Conversational AI transforms customer service across commerce
LLMAmazon BedrockAmazon Chime SDKAmazon Web ServicesAWS Cloud Development KitAmazon Bedrock AgentCoreAmazon Nova 2 Sonic

AWS builds AI restaurant host using Bedrock AgentCore and Nova 2 Sonic

AWS published a guide for building a voice-ordering system that answers restaurant phone calls, takes orders, and confirms them using Amazon Bedrock AgentCore and Amazon Nova 2 Sonic connected to a restaurant backend via the Model Context Protocol. Restaurants miss an average of 150 calls per location monthly, with 60 percent attempting to place orders or book tables—a direct-to-revenue problem that AI telephony agents can solve without pulling staff off the floor.

AWS published a technical walkthrough for building a voice ordering system that answers phone calls, takes orders from greeting to confirmation, and deploys the full stack using AWS Cloud Development Kit. The system uses AWS Machine Learning Blog (AWS Machine Learning Blog).

The problem is acute: AWS Machine Learning Blog (AWS Machine Learning Blog). For commerce practitioners, this architecture decouples the agent from backend systems via an open standard (MCP), meaning new channels like mobile apps or kiosks can reuse the same ordering logic without rewrites, and the agent supports bidirectional audio with interruption handling in a single stream.

The solution pre-warms the agent session while the phone is still ringing to eliminate dead air, uses Amazon Chime SDK Voice Connector for SIP trunking and toll-free provisioning, and runs the SIP gateway on Amazon ECS Fargate behind a Network Load Balancer. A sample repository on GitHub enables immediate deployment with AWS CDK, seeding sample menu and location data for real orders.

Sources:1 report