Anthropic Upgrades Claude Managed Agents with Dreaming

Anthropic announced significant upgrades to its Claude Managed Agents on Wednesday at the Code with Claude developer conference in San Francisco, introducing a new "dreaming" feature alongside expanded access to outcomes and multiagent orchestration.

The dreaming capability, now available in research preview, enables agents to conduct scheduled background reviews of past sessions, including conversations, logged memory, and completed tasks. This process identifies patterns such as recurring mistakes, preferred workflows, and shared preferences across multiple agents. Developers can configure dreaming to update agent memory automatically or require manual approval of changes before implementation. "Dreaming surfaces patterns that a single agent can't see on its own, including recurring mistakes, workflows that agents converge on, and preferences shared across a team," Anthropic stated. The company noted that dreaming restructures memory to maintain high relevance over time, making it particularly valuable for long-running projects and multiagent setups.

Complementing dreaming, outcomes allows developers to define specific standards or examples of successful outputs for tasks. A...