This paper presents Grid-Agent, an autonomous AI-based framework for addressing the increasing complexity of modern power grids driven by distributed energy resources (DERs), electric vehicles (EVs), extreme weather conditions, and cyberattacks. Grid-Agent leverages large-scale language models (LLMs) within a multi-agent system to detect and correct violations. A planning agent generates coordinated action sequences using a power flow interpreter, and a verification agent ensures stability and safety through sandbox execution with a rollback mechanism, integrating semantic reasoning and numerical accuracy. To enhance scalability, we utilize an adaptive multi-scale network representation that dynamically adjusts encoding schemes based on system size and complexity. Violation resolution is achieved through optimization of switch configuration, battery placement, and load shedding. Experiments on IEEE and CIGRE benchmark networks, including the IEEE 69-bus, CIGRE MV, and IEEE 30-bus test systems, demonstrate excellent mitigation performance, highlighting Grid-Agent's suitability for the rapid, adaptive response required by modern smart grids.
