Judgement needs practice too

One of the more interesting risks from AI in software isn’t that senior developers disappear. It’s that junior developers get very good at producing software before they’ve had enough experience to know when the software is wrong.

That distinction matters.

A lot of judgement is built through repetition: debugging something that should have worked, discovering the edge case you missed, reading somebody else’s code, making a bad design choice and then living with the consequences.

AI can remove a surprising amount of that friction. Mostly that’s a good thing. I use it precisely because it lets me build beyond what I could comfortably produce unaided.

But some of the friction we remove was also practice.

So I don’t think the answer is to keep junior people away from AI. That would prepare them for a world that is disappearing.

The harder question is how we deliberately create the experience that used to arrive as a by-product of doing the work.

Perhaps junior developers need to spend less time proving they can type the answer and more time explaining it, testing it, breaking it and challenging what the model produced.

AI can accelerate the doing. We still need to design for the learning.

Editorial decision (25 September 2026): publish on Threads, X and Bluesky. Suitable for LinkedIn as a short standalone post because it extends the existing apprenticeship-problem strand without requiring a new long-form article. This is a draft, not a claim of publication.