Routing yes/no decisions to Jev inside Claude Code
The author notes that 31 yes/no questions in one refactor session were billed as full generations, so he switched to Claude writing, Jev deciding, and low-confidence cases falling back to Claude, cutting three decision steps from 13 seconds to 0.83.
your Claude Code agent is burning cash like an idiot and you're the one paying for it one refactor session had 31 yes or no questions, you paid full essay price for every single one Claude writes → Jev decides → low confidence goes back to Claude three decision steps in one PR loop went from 13 seconds to 0.83 everyone screaming 200x never read past the headline, you only see that on boring high volume checks full routing setup and the fallback code are in the article below ↓
This post quotes another post: x.com/cyrilxbt/status/2101515130798297448