Building an Amazon Connect Customer AI Agent with a 1-Hour Conversation [AICC Builder Series #1]

When considering the introduction of an AI agent for a customer center, it's often unclear 'which service to use' or 'how to train it on business rules,' which frequently delays getting started at all. AICC Builder is an AWS open-source sample web application designed to solve this problem: it specifies requirements through about a 1-hour Korean-language interview and automatically generates the 6 types of deployment assets needed for an Amazon Connect Customer AI Agent PoC.

The overall architecture of the AI agent consists of five layers. Amazon Connect Customer receives calls and chats, and at the voice/STT layer you choose a provider such as Amazon Polly, Nova Sonic V2, Deepgram, or ElevenLabs. The Connect Customer AI Agent understands the conversation using a selected model such as Claude Haiku or Sonnet, and autonomously invokes tools via MCP. Amazon Bedrock AgentCore Gateway connects business tools defined in OpenAPI through the MCP Gateway, while the actual business processing is handled by a backend such as Lambda, RDS, and DynamoDB.

The 6 types of assets that AICC Builder generates are as follows:

• System Prompt: defines the AI Agent's role, operating rules, and response tone
• Knowledge Base documents: FAQs and reference information
• OpenAPI Spec: definitions of business tools
• Connect Flow: inbound call handling flow
• Lambda functions: sample code implementing the tools
• CloudFormation template: script for deploying the entire infrastructure

Users simply describe their business in Korean and provide their company website or existing documents. Since the tool fills in technical choices with recommended values, you can verify actual operation over a real phone call within half a day. You can follow the entire process through a demo of a delivery-tracking inquiry for the example online retail company 'AnyCompany.'