It doesn't work" gets you nothing. "The stars fall behind the score bar so you can't see them" gets you a fix. That is technical writing, observation, and hypothesis in one motion.
Taste. Eventually: is this good? Is it too complicated? Would anyone want to use it? Taste only develops through volume, and volume is exactly what the new tools unlock. A kid can now build in an afternoon what used to take a semester, which means they can be wrong, and learn from it, twenty times faster.
The caveat I owe you
None of this means reading and writing code is obsolete. The kids who go deep should absolutely learn what's under the hood, the way a serious musician learns theory even though nobody buys tickets for theory. A generation that can only prompt, with no mechanical sympathy for what the machine is doing, has its own failure mode, and we should be honest about it. But the sequence has flipped. Build first, with AI, because that is where the motivation lives. Learn the internals when the thing you're building makes you want to know. Syntax as a prerequisite filtered out most kids before the interesting part. Syntax as a consequence of curiosity filters out almost no one.
We're testing this progression with real kids in our own Build Lab prototype: remix something that works, then fill in blanks, then build open-ended, then debug on purpose. The early observation that matters: the kid who can't type a semicolon can decompose. The gate really was the gate.
What to do this year
If you're a parent, you don't need a curriculum to start. Let your kid build something they actually want with an AI tool, then ask four questions at dinner: What did you ask for? What came back? How do you know it works? What would you change? Those four questions are the literacy. The rest is practice.
The age of AI does not need a generation of typists. It needs a generation of clear thinkers who can direct enormous power precisely, check what it gives back, and want to build good things with it. That is teachable, starting around age nine, and it is a lot more fun than semicolons.
- Wing, "Computational Thinking", Communications of the ACM (2006)
- Weber-Wulff et al., "Testing of detection tools for AI-generated text", International Journal for Educational Integrity (2023)