RDF, OWL and SPARQL spent two decades being technically impressive and commercially ignored. The reason was never the technology; it was that nobody wanted to write the ontology.

What changed

A language model will produce fluent text about your domain and will also invent a customer, a policy number and a date. Grounding it needs a structure that says what entities exist and how they relate — which is precisely what a knowledge graph is. The graph provides the facts and the constraints; the model provides the language.

The model is good at expression and bad at truth. The graph is the opposite. That is a partnership, not a competition.

What still hurts

• Ontology design is still slow, political work. Two departments will disagree about what a customer is, and the disagreement is real rather than technical.
• Graph maintenance has no obvious owner in most organisations.
• SPARQL remains hostile to newcomers in a way that SQL, for all its faults, is not.

The extraction half is genuinely easier now — models are good at proposing triples from documents. The governance half has not moved at all, and that was always the expensive part.