Jev in practice: typed decisions, scoped authority

A team built safe-upgrade, an AI agent that handles dependency upgrades by separating three distinct concerns: Jev makes typed probabilistic decisions from repository evidence, LangGraph manages the workflow through those decisions durably, and Tenuo ensures each resulting action has appropriate authority scoping. The design demonstrates how to build safe agent delegation by keeping judgment (decision-making), control flow (workflow execution), and authority (action permissions) separate rather than entangled.