Jev + Opus 5.5 workflow: six rules that cut AI bills by 95%
Jev picks the next action, code catches repeated work, and Opus 5.5 handles tasks needing deeper investigation, routed through six steps from prepare to measure.
Boris Cherny, Claude Code creator: “we cut our AI bill by 95% and started getting through the work faster most of the money was going into repeated decisions and unnecessary retries, so we changed when the agent runs and what it needs before starting another attempt” that control lives in the workflow: Jev selects the next action, code catches repeats, and Opus 5.5 receives the tasks that need deeper investigation i packed the setup into 6 rules you can give your agent prepare → reuse → route → execute → check → measure start with one recurring task and define what the agent should execute, skip, or send for human review before another paid run begins completed work returns its saved result when the inputs haven’t changed, keeping the same request from triggering the same investigation twice failed checks come back with the exact requirement, relevant evidence, and one targeted correction instead of another full attempt before buying another agent course, give your current setup clear rules for when to run, retry, and reuse finished work the article below walks through the Jev + Opus 5.5 setup and the budget behind a 20x more efficient workflow ↓
This post quotes another post: x.com/0xnineinch/status/2108604801411879139