Claude Agents Get Dreaming, Multiagent Orchestration Features

Anthropic is launching dreaming, outcomes, multiagent orchestration, and webhooks for Claude Managed Agents. Dreaming reviews past sessions to extract patterns and refine agent memory over time, while outcomes lets developers define success rubrics so agents self-correct without human review. In testing, outcomes improved task success by up to 10 points. Multiagent orchestration allows a lead agent to delegate complex tasks to specialist subagents working in parallel. Early adopters include Harvey, which saw 6x completion rate improvements, Netflix, which uses it to analyze build logs at scale, and Spiral by Every, which uses it to parallelize writing drafts scored against editorial rubrics.