AIgenerated game worlds are getting playable but does procedural coherence matter if it looks cool?

Google Genie 3 dropped and people are losing their minds over the fact that you can type a prompt and walk around inside it. Which is genuinely wild. But watching the demos I keep thinking about something nobody seems to be talking about: whether the world actually holds together logically, or if we are just impressed by the surface layer. Early procedural generation in games felt magical until you realized the towns had bakers with no wheat fields and guards patrolling walls that led nowhere. AI generated worlds right now feel like that but faster and prettier. The question for me is whether coherence is a prerequisite for this to be useful in actual games, or whether players just adapt and stop expecting internal logic when AI is involved. There is some evidence people tolerate a lot of weirdness if the aesthetic is strong enough. Also curious how this interacts with game narrative. If the world is generated on the fly, authored story beats become basically impossible to guarantee. Studios that rely on carefully placed environmental storytelling would have to rethink everything from the ground up. Is anyone actually working on the coherence layer, or is the bet that raw generative scale eventually produces emergent consistency on its own?