Anthropic rolls out 'dreaming' feature for Claude managed agents
Anthropic just rolled out a "dreaming" feature for its Claude Managed Agents, letting these AIs look back at their past chats and learn from them, kind of like reflecting after a long day.
This upgrade, announced on May 6, 2026, builds on April's launch that made setting up these agents way easier with a suite of APIs.
Anthropic's 'dreaming' updates agent memories
With "dreaming," agents can now update their own memories or let developers approve incoming changes automatically, making them better at what they do over time.
Anthropic also improved how these AIs work together and stay accurate.
The company is big on AI safety too: they've studied things like Claude's sense of right and wrong, and even its emotional patterns.
Right now, "dreaming" is in research preview mode; developers need to request access if they want to try it out.