Imagine an AI tirelessly working alongside doctors in the ICU, helping to make critical decisions faster and with more information than ever before. That’s the promise of large language models (LLMs) in critical care medicine, a field where seconds can mean the difference between life and death. A new scoping review examines 24 studies exploring this cutting-edge application of AI. From pre-hospital triage to bedside diagnostics, LLMs are showing potential to assist with complex decision-making, streamline medical documentation, and even aid in medical education. One exciting area is diagnosis, where studies show LLMs sometimes outperforming human doctors, particularly in identifying rare or tricky conditions. Imagine an AI flagging a subtle symptom in a patient’s history that a human doctor might miss, leading to earlier, potentially life-saving interventions. Beyond diagnosis, LLMs can help predict patient outcomes, personalize treatment plans, and automate administrative tasks like generating radiology reports. This could free up critical care doctors to focus on what they do best: providing hands-on patient care. But the technology is still young and faces challenges. Like other AI systems, LLMs are prone to “hallucinations,” generating incorrect information. In the high-stakes world of critical care, this could have devastating consequences. Another hurdle is ensuring these AI systems stay up-to-date with the constantly evolving medical knowledge in critical care. Researchers are working to enhance model reliability, ensure fairness, and develop stricter ethical guidelines. For now, human oversight remains crucial, and more comprehensive evaluation standards are needed before LLMs become full-fledged ICU experts. This research provides a crucial snapshot of where the field stands today and where it's headed tomorrow. The vision of an AI-powered ICU is still unfolding, but this review suggests that it could be a game-changer for how we care for the sickest patients.