AWS shows a multi-agent pattern for market surveillance

AWS has published a reference pattern for a market surveillance agent that uses LangGraph and Strands on AgentCore. The example targets financial services, where systems must review trading activity, investigate suspicious patterns, and produce reports under strict compliance requirements.

The design splits responsibilities between two frameworks. LangGraph manages the larger workflow, including state, graph execution, checkpointing, and human-in-the-loop steps. Strands acts as the reasoning layer inside individual workflow nodes and can work with different large language model providers.

AWS’s point is that production agents need more than a prompt connected to tools. They need state management, recovery from failure, clear handoffs between specialized agents, and a way for humans to intervene when the risk is high.

The example is technical and specific, but the lesson is broader. Regulated industries will judge AI agents by auditability and control as much as by their ability to generate an answer.