A landmark study quantitatively benchmarks a multi-agent LLM-based AI system against board-certified clinicians in real-world virtual acute care, revealing near-equivalent diagnostic and therapeutic performance with superior documentation safety and consistency.