Enterprise scale changes the rules for agentic AI. An autonomous agent without a clear identity and access model is not a smart assistant; it is an attack surface.
Identity first
Every agent needs an explicit identity, a defined access scope and an audit trail. Delegation chains must remain traceable: who acted, on whose behalf and with which mandate.
Least privilege by default
An agent should receive only the permissions it needs, for only as long as it needs them. Escalation belongs in an explicit approval flow, not in a blanket permission granted during prototyping.
Human approval where it matters
Autonomy is useful for reversible routine work. Irreversible operations need a human checkpoint, evidence for the decision and a reliable way to stop the workflow.
The architecture question is therefore not only what an agent can do. It is how the system proves that the agent was allowed to do it.