This is commonly talked about in the design field as well (since at least the 1960s, probably earlier). On the one hand, there's a very industrial-psychology-derived HCI tradition, based on user-testing everything with statistical significance and so on, that tends to do relatively small changes, e.g. assuming that most of the design will stay as it is, how about widget A versus widget B? Against them, there's more of a design tradition, derived from architecture, graphic design, art, and other areas, that looks at more holistic sorts of issues, and tends to think that doing 1000 A/B tests does not add up to "design", and isn't likely to produce good outcomes either. HCI hasn't really resolved the issue over several decades, and it flares up now and then (my impression is that the main HCI conferences/journals are ~80% the industrial-psych sort, but the 20% dissenters have persistently stuck around for decades).