ScholarWorks@성균관대학교

초록

Agentic artificial intelligence (AI) holds significant potential for enhancing functional capabilities and effectiveness in domain-specific applications across built environments. While recent studies have primarily focused on the architectural components or technical mechanisms of large language model (LLM)-based AI agents, there remains a lack of comprehensive literature reviews addressing their various application domains, functional roles, and learning approaches within the built environment. Therefore, this study reviews the current landscape of Agentic AI applications in the built environment and proposes a classification structure that encompasses applications, functional roles, and learning approaches. First, this paper examines five representative applications within the built environment. Second, it categorizes the roles of AI agents according to the Data-Information-Knowledge-Wisdom (DIKW) hierarchy, emphasizing their progression from data interpretation to decision support. Finally, this review identifies four core learning approaches adopted by AI agents. Based on this classification framework, this paper defines Agentic Built Environment as virtual assistants embedded with Agentic AI that are capable of providing intelligent services throughout the entire building lifecycle. It also presents the current Level of Development (LoD) of the Agentic Built Environment, identifies existing limitations, and proposes future directions for developing scalable AI agents that support AI-powered services and intelligent decision-making throughout the building lifecycle. © 2025 Elsevier B.V.

키워드