AWS has published a technical walkthrough for building a production-style multi-agent system for market surveillance. The example combines LangGraph for workflow orchestration, Strands for agent reasoning, and Amazon Bedrock AgentCore for deployment, memory, and observability.
The financial-services scenario is deliberately complex: surveillance systems need to coordinate specialized agents, analyze trading patterns, investigate signals, and recover from failures. AWS uses LangGraph to manage state and directed workflows, including checkpoints and human-in-the-loop interactions. Strands handles reasoning inside individual nodes, while AgentCore supplies managed infrastructure.
The post is a reference architecture rather than a finished compliance product. Its value is in showing how enterprise agent systems are being assembled from multiple layers: orchestration, model-agnostic reasoning, state management, and production monitoring. That stack is becoming important as companies move from simple chatbots to agents that must run reliably across multi-step business processes.