Anthropic adds 'dreaming' feature to Claude agents for self-improvement

Technology Oct 06, 2026

Anthropic just gave its Claude AI agents a cool new upgrade called "dreaming."

Now, these AIs can look back at their past sessions, spot patterns (like repeated mistakes or team preferences), and tweak how they behave, all during scheduled reflection breaks.

Some changes happen automatically, while others need a user's thumbs-up.

Developers can request access to 'dreaming'

This dreaming feature helps the AIs learn and remember what really counts, making them better at tasks and teamwork.

For now, developers can try it out by requesting access.

It's another example of Anthropic's creative approach to building smarter (and more thoughtful) AI.