Hi everyone! I’ve been having a great experience with LiveKit lately—the "Agent Insights" beta and the testing kit have been game-changers for my workflow. I’m curious about how you all handle prompt engineering / optimization. There are so many ways to do it (prompt management/versioning tools, evals, or even Automatic Prompt Optimization), and I’d love to hear your perspective: • How do you iterate on your prompts? • Do you rely on customer feedback, synthetic data, or specific APO techniques? • Or just simply do the classic refine and test? Especially when working with Live AI Agent using LiveKit that directly talk to customers.

Classic refine and test for me, its tempting to use the tools available on day one, but those only make sense when you see a clear usecase. I am focusing to cutting down the point of failures as much as i can, i am near launch.