The Metric Contract Mandate: Standardizing Semantic Layers Before AI Agent Access

This article discusses the critical requirement for metric contracts—machine-readable definitions of business metrics like revenue, retention, and active users—before granting AI agents access to data platforms. It argues that text-to-SQL tools fail because they generate correct SQL but wrong business answers due to ambiguous metric definitions. The article details what a metric contract contains (calculation rules, ownership, grain, dimensions, freshness, tests, lineage, access policies), why semantic layers need to act as execution contracts, and how to build a metric contract program with versioning and change management. It emphasizes that agents need governed, open lakehouse data foundations and provides guidance on procurement, documentation, and anti-patterns. The content is highly relevant to IT/technology, focusing on data engineering, semantic layers, AI governance, and best practices for agentic analytics.