ChatGPT Work grounds data agents in semantic layers

Raw schema text-to-SQL was always an enterprise dead end because real businesses run on nuanced metric definitions rather than clean database tables, making semantic layer integration the true prerequisite for data agents.

• –Standard LLM text-to-SQL frequently misinterprets complex metrics like churn or ARR, whereas semantic layers supply explicit calculations and business logic.
• –Connecting via frameworks like dbt MetricFlow or Cube MCP enforces permissions, governance, and row-level security before queries execute on enterprise warehouses.
• –Grounded data agents shift data teams from manually building ad-hoc SQL reports to curating standardized metrics that AI can query autonomously.
• –OpenAI's architectural choice validates semantic layers and enterprise ontologies as mandatory infrastructure for production AI workflows.