#llms

Mar 26, 2026

Ontologies: What Do I Mean vs What Do I Want?

The article argues that prompt‑heavy approaches to natural language querying are fundamentally brittle because they force LLMs to manage meaning, preference, and logic simultaneously—something they are not designed to do, leading to ambiguity and inconsistency in enterprise analytics. It positions Fabric Data Agents as a more structured alternative, where instructions are deliberately scoped to behaviour and intent, while meaning is externalised—reducing the fragility seen in traditional prompt stacks. Fabric IQ is presented as the key enabler of this shift, providing ontologies as a first‑class semantic layer that lets agents move from inferring business meaning to reliably navigating explicit definitions, making natural language over enterprise data more consistent and scalable.