My personal experience is that system-level architecture diagrams are useful (i.e. which service/application talks to which service/database/message queue etc.) but code/class level ones are not. As to why - because the code tends to change a lot more often compared with the overall architecture of a platform. Requirements change (or requirements become better understood), code is refactored once the engineers get a better feel for the structure they are solving, new tools come along, etc. Code level diagrams have a habit of going out of date quickly, or (even worse) discouraging people from making useful changes based on new information.

To me, it seems that in their rush to criticize agile, the author failed to consider _why_ agile-like practices killed off UML.