Before you hand an agent to a client, how do you check that the bookings and lookups actually land correctly? Scripted test calls by hand, an automated caller running scenarios, or just watching logs for the first week? Curious what catches the most problems for the least time.

Hey all, I found this group through the voice AI side of Skool and came for the template library and the Claude Code builds. I'm mostly interested in what happens after the build: testing agents before they go live and keeping them reliable once real callers hit them. For those of you running agents for clients right now, are you still on Retell or Vapi, or has anyone actually moved a client onto a self-hosted Pipecat or LiveKit setup?

For anyone who moved a live client agent from the classic speech-to-text, LLM, text-to-speech setup to a speech-to-speech model like GPT-Live-1: did bookings and CRM lookups get more or less reliable? The voice sounds better, but I'm curious if the actions keep up.