Humans, governance and the limits of autonomy

What must remain human and what breaks if it does not?

Autonomy is only useful when it operates within a system of accountability. Supply chains already have approval rights, audit trails, safety requirements, labour agreements, compliance obligations, and people accountable for decisions. This pillar examines how agents can operate within those structures and where human ownership must remain explicit.

We distinguish between decisions an agent can recommend, prepare, execute, or never make autonomously. Examples include committing supply, changing inventory policies, selecting suppliers, releasing production, or responding to regulatory requirements. We examine human-in-the-loop design, review thresholds, traceability, policy enforcement, skill displacement, and exception management, including what happens when an agent is confident but insufficiently grounded.

The objective is to define practical boundaries for autonomy that protect service, compliance, safety, and accountability. Governance is not a final layer added after deployment; it is part of the system architecture and a prerequisite for trustworthy agentic AI.