[short_sells_poo]

I wonder if there's also another aspect: Games have to be fun for humans and this involves a lot of trial and error with actual humans in the loop.

You can't just spin up a bunch of claude agents to implement a game for you, because implementing a new feature requires that you playtest it with a fairly fine granularity as it's being implemented.

You can save some time implementing various subsystems with llms, but at some point the dev cycle will turn into: tweak things, build, play, rinse and repeat.

I'm sure asset flip and friendslop games will become cheaper to make with ai tooling, but if you want to make a genuinely good game, it will have to involve humans actually playtesting it during development.

[notahacker]

Also, game audiences are brutal about games that are flawed, and even more brutal if they think it's shovelware. If you're selling games on Steam, something with "Mostly Negative" reviews isn't serious competition (at least not unless it's part of a long established franchise with a big marketing budget).

And sports management sims are one of the most brutal of the lot. Slop might get you something that superficially looks like a sports sim (a decent third party UI library would get you quite some way in the past), but what fans actually care about is perceived realism, game mechanics, balance/challenge etc (and in real world sims, "how has my team been represented?") and that's a design decision, play testing, parameter tweaking and user engagement exercise, not a feature-adding one.