AWS publishes guide for migrating text agents to voice assistants using Amazon Nova 2 Sonic

AWS published a blog post exploring how to convert a traditional text agent into a conversational voice assistant using Amazon Nova 2 Sonic, addressing design priorities, agent architecture, and tool integration for voice use cases in finance, healthcare, education, social media, and retail.

Voice agents require fundamental architectural changes compared to text agents: voice operates with ultra-low latency requirements (hundreds of milliseconds), uses bidirectional streaming instead of stateless REST, handles real-time speech interruption (barge-in) and turn-taking, and delivers responses as short spoken phrases with confirmation loops rather than formatted text blocks.

Amazon Nova 2 Sonic supports asynchronous tool calling, allowing the conversation to continue naturally while tools run in the background, enabling the agent to accept input, run multiple tools in parallel, and adapt if the user changes their request mid-process.

AWS provides a Skill in the Nova sample repository that works with AI IDEs like Kiro and Claude Code to automatically convert text agents into voice agents, and offers a sample React + WebSocket implementation for building lightweight voice agent web clients.