A few years ago, Prof. Shafi Ahmed had the opportunity to visit hospitals across Shanghai, Beijing, Guangzhou and Chengdu. What struck him was not simply their scale, although many were enormous, but the pace at which technology was being embedded into everyday clinical practice. Digital infrastructure, automation and artificial intelligence were not pilot projects hidden away in innovation centres; they were becoming part of the fabric of healthcare.
There is a bigger transformation on the near horizon. For much of the past two decades, healthcare has been focused on digitisation. We replaced paper notes with electronic health records, imaging has become digital, and the operating theatres are filled with robots. More recently, artificial intelligence has begun writing clinic letters, interpreting scans and helping clinicians document consultations.
These are important advances, but they share one characteristic: AI has largely been used as a tool and performs a specific task when asked.
From Application to Operating System
China is now pursuing something much more ambitious.
Across several leading academic centres and hospitals, a new model is beginning to emerge, one in which AI is not simply another application within the hospital, but becomes part of the hospital's operating system. Instead of individual algorithms working in isolation, networks of specialised AI agents collaborate continuously to coordinate clinical care, education, research and hospital operations.
This is the idea of the agentic hospital.
The distinction may sound subtle, but it represents a profound shift. Traditional software waits for instructions; agentic systems observe, reason, communicate with one another and recommend actions in real time. Rather than replacing clinicians, they function as an invisible multidisciplinary team working constantly in the background.
The Headline and the Reality
At first glance, the news sounds as though China has already built hospitals staffed by autonomous AI doctors and nurses. That isn't quite true. The reality is more nuanced and, in many ways, in our opinion, far more interesting.
The project attracting the most attention is Tsinghua University's Zijing Agent Hospital. Despite its name, it is not a physical hospital caring for patients. It is a sophisticated simulation: a virtual environment populated by thousands of AI doctor, nurse and patient agents that interact with one another, diagnosing disease, ordering investigations, making treatment decisions and responding to changing clinical scenarios.
Its purpose is not to replace clinicians today. Its purpose is to learn.
A Flight Simulator for Medicine
The craft of surgery is often historically compared to aviation. Long before pilots take control of a commercial aircraft, they spend countless hours in flight simulators practising emergencies they may never encounter in real life. These simulated environments allow mistakes to be made safely, systems to be refined and judgement to develop without putting passengers at risk.
Tsinghua is attempting something similar for healthcare. Rather than training AI solely on historical medical records, researchers have created a virtual hospital where AI agents can learn through experience, collaborate with one another and continuously improve their clinical reasoning before they are deployed in real-world settings. That distinction is important because it changes how we should interpret the headlines. China has not yet built the world's first agentic hospital. It is building the world's first large-scale testing ground for agentic healthcare.
Earlier this year, Tsinghua officially launched the AI Agent Hospital, one of the first healthcare projects anywhere in the world designed specifically around autonomous AI agents. In this simulated environment, every doctor, nurse and patient is represented by an autonomous AI agent. Millions of consultations can take place simultaneously. Patients develop symptoms, clinicians investigate them, diagnoses are made, treatments are prescribed, and complications are managed, all within a digital world. The purpose is simple but powerful. Instead of training AI only on historical medical records, these systems learn through experience in a safe simulated environment, allowing models to improve before they are introduced into real clinical practice. It is, in many ways, a flight simulator for medicine.
From Single Hospitals to Shared Infrastructure
Meanwhile, Beijing Tsinghua Changgung Hospital has continued expanding its own digital infrastructure. With approximately 1,500 beds, it is deploying AI not only in clinical medicine but also across research, education, regional health collaboration and hospital management. The institution increasingly resembles a living laboratory where clinical practice and AI development evolve together.
Elsewhere in China, the concept is moving beyond individual hospitals. One of the most significant developments came through the collaboration between Southern Hospital of Southern Medical University and Huawei, which unveiled the Hospital AI Platform (HAIP). Rather than creating another clinical application, Huawei describes HAIP as an AI operating system for hospitals. The platform connects hospital data, large language models, knowledge bases and multiple autonomous agents into a single coordinated environment.
One Patient, Six Agents
This is where the concept of the agentic hospital becomes perhaps easier to understand. Imagine a patient arriving in the emergency department with severe abdominal pain. Before a surgeon has even examined the patient, one AI agent has already reviewed previous admissions and medications. Another has interpreted laboratory results. A radiology agent has analysed the CT scan. A surgical agent has estimated the probability of appendicitis or bowel obstruction. At the same time, operational agents are identifying available theatre capacity, checking bed availability, preparing pharmacy orders and estimating likely discharge dates.
Each agent performs a specific role. None of course replaces the clinician. Together they create an intelligent support system that continuously coordinates care while clinicians remain responsible for judgement, empathy and decision-making.
The West Is Asking a Smaller Question
This represents a very different vision from much of the current conversation surrounding healthcare AI. Across Europe and North America, many discussions continue to focus on ambient scribes, documentation assistants and administrative automation. These technologies are valuable because they reduce the burden of paperwork, but they remain relatively narrow in scope.
China appears to be asking a different question.
What happens if intelligence is embedded throughout the hospital itself?
What if every clinical pathway, every operational decision and every patient journey is supported by networks of AI agents working together?
The implications extend well beyond efficiency. Hospitals could continuously learn from every patient encounter and clinical pathways could evolve dynamically as new evidence emerges. Operating theatre schedules might adapt in real time and early signs of deterioration could be recognised hours before clinicians would normally detect them. Hospital management could and should, one might argue, become predictive rather than reactive.
The Problems Nobody Has Solved Yet
Of course, significant challenges remain. Clinical governance becomes more complex when autonomous agents interact with one another. Transparency will be essential if clinicians are to trust recommendations generated by multiple AI systems. Cybersecurity becomes even more critical when intelligence is deeply integrated into hospital infrastructure. Regulatory frameworks will need to evolve alongside the technology, ensuring that accountability always remains with healthcare professionals.
None of these challenges are trivial. Yet history suggests that healthcare has always been transformed not only by new discoveries, but by new ways of organising care. The stethoscope changed examination, anaesthesia changed surgery, medical imaging changed diagnosis, and electronic health records changed documentation.
What Hospitals May Be Judged On Next
Agentic hospitals may change the hospital itself. Whether China ultimately builds the world's first fully agentic hospital is almost secondary. What matters is that the question has now shifted from "Can AI help doctors?" to "What does a hospital look like when intelligence is woven into every process?". That is a far bigger idea.
The answer is unlikely to arrive overnight. Clinical governance, regulation, transparency and patient safety remain significant challenges. Hospitals are among the most complex organisations in society, and autonomy in healthcare demands a far higher standard than autonomy in almost any other industry.
But every technological transformation begins with experimentation. The Wright brothers built prototypes before commercial aviation. Autonomous vehicles learned in simulation before driving on public roads. AI itself was trained for years in virtual environments before it entered everyday life.
The hospitals of the future may not be defined by the number of beds they contain or the sophistication of their robots. They may instead be judged by the quality of the intelligent agents quietly coordinating care behind every patient journey. Just as electricity became invisible infrastructure for every modern hospital, AI may become equally invisible yet omnipotent, always learning and always supporting clinicians.
The age of the digital hospital is already here. The age of the agentic hospital may have just begun.
