The introduction of digital technologies in construction site management has shown potential to achieve better tracking of processes. However, these solutions seldom have been custom-built, lacking optimization for intelligent construction sites, that require adaptable systems to respond to changes. One of the main causes of accidents on sites is represented by crew interferences, which never are fully predictable and suffer the emergent behavior of complex domains. So far, AI tools are mostly focused on computer vision to assess risks from image recognition and in the reasoning layer of expert systems, which do not provide spatial navigation capabilities. For the reasons stated above, in this paper AI agents have been extended with 3D reasoning capabilities to detect hazards related to crews displaced in separate workspaces. The detection of spatial interferences has been enabled by the semantic voxelization of jobsite BIM models and analyses of records provided by real-time tracking systems. The approach was tested on a refurbishment project concerning a hospital located in central Italy.