Jev and Opus 5.5 split the agent decision loop
Jev handles only filtering invalid routes, choosing providers, picking one of four steps and rechecking stale decisions, while Opus focuses on reasoning and coding, so each decision can be logged and replayed.
jev + opus 5.5 turns 4 agent decisions into a much cleaner loop. instead of letting one model control every branch, jev only handles 2 narrow jobs while the host keeps the hard limits. → filters invalid routes before selection → chooses from real provider/model candidates → picks 1 of 4 steps: inspect / implement / verify / answer → rechecks stale decisions before execution opus stays focused on reasoning, coding and synthesis. jev only decides what happens next inside options the app can actually verify. each decision can then be logged, replayed and compared later, turning 1 opaque agent loop into something you can actually inspect.
This post quotes another post: “A complete guide to getting started building with Jev”