As agentic AI moves from assisting agents to independently handling tickets, drafting responses, and closing cases, support leaders are being forced to define where human oversight still matters. The challenge is no longer whether automation works, but how much autonomy is safe in live environments where errors scale quickly and quietly. Leaders discuss how they are deciding which interactions can safely run end to end, what triggers escalation in practice, and how visibility is maintained when decisions happen at scale. Expect honest perspectives on where boundaries are being set today, where they are already being tested, and what many teams are only realising once systems are live in production.