AI engineers are rediscovering ontologies — formal descriptions of entities, classes, and relationships — as a way to impose deterministic guardrails on probabilistic LLM-based agents. Frank Coyle (UC Berkeley) and Neo4j CEO Emil Eifrem both presented at AI Engineer World's Fair on how ontologies, including established Semantic Web standards like OWL, RDFS, Schema.org, and FOAF, can constrain agent loops and validate LLM reasoning. Coyle frames this as 'neurosymbolic AI': combining neural networks with rule-based symbolic systems. The key benefit is keeping LLMs 'on the rails' during agentic loops. The main challenge remains ontology maintenance, though one proposed solution is having agents self-update their ontologies when encountering edge cases. The trend reflects a broader 2026 shift back toward software engineering discipline after the vibe-coding era.
