Claude's agent has a new feature called 'Dreaming' that allows for self-improvement.

Anthropic's AI agent infrastructure, 'Claude Managed Agents,' has added a 'dreaming' feature that improves agent performance and reduces memory consumption.

New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration | Claude

Live from Code with Claude: we're launching dreaming in Claude Managed Agents as a research preview.

Outcomes, multiagent orchestration, and webhooks are now in public beta.

pic.twitter.com/p4DFRzFEd8

— Claude (@claudeai) May 6, 2026

Dreaming is a feature that reviews AI agent sessions, extracts patterns, and improves them over time. This allows the AI agent to handle complex tasks with minimal instructions while reducing memory consumption.

It can uncover patterns that a single agent cannot capture, such as recurring errors, convergent workflows among agents, and shared preferences across the team, making it particularly useful for long-term projects and multi-agent orchestration.

Users can decide how much control they want, and they can choose to have the memory updated automatically or have a human review it before the changes are reflected.

Dreaming is available as a research preview version on Claude Managed Agents, and developers can request access from the official website.