I'm seeing two effects that feel immediate and don’t get enough attention.
First, the "repair economy" of software is dying. Maintaining old software might stop being the default. In the past, repair culture disappeared once manufacturing became cheap, think shoes, clothes, electronics, all moved from repair to replace. With AI making it cheap to generate new code, tests, and scaffolding, it may actually be faster to rebuild than understand what someone wrote ten years ago.
Second, not all code is equally replaceable. AI will go after problems with “many right answers” first, like most UI, basic CRUD apps, most everyday engineering tasks. Art follows the same pattern too. But systems where domain logic is deep and architecture is intentional are harder for AI to replace because constraints and correctness matter more than variation.