Claude Managed Agents: Outcomes (Rubric-Based Grading Loop) + Multi-Agent Orchestration With Shared Filesystem — Public Beta

Anthropic·high signal

Outcomes lets you write a success rubric; a separate grader evaluates output in its own context window and loops the agent for another pass when criteria aren't met — up to 10.1% quality improvement on file generation tasks. Multi-agent orchestration lets a lead agent delegate to specialists with their own model/prompt/tools, working in parallel on a shared filesystem. Both are public beta as of May 6. Builder takeaway: eval-driven iteration and fan-out/fan-in are now first-class primitives, not custom code.