Yeah, those handoffs are exactly where I’d experiment.

Yeah, those handoffs are exactly where I’d experiment. Give Jev a narrow question with explicit criteria, then let code decide whether to advance, retry, or escalate. Keep tests as hard gates. Which handoff burns the most retries in your setup? I built a tool that handles calls to automated phone systems that our staff would otherwise make manually. I’m adding Jev at the decision points to evaluate the call state against explicit criteria, with code controlling what happens next. That’s where I’m testing whether I can cut down on repeated LLM calls.