Technology

Counterpoint: procedural generation in computer games existed at least since Rogue (1980) and is accepted as a well defined way to efficiently make randomized assets and asset combinations via set rules. It aces the tasks with trivial variation like said bookspines, but can go as complex as you set it up. Not everything can be supplemented by proc gen, but it's usually the answer when you have troubles with quantity of pieces as a priority rather than quality/uniqueness of individual ones. Examples range from Borderlands guns constructor set and Diablo maps, to Minecraft-alike sandbox worlds, to overcomplicated life sims like Dwarf Fortress where multiple systems of algorhytms shape every playthrough in a different way while also reacting to player's input.

Building such systems needs planning and thinking, but it is a potentionally endless source of intentionally designed stuff, that returns a piece on command.

GenAI though is unpredictable, unintentional and, as long as you don't build it into the game itself, consists of fixed resources you bake in.

The niche it primarily threatens is not background props, but high effort pieces like visual novel scenes, loading screens art, character's likeness, strings of dialogs and lore books that couldn't be proc gened and are created piece by piece by an artist. Making LLM create bookspines is a waste of tokens, and a waste of hard drive space too to pregen them all, it only makes sense if used in place of unique assets. And, these are the things most would prefer handcrafted for they form the face of the game more than anything else.