Ship your first Managed Agent

Claude Managed Agents streamlines the development of production-ready AI agents by abstracting infrastructure, scaling, and complex context management. This harness architecture decouples the agent's logic from its tool execution environment, significantly enhancing security and reducing latency by over 90% for p95 metrics. By utilizing session-based event logs rather than simple request-response cycles, developers gain improved reliability and state persistence. The workshop demonstrates these principles through a site reliability engineering agent capable of autonomously debugging incidents by analyzing logs and metrics. Beyond basic implementation, the platform supports advanced features like sub-agents, memory-based "dreaming" for self-improvement, and secure credential vaults, allowing developers to focus on domain-specific tasks and custom tool logic while Anthropic handles the underlying compute and infrastructure requirements.