Most interactive architecture nowadays prioritises reaction and automation over conversation, thus limiting human-building interaction. This concern, highlighted by architectural theorists including Paul Pangaro, raises the urge to revisit Conversational Architecture under the “Pandemics of today’s AI”. While Large Language Models (LLMs) offer promise, empirical experimentation is underexplored. This paper reviews three case studies of conversational design to discuss its origin, trend and current challenges and introduces a novel framework for designing conversational spaces using an agentic LLM system. It is validated through a case study from the Sustainable Future Competition 2024, focussing on a meditation zone design. The paper concludes with the implications and challenges of Conversational Architecture.