<p>Robotic deconstruction offers a precise and environmentally responsible alternative to conventional demolition, enabling the selective recovery of building materials for reuse. This research presents a methodology for robotic deconstruction enabled by Spatial Artificial Intelligence (Spatial AI), demonstrated through the case of brickwork. The approach comprises: (1) a deep learning-based real-time object perception system, trained on synthetic photorealistic data to detect and localize individual bricks; (2) the incremental registration and spatial mapping of these discrete elements within an evolving as-built digital model; and (3) reasoning and control routines that enable perception- and mapping-informed, stepwise robotic deconstruction. This methodology was validated in two progressively complex case studies involving the deconstruction of dry-stacked and mortar-bound brickwork structures with unknown geometries. A mobile robot equipped with an RGB-D camera, gripper, and drill enabled the perception and recovery of individual bricks. In the first dry-stacked case study, we achieved first-attempt success rates of 100% for picking and placement, and in the second, more complex mortar-bound case study, 82% for mortar joint breakage, 94% for picking, and 100% for detachment and placement. Across both case studies, the system achieved complete robotic deconstruction and material recovery, with all bricks estimated within manipulation-relevant tolerances and registered in incrementally generated digital models of the deconstructed structures. These results demonstrate the feasibility of the proposed system across two prototype deconstruction scenarios and indicate its potential transferability to robotic recovery workflows for other classes of discrete building elements.</p>