Why agentic analytics starts with a well-governed data layer

AI-driven analytics and agentic workflows are exposing long-standing weaknesses in enterprise data foundations: fragmented metric definitions, inconsistent governance, and proprietary BI tool lock-in. Nick Eayrs, VP of Field Engineering at Databricks, argues that trusted AI outcomes require a unified, open semantic layer with certified business metrics, end-to-end lineage, and role-based access controls built at the platform level. He outlines a sequenced approach: establish data foundations and a governed catalog first, then build the semantic layer with certified metrics, and only then layer AI and evaluation frameworks on top. Per-seat BI licensing is identified as an anti-pattern that limits democratization. Real-world examples from NTT Docomo and Net One Systems illustrate how proper data foundations enable dramatic efficiency gains when AI is applied on top.