AI agents can write syntactically correct SQL but fail in production because they lack domain knowledge about what database columns, tables, and business logic actually mean. The solution is a semantic model — a structured file encoding tacit domain knowledge such as table relationships, column meanings, deprecated fields, and valid query patterns. This approach, to be demonstrated by Kris Jenkins (Lead Developer Advocate at Snowflake) at Codemotion Rome 2026, aims to build reliable AI database analysts from day one rather than through months of prompt engineering and fixes.
Table of contents
AI Literacy in 2026: How to Lead a Team of Synthetic Agents — and WinThe problem no one expectedThe solution: give the agent a map of the territorySort: