After much thinking and deliberating on these tweets from a few days ago, it occurred to me that I've been using AI somewhat wrong. I'm still stuck to my previous mode of operation, make a game piece by piece, because I need to have control over the small details (juice, sounds, effects, etc) that make the game good. But the focus on "not being slop," let's say, made me blind to alternative paths. If I want a build-heavy game, why can't I just have the AI generate a game with 300 items and then work backwards from that artifact? It shouldn't be that extreme, but for each model release, I should definitely be doing a... like a shake sort, I should be pushing the model to its limits to figure out what those are, and then building everything against those limits instead of trying to fix the limits first. Concrete example, with Horse Game I'm building it item by item, making sure each item has the right juice, sounds, effects to it, otherwise the default job the model does is bad. But after I had a reasonable gameplay prototype that I thought was fun, I could have just had the model generate a game with a complete set of ~300 items, and then worked backwards from there, fixing things as I went along. Or, even better, decide on a few basic types of effects and how they should be handled, and then once those pass my bar, have the AI go at it with the full game. Essentially I should be thinking more about building pipelines that pass my bar of quality that I can then throw the AI at, instead of focusing on doing things manually. I think I will try to go in this direction more and see what happens, the fact that I had to do manual work still was bothering me, I'm a systems type of person, why am I not building systems? Something like this, for instance, is much closer to how I should be going:

Not in style or anything, but, I should be building pipelines that allow me to exercise my quality filter at an industrial level. Ideally, the AI generates images/videos/animations of the various things it codes whenever I send it on a 10 hour quest to build a version of a game, and I say yes or no over and over again until it learns what it needs to learn. This is especially true for build-heavy games where lots of things compose and combine in interesting ways, so there's a lot you can do to be clever about how much work is to be done. This game also made me think about this a lot:

I guess I just need to think in a more maximalist way.