Large language models are increasingly used in clinical practice and are evolving from information retrieval tools towards agentic systems that support complex decision-making. Although key challenges remain, these models have the potential to reshape diagnostic workflows, improve clinical efficiency and reduce health inequities.