AWS published a guide to building an AI voice agent for phone orders in restaurants
AWS described the architecture of an AI voice agent that takes restaurant orders over the phone without an app or login, using Amazon Connect Agentic Voice and Amazon Bedrock AgentCore.
AWS published a guide on its blog to building a voice ordering system for restaurants. The caller dials a phone number, an AI host greets them, answers questions about the menu, finds the nearest pickup location and confirms the order aloud – without an app, website or login. The solution is intended to handle situations where staff at the counter process phone orders while also serving customers on site.
The architecture uses Amazon Connect for the phone channel, Amazon Lex V2 with Amazon Connect Agentic Voice for real-time speech recognition and synthesis, and an Amazon Connect AI agent that manages the conversation. The agent connects to the restaurant backend via Model Context Protocol (MCP) through AgentCore Gateway, part of Amazon Bedrock AgentCore. According to AWS, this separates the ordering logic from the phone channel, allowing the backend to be changed without modifying the agent. According to the description, the agent uses a system prompt built on the Anthropic Claude Haiku 4.5 model and has a safety guardrail attached to check content.
The backend in the guide consists of an Amazon DynamoDB database (customers, orders, menu, carts, branches), Amazon Location Service for addresses and routes, the AWS Lambda compute service for business logic, and Amazon API Gateway with authorization via IAM. During deployment, AgentCore Gateway loads the OpenAPI schema of the REST API and registers each endpoint as a named MCP tool, verifying access using custom JWT tokens validated against the Amazon Connect instance.
The source text describes the architecture and deployment process; it is not a generally available product with its own price. Details can be found in the source article.
Why it matters
For developers, this is a ready-made reference architecture for connecting a phone channel, speech recognition and an AI agent to an existing backend via the open MCP protocol, without having to build the individual layers from scratch. For restaurants and similar businesses, the described solution offers a way to automate the handling of phone orders without requiring an app or website and reduce staff workload during peak hours, without losing customers who prefer a phone call.
Two audiences, two different impacts
What this means
For individuals
Developers can use the published reference architecture as a template for building their own AI voice agents connected to an existing backend via the MCP protocol and Amazon Bedrock AgentCore.
For a business
Restaurants and similar businesses can use this architecture to automate phone orders without requiring an app, which may reduce the workload of counter staff during peak hours and reduce missed calls.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.