Smart energy management systems (EMS) are entering a phase of rapid transformation. Artificial intelligence (AI)—including machine learning (ML)

deep learning (DL)

and reinforcement learning (RL)—has become the computational backbone for real-time forecasting

scheduling

and control of renewable-rich power systems. Yet the long-term efficiency

resilience

and social acceptance of these systems depend critically on human-AI synergy: well-structured roles for people to supervise

override

and enrich AI decisions. This review integrates recent literature on AI-powered EMS for smart grids

buildings

microgrids

and isolated hybrid systems with emerging human-in-the-loop (HiTL) paradigms that explicitly incorporate occupants

operators

and policy makers. It offers an evaluation of data-driven forecasting

adaptive optimisation

and edge intelligence

and highlights research gaps in transparency

interoperability

and co-optimisation of technical and human objectives. We show that well-designed human–AI collaboration improves not only energy efficiency and renewable integration but also the robustness and trustworthiness of future energy systems.