Anthropic will let its managed agents dream

Anthropic is upgrading Managed Agents with new memory and coordination tools.

A scheduled “dreaming” process lets agents review past work and update memory over time.

A new outcomes system grades agent output against user-defined criteria, while multi-agent orchestration now supports parallel work.

The features are designed to improve reliability on detailed, subjective workflows with less human oversight; outcomes and orchestration are in public beta, while dreaming has limited access.