We study the application of cyber-physical hierarchy on a class of smart grid systems to improve scalability. Our framework employs a multi-agent flocking-based approach to study the transient stability problem in emerging power systems. An agent in this context embodies a coherent group of system generators. We demonstrate how our paradigm conveniently facilitates the identification of coherent machine clusters through spectral bisection of the associated Kron-reduced power system graph. This enables a state-dependent system hierarchy whereby inter-agent interactions are cyber-physical (tier-1) and intra-agent synergies are physical (tier-2). By leveraging this layered perspective, active control can be employed only at a select “lead” generator of each agent; secondary generators that are necessarily coherent to a lead generator will naturally follow suit. Thus this cyber-physical hierarchy improves communications and energy overhead by introducing cyber couplings only within components of the smart grid where physical relationships are insufficient for transient stability in the face of a incidental fault or intentional attack. We demonstrate the performance of our approach on the 9-bus WECC system demonstrating its lower overhead and greater robustness to cyber attacks resulting in information delay.