To add on to the exceptions section at the end, in data engineering you generally need to denormalize/flatten/materialize (or whatever term your org uses) data and tables downstream of your gold source.

While you can apply some principles from the article in DE many devs make the mistake of doing it fully and use views which usually end up far less performant than a standard scheduled create/replace statement would produce.

This is a good insight, I hadn't thought of that perspective

The idea of trying to not store calculated data seems to be a way to say "avoid caching if you can". That makes sense to me since caching can be really hard to deal with, and if you don't need the performance avoiding it makes the code simpler.