Anthropic adds 'dreaming' to Claude agents to learn from mistakes

Technology Oct 03, 2026

Anthropic just gave its Claude AI agents a cool new skill: "dreaming."

With this, the AIs can look back at their past conversations, learn from mistakes, and get better at helping out next time.

The update was announced on Wednesday, May 6, 2026.

Claude agents schedule memory reflection sessions

The dreaming feature lets Claude agents schedule reflection sessions to update their memories, either automatically or with your say-so.

Anthropic says this helps the AI spot patterns (like team preferences or recurring issues) and make smarter decisions in the future.

The platform also got upgrades for better teamwork among multiple AIs.

Right now, dreaming is in research preview and available if you request access.