Create templates to quickly answer FAQs or store snippets for re-use.
I have used only class diagrams from UML for 20 years - it's so much easier to deal in 2 dimensions i.e. a class diagram, when designing or documenting code than living in a world with only a one dimensional stream of characters - i.e. source files :)
Class diagrams are really the only thing that have a one to one representation with the structure of the artifacts in your code base so they are never too far distant from your code.
I like Grady Booch's first few comments you listed in your post.
I still use UML but only really class, sequence, and state diagrams. I use UML to communicate new designs, reverse engineer legacy designs, and even during bug analysis.
I think it is great for getting an overview of object-oriented designs and behaviour especially for large code bases. In my experience, people unfamiliar with UML have been able to understand the diagrams I have presented. But likewise other people use ad-hoc diagrams which work fine too. However I think UML helps developers explore all the facets of a architecture/design in a structured, methodical way.
I think UML was useful for the first revolution (object orientation) but I agree with Grady's observations that what happens during the next revolution (AI) is up to those who work in the field. Whether anyone in that field sees value in having a common modelling language or not remains to be seen. However since agile development dominates at the moment, I think a modelling language for the next revolution needs to be a language primarily for collaboration.
Fully agree with the importance of collaboration. Still, this collaboration support does not only involve the language (which indeed needs to be flexible enough for this) but also the tooling. In this sense I think our preliminary work on using native collaboration platforms like Slack for modeling could be a good first step in this direction
We're a place where coders share, stay up-to-date and grow their careers.
We strive for transparency and don't collect excess data.