Samsung Electronics Migrates LangGraph-Based TV App Data Query Agent to Production with Amazon Bedrock AgentCore

Samsung Electronics' TV Apps Service team built a LangGraph-based natural language query agent to manage complex environments handling app data across more than 200 countries and thousands of app types. However, during the prototyping phase, the team encountered five structural limitations hindering the transition to production: multi-data source management, lack of dynamic scalability, black-box execution, and infrastructure and memory management burdens.

To address these issues, the team adopted a strategy of replacing the operational layer with Amazon Bedrock AgentCore while keeping the existing LangGraph code largely intact. AgentCore provides modular services required for production, such as Runtime, Gateway, Memory, Identity, Observability, Evaluations, and Policy, allowing AWS to handle authentication, routing, state management, observability, and quality control without altering the agent's logic.

Specifically, AgentCore Gateway integrated fragmented data sources, such as validation databases, operational databases, and KPI stores, into a single Model Context Protocol (MCP) endpoint. AgentCore Memory managed multi-turn session states in a managed manner, while AgentCore Observability provided transparent execution tracking through OpenTelemetry-based CloudWatch integration. Additionally, AgentCore Policy enforced Cedar-based read-only policies, and AgentCore Evaluations integrated with CI/CD pipelines to verify quality before deployment.

This approach serves as a case study for achieving production-level performance, security, and governance without being tied to a specific framework, offering a practical architecture and lessons learned for other teams attempting similar agent operational automation.