Building a restaurant telephony AI host with Amazon Connect
Amazon Web Services Sergio Barraza
AWS shows a phone-ordering AI host for restaurants. It answers calls, takes orders, and skips the app-and-login dance.
Based on reporting by Amazon Web Services, Sergio Barraza — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Restaurants still lose a lot of orders to the phone. That sounds quaint until you picture the reality: a staff member juggling the counter, a caller on hold, and a handwritten order that has to survive the rush. AWS’s answer is a voice system that picks up the phone itself, greets the caller, answers menu questions, finds a nearby pickup spot, and confirms the order out loud.
The setup is built around Amazon Connect, but the point is separation. Amazon Connect handles the call, an Amazon Connect AI agent handles the conversation, and the restaurant backend keeps the menu, carts, orders, and locations. The caller never needs an app or website. The system identifies them by phone number, not by login, and the backend is exposed through MCP tools so the agent can call it by name.
Amazon Lex V2 provides the speech layer, with Amazon Connect Agentic Voice doing the listening and speaking. Amazon Connect Customer AI agents run the orchestration, powered here by Anthropic Claude Haiku 4.5 in Amazon Bedrock. Amazon Connect AI Guardrails keep the conversation from wandering off or getting ugly. Under the hood, AWS Lambda does the business logic, Amazon DynamoDB stores the data, Amazon Location Service handles geocoding and route calculation, and API Gateway fronts the backend with IAM.
AWS says the whole thing is deployed with CDK in eight stacks. It also says the contact flow captures the caller’s phone number, opens the AI agent session, pushes that number into the session, and then hands the caller to Lex V2 for the live voice interaction. The idea is simple enough: let the phone call stay a phone call, but give it a brain.
There’s a practical edge to this that a lot of “AI ordering” demos miss. The telephony channel is the product here, not a browser clone with a voice skin. And because the backend is kept separate from the channel, the restaurant logic can change without ripping up the conversation layer. That’s the part worth paying attention to, not the shiny robot-host vibe.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI automation: boring, specific, and aimed at a place where phones still matter. The industry keeps trying to make every customer behave like a web user; this does the opposite and meets people where they already are. Also, the fact that the caller is identified by phone number instead of another login to forget is just common sense, which is why it feels almost rebellious.
Read more about this at: Amazon Web Services