I'm trying to establish convention in our lab that bibliography generation is the last step before submission. It always ends up that the bibliography (ie field codes with ref info in Word) is the most fragile part of the document, and everyone that isn't the person that generated the bib is terrified of changing anything in it for fear of ruining the references.

It seems a lot more sane to use nicknames for references and then format references right at the end when you know it's not going to be edited any further. It still seems like it's an unsolved problem in the process.