Transforms language models into dynamic multi-agent systems where the model alternates between a Conductor role that breaks down complex problems into subtasks and specialized Expert roles that execute each subtask, enabling structured multi-perspective reasoning and self-critique for robust problem-solving workflows.