The stuff that can be finished in a week often feels like my best work, though, at least intellectually (and you can get a good amount done in a week if you're excited about it). The week-long side projects tend to have more novelty, which sometimes even results in more impactful published papers: you're proposing the first X to do Y, and you have a working system to boot (in academia not always required, but a nice plus if you have it).

My biggest projects tend to require some focus and months, but they tend to be more of the grind variety. Take that initial prototype, either one I've produced or one someone else has published, and build a Real System on it, working out all the details, including engineering and theory as appropriate.