An Agentic AI-Enabled Physics-Informed Machine Learning Framework for Grid-Interactive, Decarbonized Building Operations

The urgent need for building decarbonization calls for a fundamental paradigm shift in future autonomous building energy operation, from human-intensive engineering workflows toward intelligence agents that can interact with physics-grounded digital environments. To support this transition, this study proposes an agentic AI-enabled Physics-Informed Machine Learning (PIML) environment for scalable building energy modeling, simulation, control and automation. The proposed framework consists of: (1) a modular and physics-consistent PIML digital environment spanning building, Heating, Ventilation, and Air Conditioning (HVAC), distributed energy resources (DER) to support grid-interactive energy management; and (2) an agentic AI layer implemented with 11 specialist agents and 72 Model Context Protocol (MCP) tools to enable end-to-end execution of multi-step energy analytics. A representative case study demonstrates multi-domain, multi-agent coordination in assessing how system retrofits and control upgrades affect energy use, operation cost, thermal comfort, and flexibility in a 20-building residential cluster. Furthermore, a large-scale benchmark (6,156 runs) is conducted to systematically evaluate workflow performance in terms of accuracy (planning, agent selection, tool selection, and parameter extraction), token consumption, execution time, and inference cost. The benchmark results quantify the impacts of intelligence mode design, model-size configuration, task complexity, and orchestrator–specialist coordination on overall performance. Four key lessons learned and six limitations with corresponding future research directions are identified to inform the development of reliable agentic AI systems for real-world building energy applications. This work establishes a scalable, physics-grounded foundation for deploying agentic AI in decarbonized and grid-interactive building energy operations, and highlights key research directions toward adaptive self-driving building intelligences.