A useful explanation of how much orchestration sits behind an apparently simple phone conversation. Speech recognition, intent handling, business-system lookups, response generation, and text-to-speech all need to work within a very small latency budget. In practice, details such as interruption handling, background noise, confirmation of names or numbers, and a smooth transfer to a human agent are just as important as the language model itself. I would also be interested in how the system monitors failed tool calls and prevents sensitive customer information from entering logs. Those operational details often determine whether an AI voice agent feels genuinely helpful or merely impressive in a demo.