Anthropic Ships Claude Managed Agents 'Dreaming' Memory Consolidation
Anthropic has released a "dreaming" feature for its Claude Managed Agents product, now in research preview. The dreaming phase is a scheduled overnight process that reviews recent agent work and updates the agent's long-term memory — implementing off-cycle consolidation rather than relying entirely on in-context retrieval at call time. The same update also adds multi-agent orchestration (spawning additional agent sessions at runtime) and an "outcomes" mode (goal + graded loop + acceptable-outcome criteria). Reported by The New Stack and confirmed across multiple sources.
Why It Matters
This is the first time a leading AI lab has shipped productized off-task memory consolidation for agents — a capability the research community has been theorizing about under names like "neocortical analog" consolidation. Combined with multi-agent orchestration, it materially advances what a managed agent can sustain across sessions without human re-briefing.