A 48-hour jam has almost no slack in it, which makes it a strange place to experiment with new tools for the first time, and also exactly where a tool that actually works pays off the most. The difference between those two outcomes is preparation you do before the jam starts, not during it.
Decide your AI usage before the clock starts
The worst time to learn a generator's quirks, its failure modes, how long a call actually takes, is at hour six when you're already behind. If you're planning to lean on AI tools during a jam, spend an hour beforehand generating throwaway test assets with the exact tool and workflow you intend to use live. You want zero surprises about how it behaves once the clock is running.
Where it earns its keep in 48 hours
Anything that's normally a time sink but isn't your game's core hook: placeholder art you'll never see again after the jam ends, a batch of item names, filler dialogue, a first draft of a settings menu's copy. None of this is where your jam entry wins or loses, so speed matters more than polish, and that's exactly where generation is strongest.
Where it costs you time instead of saving it
Iterating endlessly on a hero asset trying to get AI generation to match an exact vision in your head. If it's not converging in three or four attempts, stop and either simplify what you're asking for or do it by hand. A jam's clock punishes perfectionism harder than almost anything else, and a generator can absolutely become a new outlet for that same perfectionism instead of curing it.
The actual jam-winning use of AI
Protecting your scarcest resource, which in a jam is never art or code, it's your own attention and decision-making energy over 48 sleep-deprived hours. Every minute a generator saves on a low-stakes asset is a minute you get back for the one decision that actually determines whether the jam entry is good: is the core mechanic fun in the first ten seconds.