I'm personally weakly +1 on this. It makes the combination of Arg and Max much more convenient to use and points directly to how to use it without code duplication. The potential downside is using a big scary extension in what will become a fairly mainstream module in 8.4, but I think we can phrase our way around the issue when it comes to the libraries part of Haskell-Prime.

This opens the door to possibly being smarter about the existing Data.Monoid.First (and Last), encoding them as pattern synonyms for Maybe (Data.Semigroup.First a) (or Last) as well, but we'd likely need to give that a lot more thought / testing.