Looks like it based on Anthropic’s own multi-agent orchestration research: https://www.anthropic.com/engineering/multi-agent-research-system https://www.anthro

Looks like it based on Anthropic’s own multi-agent orchestration research: https://www.anthropic.com/engineering/multi-agent-research-system https://www.anthropic.com/engineering/multi-agent-research-s... Their findings suggest multi-agent systems result in better performance attributed mostly to token usage (80% of variance).