Ontologies Are So Back: Why AI Agents Are Reviving the Semantic Web

Artificial intelligence engineers are reviving ontologies, which are structured descriptions of data classes, properties, and relationships that function as data graphs. While large language models excel at probabilistic reasoning, they require logical guardrails to operate effectively in agentic systems. By combining neural networks with symbolic AI and knowledge graphs, developers can build neurosymbolic systems that keep language models on track.

Industry leaders are already implementing this approach. Neo4j uses three types of ontologies to create a shared semantic layer for agents at scale, covering business concepts, technical enterprise metadata, and runtime execution traces. Other practitioners, such as OpenLink Software, are building agent engineering stacks that utilize Semantic Web technologies like RDF memory. Because established web ontologies like Schema.org and OWL are already present in training data, developers can simply prompt for them rather than building from scratch.

Although maintaining ontologies historically stalled the Semantic Web vision from the nineteen nineties and two thousand s, modern AI agents can potentially help update definitions when encountering edge cases. This resurgence reflects a broader shift toward software engineering discipline in twenty twenty six, moving away from fully automated approaches and prioritizing quality control, guardrails, and humans in the loop.