Mechanical deformation and atomic transport in metallic glasses are controlled by atomic cooperativity, but the nature of cooperativity has been debated. The leading idea is that some soft region, such as distributed free-volume, defines the cooperativity. I propose an opposite view, that the cooperative resistance to atomic motion, hardness, determines atomic transport. The scale of cooperativity is represented by the coherence length of the medium-range order (MRO). We show that the activation energy for motion in supercooled liquid and liquid fragility are proportional to the coherence volume. Defect-like sites, soft spots, may initiate local atomic movements, but the defect structure disappears during the motion. At the saddle-point of the potential energy landscape the memory of prior thermal history is lost by configurational melting. What determines the outcome is not where the movement started, but how it is resisted by the organization of the system. The work is supported by DOE-BES.