Overcoming the Trust Deficit: Taking AI Agents from Prototype to Production - Manish Bhide

This keynote discussion addresses the challenges organizations face in deploying AI agents to production, despite significant interest among developers. The speaker examines the trust deficit related to issues such as robustness, reliability, safety, and policy alignment that hinder the transition of AI agents from development to real-world applications. Key topics include adversarial robustness testing and the concept of natural robustness, where minor input changes can lead to unpredictable outputs. The session also highlights the importance of implementing continuous guardrails and runtime policy checks to support trusted deployment. Attendees will gain insights into designing, evaluating, and hardening AI agents, ensuring evaluation stability, and building resilient production pipelines to navigate real-world variability.