They're not that different from the classic nix files. Their main difference is that their inputs are always well-defined (as opposed to a channels registry, i.e. you can get totally different systems by reapplying a configuration when you change channels which doesn't change your nix file at all). A configuration is always exactly described by a flake.nix and flake.lock.

I mean there is more to it, but this is the primary motivation. What you would normally put into use case specific nix files goes into a flake's output section. The stuff in your input sections is what you can use in there.