<p>Architectural design has long relied on 2D construction drawings as mediators between design ideation and construction, translating intent into material action. As robots increasingly join humans on building sites, these established communicative practices face epistemic tension: humans interpret drawings through tacit and experiential knowledge, whereas robots require explicit, procedural, and unambiguous instructions. This research investigates how architectural design intent, as represented through drawings, can be effectively communicated to facilitate shared interpretation and collaboration between human and robotic agents. We propose an integrated dual-model approach that distinguishes between what to build (spatial–topological) and how to build it (procedural–topological). Building on this, we develop a speculative framework that automates the generation of these models from construction drawings using multimodal AI, fostering a collaborative human–robot communication in architectural construction. We present initial examples and results of this workflow and discuss their implications for future human–robot collaborative practices.</p>