AI Operations · Glossary
Persistent Policy
A Persistent Policy is a Policy defined by scope: it applies across all Routines for an Interactive Agent, agent-wide rather than scoped to any single Routine or Routine step.
Persistent Policies are defined by scope. The scope is agent-wide. Scope is the defining axis of this class, independent of cadence. A Persistent Policy can also be Continuous (evaluated every turn), or one-shot at session entry, or routine-step gated. Persistent (scope) and Continuous (cadence) are orthogonal. A Persistent Policy can also be Continuous. A Routine-Scoped Policy can also be Continuous. The scope axis and the cadence axis are independent. Persistent Policies express rules that hold everywhere the agent operates: regulatory disclosures, brand voice constraints, escalation thresholds, identity and authentication requirements. The class is distinct from Routine-Scoped Policies (scoped to one Routine). Persistent Policies are typically High-criticality, executed under Deterministic Enforcement. They are the floor of agent behaviour.