💻 What Comes After Human-in-the-Loop: Engineering the Agent Control Plane

AI agents now open pull requests, triage incidents and run migrations. Capability is no longer the constraint, verification is. In most organisations, the control plane for AI agents is still a person: someone reading diffs and clicking approve, quietly becoming the bottleneck and the single point of failure. This session is about what replaces that person; not nothing, but infrastructure. It presents four architectural patterns for building oversight into agentic systems, a decision framework for when an agent has earned more autonomy, a measurement model for whether oversight is actually improving, and a map of the problems the industry has not solved yet. It closes on a single reframe: the question is not whether the agent can act, but whether the system can explain why it was allowed to.