Learning-in-the-Loop Power System Control integrates data-driven learning models directly within the operational control loops of modern power grids. Instead of treating AI as an external optimization layer, this approach embeds learning agents that continuously adapt to real-time measurements, system uncertainties, and dynamic operating conditions. By coupling deep reinforcement learning, online optimization, and model-based control, this research enables intelligent, safe, and interpretable decision-making for voltage regulation, frequency stabilization, and energy management across distributed resources.