Anthropic launches dreaming feature for Claude Managed Agents

5 articles · Updated · ZDNet · May 6

Available as a research preview, the update lets developers request access and automatically or manually approve memory changes after agents review past sessions.
Anthropic said dreaming helps agents spot recurring mistakes, shared preferences and converging workflows, while keeping memory high-signal for long-running work and multi-agent orchestration.
The company also expanded outcomes and orchestration tools, building on Managed Agents, launched on 8 April, which Anthropic says speeds agent deployment tenfold.

As AI agents learn to 'dream' on our data, who will ultimately own the valuable memories they create?

Could an AI that 'dreams' develop false memories, making it more dangerously confident than a forgetful one?