AI Operations · Glossary
Priority
A Priority is the authored ordering principle the agent applies when Policies conflict on a decision or when the Universal Router selects between candidate models. It is a routing and conflict-resolution control, not one of the authored components.
Priorities are not heuristics, and they are not a content type. A Priority is authored by a domain expert and applied deterministically at runtime across two surfaces. Inside the agent, Priorities resolve conflicts between Policies that fire on the same decision: when a customer-experience Policy and a regulatory Policy disagree, the authored Priority decides. Inside the Universal Router, Priorities order the model-selection dimensions of cost, latency, capability, and jurisdiction, with the operator authoring which dimension dominates for which class of decision. Priorities are inspectable per decision: the production trace records which Priority resolved which conflict. They make the agent's behaviour intentional under contention. Priorities are configured as relationships among the authored components rather than authored as a component themselves.