Managed Agents: dreaming, outcomes and multiagent orchestration
Claude Managed Agents add 'dreaming' (scheduled review of past sessions to improve memory), outcome rubrics with a separate grader, and multiagent orchestration.
Memory (public beta 2026-04-23) lets agents learn per session; dreaming curates memory between sessions across agents; outcomes grades work against a written rubric in its own context window; orchestration and webhooks let agents coordinate. Dreaming is a research preview.
- Date
- Wednesday, 6 May 2026
- Lab
- Anthropic
- Kind
- feature
- Access
- closed API
The dates conflict. Platform release notes list orchestration, outcomes and the dreaming preview on 2026-05-06, while the blog post is dated 2026-05-19, and the earlier date is used.
Sources
- claude.com/blog/new-in-claude-managed-agents
- claude.com/blog/claude-managed-agents-memory
- platform.claude.com/docs/en/release-notes/overview
This record was checked against its sources on 6 October 2026. How we check