Anthropic has introduced a new “dreaming” feature for Claude Managed Agents, allowing AI agents to review past work sessions and improve over time. The system can identify recurring mistakes, refine workflows and organize memory between tasks, helping agents perform better in long-running projects.
The company also announced new tools like “Outcomes,” which checks whether an AI successfully completed a task and multiagent orchestration, where a lead AI delegates work to specialized agents.
Anthropic says companies including Netflix and Wisedocs are already using these tools, with Wisedocs reporting document reviews that are 50% faster. The dreaming feature is currently available as a research preview.
Source: Android Authority
