A guided video lesson

Why Agentic Systems Need Ontologies

A source-grounded lesson on pairing probabilistic LLM agents with formal ontologies, graph representations, and validators so flexible generation and action planning can be checked against domain-aware rules.

Open the complete source

日本語版を読む

Chapter pages

Choose where to begin

  1. 01Why this matters: build to learn in an AI transition0:00 - 2:23
  2. 02Two lineages: agents and ontologies2:23 - 4:08
  3. 03Convergence: probabilistic agents with symbolic guardrails4:08 - 5:23
  4. 04Ontology as graph: entities, relationships, and extensibility5:23 - 6:26
  5. 05Top-down modeling and the limits of early symbolic AI6:26 - 7:52
  6. 06Bottom-up modeling and reusing existing vocabularies7:52 - 9:15
  7. 07RDFS domain and range: deriving types from graph statements9:15 - 10:43
  8. 08OWL properties: transitivity, functionality, and constraints10:43 - 12:23
  9. 09Why agent loops are powerful and risky12:23 - 14:23
  10. 10Inside the Claude-agent tool loop14:23 - 16:34
  11. 11Putting ontology in the tool-result validation loop16:34 - 17:45
  12. 12Pydantic at the input, ontology at the result17:45 - 18:45
  13. 13OWL checks semantic errors that text can miss18:45 - 20:05
  14. 14The proposal: reason over the ontology, then let the agent act20:05 - 21:18