How a Call Works

Understanding these three steps explains almost every behavior you will later observe in testing:

1. Listening — speech is converted to text. The result is text, not sound. That is why unclear pronunciation, strong accents, background noise, and spelling things out are the hardest cases.
2. Thinking — your prompt decides what the agent does with that text: what it asks, what it bases answers on, when it calls a tool, when it transfers, when it ends the call. Everything your agent does or fails to do is decided here.
3. Speaking — the answer is converted to speech. The agent reads out exactly what the text says. +1 415 555 1234 becomes an unintelligible run of digits; Mon–Fri 8–18h becomes gibberish. That’s what Pronunciation is about.

These three steps repeat on every conversational turn. Each takes time — which is why short pauses occur, and why it matters that your agent answers briefly (see Writing the Prompt → Response Rules).

Continue to Strengths & Limitations for what this means in practice.