Ontologies Are So Back: Reading a Semantic Web Revival Against a Live Corpus

"This stuff has been out there underlying a lot of what we already do — so take advantage of these things that already exist." — on reusing schema.org, FOAF, and Dublin Core instead of inventing ontologies from scratch.

"Thick agents with manually wired data sources" give way to "thin agents on a smarter shared ontology-based semantic layer" — business, technical, and execution-trace ontologies running underneath every agent.

"The beauty of LLMs is that they are powerful processors of language. The beauty of an ontology is that it defines the types of entities and relationships through which language acquires computable context. Together, they bring the expressive power of language to computing's UI/UX stack."

"When an agent maintains the ontology as part of its own operation, updating definitions when it encounters edge cases, the maintenance problem changes character" — though it remains a hard problem.

"Ontology maintenance is a solved problem — because the missing piece was never willpower, it was a dynamic ontology creation process built on the actual RDFS/OWL toolkit, not just a flat lookup table. When agent-rdf-memory encounters a genuinely new concept at an edge case, it doesn't just check for an exact duplicate — it places the new term properly within the existing structure using rdfs:subClassOf to slot it into the right class hierarchy, rdfs:isDefinedBy to record which ontology owns it, and owl:equivalentClass (for classes/properties) or owl:sameAs (for individuals) to align it with anything already canonical elsewhere in the corpus or in a standards vocabulary. That's the same toolkit OWL and RDFS were built to provide in the first place — it was just never paired with something fast enough to apply it continuously. Sarkar's proposal names the right actor but not the right method; this is the method.

"you have to be over 40 to even discuss ontologies" — a skeptical aside the article quotes alongside the maintenance-problem debate.

And this is only possible because of the LLM/RDF symbiosis this whole article is really about. The Semantic Web vision didn't die — it stalled, starved of the one thing that could apply its own RDFS/OWL toolkit continuously and at scale: ontologies could only be extended by the small number of humans who understood them, so growth calcified into maintenance overhead. LLMs reopened that pathway by creating exactly what Coyle calls neurosymbolic AI — a neural system fluent enough in language to judge where a new term belongs in a class hierarchy and recognize an equivalence match against the existing registry, at a speed and volume no ontology committee ever could. RDF supplies what the neural half alone cannot: a durable, queryable, dereferenceable record of the decision. Neither half regalvanizes the vision alone. Together, through that neurosymbolic pathway, they're why ontologies are so back — not something only a fortysomething could love.