Users declare desired workloads, networking, storage, and configuration through API objects. Controllers continuously compare desired and observed state, scheduling containers and replacing failed instances. Operational complexity includes cluster security, upgrades, capacity, and application-specific reliability.