A multi-agent system can fail in ways that traditional monitoring methods do not detect. This post introduces a dual-layer monitoring strategy. The first layer utilizes Amazon Bedrock AgentCore Evaluations to continuously assess the quality of the agents. The second layer employs AWS DevOps Agent to autonomously investigate infrastructure issues. This approach was demonstrated on a four-agent airline reservation system. AgentCore Evaluations provide continuous quality scoring, while AWS DevOps Agent performs autonomous infrastructure investigation. This dual approach allows for a more comprehensive understanding of agent system behavior and potential issues in production.



