Jev handles four decision forks, cutting Opus daily cost from $110 to $4
The author lists the four forks Jev decides, from which notes enter context to which checks run before the full suite, with a confidence floor that sends low-confidence choices back to Opus.
Jev + Claude Opus 5.5 is a f*cking goldmine and almost nobody has wired it up yet. Once Jev takes over the forks they drop from about $110 a day to $4. It's the best way i've found to run Opus so far. The four places Jev decides: -> Which project notes deserve a spot in context before it loads. -> Which easy steps go to a faster worker. -> Which recovery path to take when a tool fails, so there's no blind retry. -> Which targeted checks run before the full suite. Opus keeps the hard reasoning. Jev only chooses between options the harness has already built and checked. Each choice arrives with a probability. Set a confidence floor and anything below it goes back to Opus. Jev engineering fits in one line: the expensive model reasons, the cheap one decides. That's $106 a day on the same work, and it ends up in your performance review. Read the paper first, then the full Jev guide below ↓
This post quotes another post: x.com/polydao/status/2103689373774483815