Anthropic introduces Managed Agents on Claude, a managed execution layer for agent-based workflows. It separates agent logic from runtime concerns like orchestration, sandboxing, state management, and credentials. The system supports long-running multi-step workflows with external tools, error recovery, and session continuity via a meta-harness architecture. By Leela Kumili