Anthropic introduces 'dreaming' for Claude agents

Anthropic has introduced asynchronous memory consolidation, dubbed dreaming, for Claude Managed Agents, enabling periodic retrospective review of past sessions to distill actionable long-term memory. Targeting coding, finance, and legal workflows, this mirrors neuroscience-inspired replay mechanisms seen in reinforcement learning systems. Unlike stateless context-window approaches or naive RAG pipelines, dreaming offers structured, agent-curated memory persistence across sessions, directly improving multi-session task continuity for enterprise agentic deployments.