Comment by ghaff
I've even had this discussion with people in at least adjacent roles to mine. We could put together ten very serviceable conference submittals tomorrow that we could spend a week turning into very competent conference presentations within a week. They wouldn't be brilliant but they'd be "good" relative to the norm. It's just one of the things we can crank out like we crank out blogs/columns.
Depends of course on the field and the conference tier. I doubt you could write an accepted paper for ICLR/ICML/NeurIPS/CVPR/ICCV in a week from scratch. These typically involve writing code, doing experiments, evaluating on benchmarks and obtaining good, if not state of the art, scores.
Typically this does involve 4-6 months of work even in high-output labs. It just involves many iterations, polishing the text and figures to "look like" an accepted paper by gestalt etc.
I wasn't trying to imply there's no work or effort needed. I'm more saying that there's a lot of learnable, repetitive parts and if you get them right, your output can have a consistent good quality but nothing groundbreaking. Chipping away at problems, permuting the assumptions and tasks, throwing in the latest hot new method and combining it with another method etc. It isn't something that anyone could do, but it doesn't rely on big eureka moments or flashes of deep insight that one might romantically expect.