Back to Speakers

November 23–27, 2026

Craig Tucker

CEO

VERN AI

Session Details

Your AI Agent Is Out of Control!

Most AI agents are not actually under control. They are guided by prompts, policies, and model behavior that remain probabilistic. The same input can produce different behavior, instructions can be interpreted inconsistently, and a model update can change how an agent behaves without changing a single line of your application. That creates a serious problem for anyone deploying agents into healthcare, financial services, customer support, education, or other environments where consistency matters: How do you put deterministic behavioral boundaries around a probabilistic system? This talk presents a different architecture: An independent control layer that sits between the model and the application. Using examples from more than 50,000 production conversations, we’ll show how deterministic behavioral controls and human-signal detection can constrain what an agent is allowed to do, detect when an interaction changes, and alter the agent’s behavior in real time. Attendees will see examples involving hallucination prevention, emotional escalation, role containment, persona consistency, and graceful exits. The goal is not to make the model more intelligent. It is to make the system more controllable. Attendees will leave with a practical framework for separating intelligence from control when building production AI agents.

Full conference agenda

See every session across the AI Agents Conference November 2026 line-up.

View Agenda