Toward a Human–AI Collaborative Workflow in Urban Design: Insights From the Dongyang City International Design Competition (2023–2025)

Source abstract

Generative artificial intelligence (AI) is reshaping the workflow of urban design. This study aims to explore and validate a human–AI collaborative framework that aligns generative AI capabilities with the practical demands of urban design, thereby promoting the efficient and reliable integration of AI into real-world projects. To this end, we propose a three-phase collaboration framework—progressive AI agenting, co-creative interaction, and responsibility-based governance—and conducted a three-year longitudinal application within the Dongyang International Urban Design Project (2023–2025). By embedding GPT and image generation models into key stages such as research analysis, concept development, and design representation, the collaborative workflow significantly enhanced design efficiency and expressive capacity. Meanwhile, we established three governance mechanisms—expert pre-review, public feedback, and prompt traceability—to address challenges related to responsibility attribution, contextual reasoning, and ethical oversight. Through this iterative and practice-oriented research process, the study contributes methodological insights and practical reflections for developing scalable, adaptive, and accountable human–AI collaboration models in future urban design.