A single AI agent requires a harness—a control layer managing context, tools, execution, and verification. However, an enterprise fleet of agents needs an operating system to manage contention and resource allocation across multiple agents competing for shared resources, fundamentally different from scaling up a single harness.