Conversational Application of Agentic Multimodal AI in Collaborative Architectural Design Environment: An architectural-focus AI design partner for early-stage design exploration

Nowadays, most AI-assisted architectural design processes are limited to one-to-one interaction and one-directional design processes and require a long learning curve for applying sophisticatedly designed AI applications. Despite agentic AI systems showing promising potential in enhancing human-AI collaboration in creative scenarios, they are underexplored in architectural design. This paper proposes a conversational design framework for early-stage architectural design exploration, with a developed architectural-focus agentic multimodal AI system deployed on WeChat. This common messaging application allows group messaging as a platform for collaborative design. A case study is presented by a mixed-use renovation development in Suzhou; tutors and three undergraduate architecture students joined a WeChat group with the AI bot as a design team. Each student developed a design proposal, and three AI collaboration modes were observed, including performance-based, evaluation-based and designerly-based. The proposed framework and agentic AI tool enable designers to dedicate themselves to the design process with minimum learning curves. Collaborative environment integration allows more sophisticated designers, such as tutors, to enhance design conversation by asking follow-up questions. Students displayed mutual-learning AI-collaborated design methods by observing others' conversations in group chat. With the observed challenges and opportunities, the implications of research, AI education, and real-world practice are discussed.