Show what it takes to put an AI agent on top of enterprise data in industries where every answer may be audited. Using two live builds on public data — a clinical trial landscape assistant for pharma and a near-real-time fraud-scoring pipeline for banking — we demonstrate an agent that discovers schema over MCP, applies the organisation’s definitions through a semantic layer, runs inference inside the database, and leaves a query log that can be replayed months later. The aim is that attendees leave knowing the difference between an agent that has been handed database access and a database built for agents.