Anthropic unveils 'Dreaming' feature... AI can retrace its past work on its own

Anthropic has added a new “dreaming” feature to Claude managed agents that reviews past work and interactions to store only useful memories.

The feature is designed to analyze recurring mistakes and patterns, helping improve long-term agent performance over time.

It is currently available only as a limited research preview.

Anthropic also widened access to the Outcome and multi-agent orchestration features in the managed agents beta.