Who watches the watchmen: establishing trust in agent interactions

The proliferation of agentic tools is one of the defining technological moments of our time. But anyone who has actually used an agent has hit the same problem: they respond with total confidence, even when they're wrong. A correction, a "you're totally right, I missed that", and you're back on track. Until the stakes are high. Then confidently incorrect answers can be devastating to a business.

dbt's approach is simple: rather than trying to make agents perfectly accurate, make them more self-aware, forcing them to acknowledge and address uncertainty before it compounds.

This session walks through how we've built that into dbt's agentic workflows, and what it means for teams relying on agents for production data decisions.