Anthropic's Claude Managed Agents can now "dream," sort of

Anthropic unveiled a new “dreaming” capability for Claude Managed Agents at its Code with Claude conference.

The feature periodically reviews past sessions and memory stores to curate useful memories for later use.

It is available as a research preview, limited to Managed Agents on the Claude Platform.

The update is meant to help long-running agents retain relevant information and reduce context-window loss.