Anthropic introduces 'dreaming' for Claude agents to learn from chats

Technology Oct 02, 2026

Anthropic just introduced a "dreaming" feature for its Claude AI agents, letting them learn from their own past chats.

Basically, these agents can now look back at conversations, spot patterns, and get better at their jobs over time.

The feature is in research preview right now, so only select developers can try it out.

Claude agents gain self-improvement, coordination upgrades

With "dreaming," Claude agents can self-improve, either automatically or with user approval, making them more accurate and helpful as they go.

Anthropic has also upgraded how these AIs work together and stay on task, all part of its push to build safer and smarter AI tools for everyone.