1. Games are highly complex systems, so for example adding a fun mechanic from one game to another game can actually make that game less fun, and AI has no way to know, you just have to playtest it.
2. Games are evergreen (people still play and love the original Zelda/Mario/Tetris) and social (you want to share your love of game XYZ with others), so there's literally no reason for anyone to play a purely derivative game, when you can just play the original.
Eventually AI will be able to make perfect assets and all the other stuff AROUND making a fun game, but the game design will always be a human thing. At best, AI will come up with a bunch of game design experiments that a human will need to playtest and provide feedback on.
And I genuinely hope there will be some categories of games/music/books/art where human input is still needed and preferred. Currently I'm already a handful of bulletpoints down on the list of stuff I thought AI would never be able to do, and it's been kind of destabilizing
if we can't even write it down, how is an llm going to be able to ace it?
I’m having a lot of fun creating an “improved” clone of a very old school game I loved.
It takes lots of back and forth but there is definitely (for me at least) to play this derivative game since it adds a lot of levels and stuff that weren’t available in the original.
Without AI I wouldn’t be able to do anything; but with it it can make some changes, I test them and we iterate on it.
So for me AI solves “making a fun game”. It doesn’t “one shot GTA7” if that is your goal but that also sounds boring
*: I don't consider a project you built with an LLM to be "vibecoded" if all the important ideas are from your head, you know what the code is supposed to be doing (domain model is in your head) and you guided the project every step of the way. Otherwise we're all vibecode. So yea, I don't doubt you can create amazing games with AI, but "with" as in "together", not as in "Claude, build me a Minecraft clone, make it fun, no mistakes."