Existing AI benchmarks fail to measure the growing capabilities of multi-agent systems. New metrics are needed as these systems begin to reshape society.