Allows agents to review past interactions and improve memory.
At its Code with Claude developers’ conference, Anthropic unveiled a new feature called ‘dreaming,’ enabling managed agents to periodically review their interactions and curate important memories for future use. This update aims to enhance the capabilities of multi-agent systems in handling complex, long-term projects by ensuring context is retained more effectively.
Managed Agents are pre-built, configurable agent harnesses designed to facilitate collaborative work among multiple AI-driven entities on a specific task or project over extended periods. The ‘dreaming’ feature addresses the limitations of context windows typically found in large language models (LLMs) and aims to maintain high signal-to-noise ratios as agents interact with each other and their environment.
For builders, operators, and enterprise teams, this new capability is crucial for managing complex AI-driven workflows where multiple agents need to coordinate effectively. By surfacing recurring mistakes, shared preferences, and optimal workflows across a team of agents, ‘dreaming’ enhances the overall performance and efficiency of such systems.
Currently in research preview, developers can request access to the ‘dreaming’ feature through the Claude Platform. Anthropic is also doubling rate limits for Pro and Max subscription plans to address user demand and improve service reliability.
What matters
- Anthropic introduced a ‘dreaming’ feature for Claude Managed Agents.
- Improves context retention in long-running projects with multiple agents.
- Feature now available via research preview on the platform.
Why it matters
Feature now available via research preview on the platform.
This GenAI News article was prepared in original wording using reporting and materials published by Ars Technica. Source reference: https://arstechnica.com/ai/2026/05/anthropics-claude-can-now-dream-sort-of/.
Drafted by the GenAI News review pipeline.