Opus thinks, Jev decides: a cheaper agent architecture

Your AI coding agent doesn't need a frontier model to decide which file to load. Yet that's exactly what you're paying for. Every time your agent picks project notes, routes a task, recovers from a tool error, or chooses which tests to run, it may be burning expensive reasoning tokens on a multiple-choice question. There's a better architecture. Opus 5.5 thinks. Jev decides. Your code stays in control. Here's how to cut the waste: → Use Jev to select relevant context instead of loading eve