Agentic AI in Production explores what happens when AI systems move beyond answering questions and begin taking action across tools, workflows, and live enterprise environments. The talk argues that production agents are not just a model-performance challenge, but a delegation, governance, and accountability challenge.

The session introduces a practical framework for evaluating agentic systems by their function, authority, reversibility, and operating context. Using real-world examples from customer support, sales chatbots, healthcare-adjacent tools, coding agents, and large-scale automation, it shows how failures often occur when organizations hand over responsibility faster than they build controls.

The talk highlights key production concerns for product, engineering, and cloud teams: observability, source-of-truth grounding, least-privilege access, dev/prod separation, rollback, human escalation, and vendor tradeoffs. The central takeaway is simple: before increasing autonomy, teams must first ship the control layer.