Anthropic Introduces 'Dreaming' and 'Outcomes' to Enhance Claude Managed Agents' Self-Improvement and Goal-Setting

Anthropic has expanded its Claude Managed Agents platform with three new capabilities: dreaming, outcomes, and multiagent orchestration. Dreaming, currently in research preview, extends Claude’s memory by periodically reviewing past agent sessions and memory stores to extract patterns and curate memories. This scheduled process helps agents self-improve over time, either automatically or with human review. Outcomes, meanwhile, lets users define a rubric for what constitutes a successful result. A separate grader evaluates the agent’s output against those criteria in its own context window, ensuring unbiased assessment. When something falls short, the grader pinpoints what needs to change and the agent retries. Users can set an outcome and receive a webhook notification when the agent finishes, streamlining task completion and quality assurance.

The third feature, multiagent orchestration, allows a lead agent to decompose a complex job into sub-tasks and delegate each to specialist agents with their own models, prompts, and tools. These sub-agents work in parallel on a shared filesystem and contribute to the lead agent’s overall context. Because events persist and every agent remembers its actions, the lead can check in mid-workflow. For example, Netflix’s platform team already uses multiagent orchestration to run investigations where sub-agents fan out through deploy history, error logs, metrics, and support tickets simultaneously. Together, these features make Claude Managed Agents more autonomous, adaptable, and capable of handling sophisticated enterprise workflows without extensive manual oversight.