Why Agentic Systems Need Ontologies — Frank Coyle, UC Berkeley

Checking that an agent supplied correctly formatted data does not establish that its proposed action makes sense. This ontology proposal adds a representation of a business's entities, relationships and rules, so a validator can examine the meaning of an action before it changes records.

The talk uses customers, orders and refunds to explain the distinction between checking data types and checking domain relationships. Its intended benefit is to make business constraints explicit at the point where an agent acts. That requires software which enforces those constraints; defining an ontology alone does not establish that an erroneous payment will be blocked.

“People worry about hallucinations, but that's the feature.”

Frank Coyle

“LLMs can't do anything. All they can do is give us the next word with a high probability.”

Frank Coyle

“you want to check your types with Pydantic and then check your results with the ontology”

Frank Coyle

“Nothing is a mistake. There is no win. There's no fail. There's only make.”

Frank Coyle

“Loops can drift as agents start talking to each other, things get all go off off the rails.”

Frank Coyle