Migrating a text agent to a voice assistant with Amazon Nova 2 Sonic

Amazon Nova 2 Sonic addresses the fundamental differences between text agents and voice assistants. Text agents deliver paragraphs with tables and links, tolerating multi-second latency with loading indicators. Voice agents require concise, conversational responses delivered in real time with sub-second latency. Nova 2 Sonic supports asynchronous tool calling, allowing the agent to continue speaking while tools run in the background, and can handle multiple tool calls in parallel. It also enables barge-in, where users can interrupt the agent mid-response, and uses voice activity detection (VAD) to manage turn-taking fluidly.

For industries like finance, healthcare, and retail, this migration is critical. A banking agent example shows text returning full account summaries with links, while a voice agent breaks information into digestible chunks with confirmation loops. The key architectural changes include moving from HTTP/REST to bidirectional streaming, implementing ultra-low latency pipelines, and designing responses for listening rather than reading. Amazon provides a sample Skill in the Nova repo that works with AI IDEs like Kiro and Claude Code to automate the conversion process.