AI really does help a beginner write things that used to be out of reach. But working code isn’t yet a solution to the problem.
An experienced developer notices earlier: the problem statement is missing data, the requirements contradict each other, the “quick” option will break maintenance in a month. AI doesn’t change that. It just implements both good and bad ideas faster.
A practical technique: don’t start the prompt with “write the code”. First ask the assistant to run a short interview about the task.
Before proposing a solution, ask up to 5 questions without answers to which there is a risk of choosing the wrong architecture. Separate critical questions from nice-to-have ones.
Answer the critical questions — and only then ask for a plan or implementation.
Why this works: you shift the value from typing lines of code to choosing constraints. That’s where expensive mistakes most often live.
When you don’t need it: for a local, well-understood edit — for example, renaming a field or adding a check in a module you already know.
The market hasn’t become easier. It’s just that “being able to type code” has become a less rare skill, and being able to pick the right problem is even more valuable.