trying to improve agent replies so I started to build out an internal feedback loop. Simulates a conversation (diff persona) shows me the traces and lets me refine the prompts etc... I think this is the way to rapidly iterate #catch