Routing with Jev so Opus 5.5 only handles hard decisions
The author puts Jev in front of expensive actions as a decision router, leaving Opus for architecture and deep reasoning, with a copyable prompt and a seven-step setup.
opus 5.5 feels f**king slow send it this prompt and make the whole workflow feel 5x faster i wired jev into the loop so opus stops wasting expensive reasoning on decisions that should take milliseconds. Here's how the setup works: step 1 → map where the model wastes effort. deep research, repeated retries, unnecessary tool calls, spawning agents, and actions that could have been skipped entirely step 2 → put jev before the expensive actions. instead of sending every decision back through opus, build a small state and let jev choose the next path step 3 → keep opus for the hard stuff. architecture, ambiguous problems, deep reasoning, planning, and anything where a wrong decision actually matters step 4 → skip jev for trivial work. quick replies, simple edits, basic math, and obvious actions should stay fast step 5 → run the router in shadow mode first. watch what it would choose before allowing it to control the live workflow step 6 → test it on real behavior. easy question? skip jev. expensive research? call it. repeated failure? change the approach instead of retrying blindly step 7 → verify every important action. the llm generates, jev decides, code executes, verification catches bad routes the result: - less deep reasoning - fewer useless retries - fewer wasted tokens - faster agent loops opus stays the brain. jev decides when the brain actually needs to work. copy the prompt below and try it on your own setup ↓
This post quotes another post: x.com/0xmorlex/status/2101268094387704311