Enabling spatial intelligence in the AEC lifecycle: a framework and review of large language models

Abstract The Architecture, Engineering, and Construction (AEC) industry is undergoing a profound digital transformation driven by the urgent need to integrate fragmented, heterogeneous data into actionable intelligence. This paper provides a systematic review of 76 representative studies to examine how Large Language Models (LLMs) enable spatial intelligence across the entire building lifecycle. A layered spatial intelligence framework is proposed, encompassing the dimensions of design, construction, operation, as well as research and education. The review demonstrates that LLMs are evolving from passive text processors into active cognitive partners capable of semantic interpretation, spatial reasoning, and complex decision-making. Key findings highlight the effectiveness of LLMs in requirement translation, multimodal design generation, regulatory compliance, and lifecycle Building Information Modeling (BIM) integration. Despite these advancements, significant challenges persist regarding spatial reasoning accuracy, the lack of physical embodiment, and institutional data governance. This research outlines a strategic roadmap toward embodied intelligence, AEC-specific world models, and enhanced human-AI collaboration, providing a structured foundation for the advancement of LLM-driven spatial intelligence in professional practice.