Is it still worth learning to code?

This is the question we get asked most, usually by a parent whose child is hesitating over their studies, sometimes by a business owner wondering whether to keep hiring developers. The wording varies, the worry is the same: why learn a skill a machine performs better and faster?

The short answer is yes, but not for the usual reasons, and not the way it worked five years ago.

Why the question is framed wrong

"Learning to code" covers two very different things that get routinely conflated. The first is syntax: knowing how to write a loop, remembering an array's methods, recalling how to open a file. The second is reasoning: breaking a vague problem into executable steps, anticipating what will break, deciding what not to build.

The first skill is genuinely losing value, and that is good news: it never had much. The second is gaining value, because it becomes the only place where a human contributes something the machine does not.

Judging is harder than writing

People underestimate how demanding review is. Writing code imposes its own discipline: you advance step by step, every mistake blocks you, the compiler complains. Reading code that is already written, coherent and neatly presented triggers none of those signals. You have to go actively looking for what is absent.

And what is absent is precisely what you do not think to look for if you have never had to build it yourself. You notice missing error handling only if you have already spent a night on an outage caused by missing error handling. Judgement is built by writing, even if it eventually replaces writing.

That is the paradox of the current moment: you need to have written a lot of code before you can afford to stop writing it.

What changes about learning

We do not think you should learn the way people used to. Three things have shifted in weight.

• Memorising syntax is pointless. Time spent recalling a function's argument order is wasted, and it already was.
• Reading code becomes the central exercise. A good drill today is to have a solution generated, then find what is wrong with it — and there is always something.
• Fundamentals matter more, not less. What a query is, what a transaction is, why a cache goes stale: these explain the errors AI produces, and they cannot be inferred from the generated code.

What about hiring?

The underlying question is often: can a non-developer with an AI replace a developer? Our experience says no, though not for the expected reason. The non-developer produces the first version perfectly well. They get stuck on the second, when the existing code must change without breaking, and they have no way to assess whether the result is safe.

Team composition does change, though. One experienced developer now covers ground that took three people. This is not the disappearance of the profession but its concentration: fewer junior roles doing execution work, and a greater need for people able to say no to an attractive architecture.

Our answer

Yes, it is still worth learning to code, provided you accept that the goal has changed. You no longer learn in order to produce code: you learn to be able to judge what gets produced, understand how it fails, and decide what deserves to exist.

Put differently, you learn to code for the same reason you learn to cook when you could order in: not to feed yourself, but to know what you are eating.