Ontology engineering: what it is, why it's back, and the half agents need that we haven't built

Ontology engineering — the practice of formally defining what entities exist in a domain, how they relate, and what rules govern them — is experiencing a revival driven by AI agents. When humans are removed from decision loops, the implicit tribal knowledge they carried must be made explicit. LLMs don't hallucinate so much as navigate without a map: give them a clean schema with semantic context and accuracy jumps significantly (BIRD benchmark shows mid-50s to 70s execution accuracy). The post argues that existing ontologies were built as comprehension layers for human readers, but agents need an 'agentic half' that includes decision logic and write-back capabilities. A practical test using GPT on finance questions shows that adding taxonomy and ontology layers reduces hallucination and produces semantically correct metrics (e.g., ARPU calculated against billed users in the same period, not all active users). The author advocates for lightweight ontologies in markdown files rather than formal OWL/SPARQL, since LLMs consume natural language natively. The post also promotes dltHub's AI Workbench and an Agentic Data Engineering course.