Anthropic unveiled a new dreaming capability for its Claude Managed Agents during the Code with Claude developers conference in San Francisco. The feature allows agents to review recent sessions and store key patterns in memory for future tasks. The company also plans to expand access to other tools and increase usage limits for subscribers.
The dreaming process runs on a schedule and analyzes sessions across multiple agents to identify recurring patterns, mistakes, and shared preferences. Users can opt for automatic memory updates or review proposed changes themselves. This approach helps overcome limits in context windows during extended projects involving several agents working together over hours or days.
