This paper explores the potential of vision-language models (VLMs) in radiology, focusing on their clinical utility and design considerations. The authors, a multidisciplinary team including human-computer interaction (HCI) researchers, AI researchers, and radiologists, conducted an iterative design process to identify and design clinically relevant VLM applications. They identified four key use cases: Draft Report Generation, Augmented Report Review, Visual Search and Querying, and Patient Imaging History Highlights. These concepts were then evaluated by 13 radiologists and clinicians, who found them valuable but noted several design requirements, such as the need for near-perfect AI performance, workflow integration, and context-specificity. The study highlights the importance of human-centered design in integrating VLM capabilities into radiology workflows and discusses implications for future research and implementation. The authors emphasize the need for responsible AI development and deployment, considering factors like data quality, domain specificity, and societal biases.This paper explores the potential of vision-language models (VLMs) in radiology, focusing on their clinical utility and design considerations. The authors, a multidisciplinary team including human-computer interaction (HCI) researchers, AI researchers, and radiologists, conducted an iterative design process to identify and design clinically relevant VLM applications. They identified four key use cases: Draft Report Generation, Augmented Report Review, Visual Search and Querying, and Patient Imaging History Highlights. These concepts were then evaluated by 13 radiologists and clinicians, who found them valuable but noted several design requirements, such as the need for near-perfect AI performance, workflow integration, and context-specificity. The study highlights the importance of human-centered design in integrating VLM capabilities into radiology workflows and discusses implications for future research and implementation. The authors emphasize the need for responsible AI development and deployment, considering factors like data quality, domain specificity, and societal biases.