Claude's new "Dreaming" feature is designed to let AI agents learn from their mistakes
Anthropic just shipped 'Dreaming'—an async process that lets Claude agents learn from past sessions without human intervention.
Why it matters
Anthropic is moving beyond single-session agents to multi-session learning systems. Combined with Outcomes and Multiagent Orchestration (now in public beta), Claude agents can now persist knowledge, reduce redundant processing, and improve autonomously—a competitive move against OpenAI's agent roadmap.
The key facts
7 to knowDreaming feature: asynchronous review of past agent sessions
Automatic memory cleanup: deduplicates and removes outdated entries
Insight distillation: extracts learnings across sessions
Outcomes feature: in public beta
Multiagent Orchestration: in public beta
Platform: Claude Managed Agents
Use case: autonomous agent improvement without human intervention
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Anthropic is adding "Dreaming" to Claude Managed Agents, an asynchronous process that reviews past agent sessions, cleans up duplicate or outdated memory entries, and distills new insights. Along with Outcomes and Multiagent Orchestration, now both in public beta, the updates are designed to help…