Most AI agent deployments fail because we're solving for the wrong problem. We optimize for accuracy when enterprises need trust. We measure task completion when they need consequence tolerance. We build for automation when they need collaboration.
This talk introduces a battle-tested framework for building agentic systems that cross the chasm from POC to production: the Five Pillars of Calibrated Trust. Drawing from real deployments, we'll explore how Success Measures, Transparency, Value Delivery, User Experience, and Consequence Acceptance create a complete trust equation for autonomous systems.
You'll learn how to implement meaningful transparency through source attribution and reasoning traces, why context management metrics like precision and recall matter more than accuracy alone, and how task decomposition enables both adaptability and auditability. We'll examine the UX continuum from HITL to HOTL, showing why interface design determines adoption more than model performance.
Most critically, we'll explore calibrated trust: the balance where users appropriately match their reliance to agent capabilities. Through examples of sequential processing patterns, decision point architectures, and progressive disclosure interfaces, you'll leave with actionable strategies for building systems where imperfection becomes a managed feature rather than a hidden risk.
This is a practitioner's guide to making agents work in enterprise environments, where success means building systems teams trust enough to use and reliable enough to depend on.

Intermediate > Expert