Anthropic has launched several new features for Claude Managed Agents: 'dreaming' (a scheduled process that reviews past sessions to extract patterns and self-improve agent memory), 'outcomes' (a rubric-based grading system where a separate evaluator checks agent output and triggers self-correction), and multiagent orchestration (a lead agent delegates subtasks to specialist subagents running in parallel on a shared filesystem). Webhooks are also now available for async notifications. Real-world users include Harvey (legal drafting, ~6x completion rate improvement), Netflix (parallel log analysis), Spiral (parallel writing with quality enforcement), and Wisedocs (50% faster document reviews).
Table of contents
Build self-improving agents with dreamingDeliver better outcomesHandle complex tasks with multiple agentsWhat teams are buildingSort: