Automatic selection of data representations is a problem with an interesting history and a bright future. This talk will give an overview of that story, with a focus on work with Mooly that took the subject in an unexpected direction.