The floor came up. The ceiling did not.

AI made competent work cheap. An average landing page, component or paragraph of marketing copy is now free and instant, so the minimum quality of work went up, but the best work did not get any better. The models also learned from millions of artists, writers and developers who were never asked, and now they compete with those same people. Both facts are true at once, and the usefulness of the tools does not cancel the second one.

Speed changes what work means. Almost anything can be built and thrown away in an afternoon, so a prototype is already real code, and a week of effort no longer proves that an idea is serious. But building and maintaining are different jobs. A new thing can take destructive changes, because nothing depends on it yet. A product that people use every day needs someone who knows how it works, because a bug given to a model without that knowledge gets moved, not fixed. Most of the disappointment with these tools comes from people who crossed from building to maintaining without noticing.

Taste did not get cheap, and neither did patterns. Code works or it does not, and a test tells a model when it is correct. Good has no test like that, because it changes with the context, the audience and the time. A model also works one screen at a time, so it can make a good screen, but not the pattern that keeps a hundred screens in agreement. Taste is what remains after enough bad decisions to recognize the next one before it happens, and a model that learns from the average of everything makes average work that nobody remembers.

Photography did not end painting. It ended painting whose only goal was accuracy, which was most painting at the time, and the painters who continued did what a camera could not. AI will probably do the same, and it will be painful, because much of what all of us get paid for is accuracy. The distance between "I wish this existed" and "I have it now" is almost gone, and that is a delight. But a technology that can be mastered completely is not worth much, so AI should continue to feel overwhelming.