Summary
In this module, you explored how voice agents listen, reason, and speak. The Contoso returns assistant provided a focused example for creating an agent, configuring its behavior, and testing a spoken conversation in the Foundry playground.
Key takeaways
- Start with a focused task: A sample policy conversation helps you test voice behavior without an order system or telephone connection.
- Write for listening: Use short replies, ask one question at a time, and define what the agent must do when information is missing.
- Separate advice from actions: An agent shouldn't claim to check an order or issue a refund without tools that perform those actions.
- Test with audio: Spoken requests reveal problems with pauses, interruptions, and clarity that typed questions don't expose.
- Make targeted changes: Match the symptom to the relevant setting, save the agent, and repeat the same test.