Anthropic Gives Claude Agents Ability to Dream

Anthropic has introduced a dreaming capability for its Claude Managed Agents, allowing them to review past sessions, extract patterns, and store useful memories to improve future performance. Unlike standard context compaction, dreaming works across multiple agents, helping them identify recurring mistakes and shared preferences over time. The company also made two previously preview features widely available: outcomes, which uses example-based grading to improve task success by up to 10 points, and multi-agent orchestration, which lets a lead agent delegate subtasks to sub-agents. Pro and Max subscribers also received a doubled usage limit, now 10 hours.