Is the “semantic web” used in sensible, productive ways these days or does it remain an academic daydream?

I’m highly skeptical of the whole concept and feel like I should have organically come across it by now.

Yeah, it is - not to the level people envisioned years ago where people made up their own tags and found a blend of XML and HTML, but it absolutely is in place for more meaningful HTML. We have tags that tell you the purpose of content - header, nav, section, article, footer, etc. Those tags are read by screen readers and other accessibility tools, and do bring semantic value to those readers.

Those things are semantic HTML elements, not the semantic web. The semantic web does have some HTML integration though, e.g. RDFa.

If you get job where you need to merge graph data from a bunch of different sources, you will likely run into the semantic web stack. The semantic web stack is basically a bunch of technologies for building graph databases and sharing graph data across domains. It isn't some huge network of interlinked data sources like originally envisioned, but it is definitely not dead. I think the concept of graph databases is still something people are only beginning to discover the uses for. Most people only know of the initial XML serialisation of RDF and OWL which rightfully made people skeptical. Nowadays, the most common serialisation format is TURTLE which is basically just a subset of SPARQL and it is very readable.

In fact, the underlying technologies are all quite mature (triplestore implementations and the RDF, SPARQL, OWL, etc. standards) and used heavily in certain domains, such as biology and language. The standards are designed for easy serialisation and interoperability which is the antithesis of the "modern" walled garden and/or commercial API approach of the big Internet companies, although they also use semantic web tech to a certain extent.

In Europe, you will run into RDF quite often in academia (and research in general). I'm working on the newest version of the Danish WordNet, for example, which is published as RDF and OWL (in the turtle format) and which runs in Apache Jena. Coincidentally, I am actually developing this new version in Clojure which stole the triplestore concept from the semantic web stack and made it a mainstream database type in the Clojure ecosystem (Datomic, Datascript, Datahike, Datalevin, Asami).

The EU is heavily investing in semantic data for education and digital identity, leaning on ontologies, RDF, json-ld and verifiable credentials.

Banks, research institutes, universities are joined in large consortiums. The German hotel reservation example above is one of the projects that’s getting attention.

There is some neat application for example many tickets/hotel reservation/restaurant do include them and this allow to make some email clients displays these tickets in a standard way.

KDE Itinerary and Apple Wallet make use of this. For the KDE Itinerary part, you can read more on this here: https://volkerkrause.eu/ (look at the KDE Itinerary posts)

I do miss the glory days of when the W3C was still in control and had some very good ideas that browsers did not like.

Some of these old AAA XHTML 1.1+RDFa websites were well engineered and a joy for both man and machine to behold.