Migrate agentic workloads to Amazon Bedrock AgentCore

This article provides a three-stage migration guide for moving agentic workloads from LangGraph to Amazon Bedrock AgentCore, progressively shifting operational burdens from user-managed infrastructure to AWS-managed services.

• Stage 1: Migrate agent runtime to managed compute, publish tools via Gateway, and move conversation state to AgentCore Memory
• Stage 2: Replace hand-written routing logic with model-driven planning using Strands Agents and Cedar policy enforcement
• Stage 3: Hand the entire agent loop to AgentCore harness for fully managed orchestration
• Each stage reuses previous infrastructure; teams can stop at stage 1 or 2 depending on needs
• Sample repository demonstrates migration with a LangGraph customer support agent, showing code changes and testing approach
• Common pitfalls include assuming feature parity, holding state in-process, and overlooking authentication architecture

The migration reduces operational overhead by moving OS patching, session isolation, tool authentication, and state management to AWS while preserving agent behavior and allowing incremental adoption of AgentCore services.