I would have no standing to write any of this if I did not say the following plainly.
Some work does go away. Not "changes" — goes.
Boilerplate. First drafts. The tenth CRUD endpoint that is identical to the first nine. The five-page brochure site that a small shop used to charge a few thousand for.
I know that last one specifically, because I built a company that automates it. Webplo takes a conversation and produces a live, hosted, working website in about a minute — the thing described throughout this book. That is a category of work that used to feed people, and we made it cost almost nothing on purpose.
So I am not an observer with an opinion about automation. I am a participant, and I am not going to tell you nothing is happening when I helped make it happen.
Here is what I noticed building it, though. Everything hard about that product is on the other side of the line. The model writes the copy. Every difficult decision — what happens when it is wrong, what it costs, what a duplicate does, what a merchant sees when a vendor is down, whether a tool can be talked into something — was engineering judgement, and there was more of it than I expected, not less. Eighteen chapters' worth.
The work that vanished was the work that never required knowing the good.
What I Actually Think
So: the anxiety is real, and it is aimed at the wrong thing.
The threat was never that a model would write code. It writes code, it writes it well, and it will write it better next year. The threat is spending a career producing only the kind of work that never required judgement — and not finding out which kind yours was until something else could produce it too.
The defence is the same as it always was, and it is unglamorous. Go deep enough in something real that you develop taste. Ship things that can fail and be there when they do. Read the code you did not write. Learn what good looks like in a specific domain until you can feel the wrongness in something plausible before you can prove it.
Do that and these tools are the best leverage anyone in this trade has ever been handed. You will produce more than you used to, and the thing you supply — knowing which of the ten plausible answers is the good one — becomes the scarce half of the transaction.
Skip it, and you are competing with something that generates plausible work instantly and never gets tired.
That is what I mean by it. If you know the good, you'll be good.
Not a promise. A description of where the value went.