November 23–27, 2026
Staff Engineer / Regional Director
Plan A Technologies
AI agents are often presented as autonomous systems that can reason, use tools, and complete tasks on their own. But in production, autonomy without boundaries quickly becomes a liability: tools fail, context is incomplete, models make surprising decisions, and retries can amplify a small mistake into a real incident. This practical session explores how to engineer bounded autonomy — systems that know what they are allowed to do, when they need confirmation, and when they must stop. Through a failure-driven demonstration, we will examine the boundary between model-driven decisions and deterministic software, then add guardrails for tool calls, structured outputs, validation, budgets, retries, human escalation, and safe recovery. Attendees will leave with a concrete mental model for designing reliable agentic systems, a failure-mode checklist, and practical questions to ask before allowing an agent to take action in the real world.
See every session across the AI Agents Conference November 2026 line-up.
View Agenda