In Anthropic's internal research evaluations, an orchestrator model delegating tasks to sub-agents outperformed a single AI agent by approximately 90 percent

Opening sign-in…