By Luyu, from Woqsik
QbitAI | Official account QbitAI
Don't tell me…… am I the only one doing this?
Now, before every hospital visit, Ihave AI run a round for me。
What's bothering me, how long it has lasted, plus my medical history — all these scattered pieces of information are prepared during the pre-consultation stage.
When I finally meet the doctor face to face, it turns out the doctor is also usingVibe Working：
Agents assist the doctor in making judgments, and follow-up visits and health information management outside the consultation room are also handled by AI.
It sounds a bit cyberpunk at first, but it's something genuinely happening in reality.
Tsinghua AIR, founded by academician Zhang Ya-Qin, has recently produced another star AI medical company —
Zijing AI Health (Zijing Zhikang)。
Its developed"Zijing AI Hospital"system has already been gradually launched for testing in hospitals across the country.
What's even more sci-fi and futuristic is that every doctor here can have their own dedicated AI avatar to assist their work — but that's far from enough.
Just consider going to see a doctor: one ailment may involve multiple departments at the same time.
Medicine is the field that most emphasizes professional expertise; you can't have an orthopedic doctor examine your lung problems……
So when doctors with different specialties each have their own avatars, can these avatars collaborate further?
Around this question, Zijing AI Health has this time brought forwardmultidisciplinary human-agent collaboration technologyin the medical field, and further upgraded "Zijing AI Hospital" (Agent Hospital).
In the AI hospital, AI doctors and human doctors work closely together……A virtual medical world connected to real diagnosis and treatment, capable of continuous evolution, is gradually taking shape.
Bringing multidisciplinary professional judgment into the same consultation room
Let's start with the old problem.
The healthcare industry has an unavoidable reality:Good doctors are always scarce。
Every doctor has their own area of specialization, and every hospital also has its own limit on resource capacity.
Getting medical care is hard—harder than climbing to the sky.
Typically, when a patient's condition spans multiple departments, the path in reality goes like this:
First register with internal medicine; if nothing is found, transfer to surgery; after surgery, book imaging; only after visiting the registration queues of several departments does it become possible to initiate a formalmultidisciplinary team consultation(MDT).
As a result, the optimal treatment window may already have been missed.
Moreover, an MDT itself is not cheap; just coordinating the schedules of experts from various departments requires paying a considerable consultation fee.
Zijing Zhikang's approach is rather direct: since it is unrealistic to have patients seek out experts, why notfirst have the experts' AI avatars assemble collectively as a team。
When the condition can be resolved within a single department,a single AI avatar is enough。
It is like inviting the AI avatar of a top-level expert into an ordinary outpatient clinic to provide guidance and suggestions for the current doctor.
Conversely, when it comes to complex interdisciplinary problems, Zijing Zhikang simply built these avatars amultidisciplinary collaboration meeting。
As Agent avatars with different specialties join the analysis one after another, the range of professional perspectives available to a single doctor is dramatically broadened.
In this way, under the scheduling of the same operating system,AI avatars can both provide same-specialty support and participate across specialties。
For example, a patient is currently admitted to the endocrinology department due to hyperthyroidism, and during the consultation, the system simultaneously collects information that the patient also has a history of asthma.
For the current clinic room, subsequent management needs to take the respiratory system into account as well.
So the respiratory medicine avatar steps in, reminding the doctor to pay attention to the related risks of subsequent medication.
This is the basic logic of multi-avatar agent collaboration: it canmove cross-specialty support forwardto earlier stages such as the first visit and pre-consultation.
In other words, when a patient walks into this clinic room, an entire team of doctors is providing backup, and the multidisciplinary support once reserved for difficult and severe cases could become standard in everyday care.
How does the collaboration actually work? Zijing Zhikang offers two forms.
One ison-demand reminders。
The doctor conducts the outpatient visit as usual; as the patient keeps providing new information, the system identifies the need for other specialties to participate, and then presents the relevant avatars' opinions to the doctor in the form of reminders.
This does not interrupt the original diagnostic and treatment process.
The other isconsultation mode。
For cases already identified as requiring joint analysis by multiple specialties, multiple avatars participate directly, presenting corresponding suggestions on the interface regarding diagnosis, medication, and other aspects.
What the doctor has to do is use their own professional knowledge to compare and weigh these different perspectives.
Of course, with more opinions comes another problem:what if the avatars say different things?
Honestly, Zijing Zhikang's approach is rather counterintuitive.
It does not actively converge the disagreements; instead, it lays the disagreements out clearly.
The system presents the suggestions from different avatars separately,highlighting the inconsistencies. The doctor can see which avatar each opinion comes from, then combine the patient's situation, the relevant evidence, and their own experience to make the final judgment.
Because the differences themselves have value.
One practicing internal medicine physician said that AI's important help at this stage is still supplementing directions the doctor may not have thought of:
When an AI suggestion is not fully consistent with my judgment, I review the information I entered to see what the reason is.
If the idea it derives from the available materials points in a different direction from what I lean toward,I would find it very meaningful。
Next, I would need to very carefully discern which one better fits the actual situation.
So the value of multi-avatar collaboration ultimately comes down to thefollow-up analysis and judgmentprocess.
Compared with other medical Agents, this is also the core reason why Zijing Zhikang's system can better fill in professional perspectives and provide traceable sources.
In fact, this is not just Zijing Zhikang's word alone.
Organizing multiple AI roles together has itself become one of the important exploration directions in medical AI.
For example, in Microsoft's MAI-DxO research, by simulating a team of doctors, information is gradually gathered around cases, diagnostic hypotheses are proposed, and tests are selected, in order to study the relationship between diagnostic accuracy and cost.
What Zijing Zhikang further emphasizes is theconnection between the avatars and specific doctors。
Here, the doctor avatars, on top of their basic capabilities, continuously learn from real doctors, achieve self-evolving growth, and ultimately hand decision-making authority back to the real doctors.
From this perspective, multi-avatar collaboration goes beyond the traditional narrative of "using AI to replace doctors"; rather, it means that behind every doctor there standsan entire on-call team of experts。
Building the foundation, letting AI avatars keep growing
However, keeping hundreds or thousands of avatars online at all times, each performing its own duties, requires a complete system behind them.
Where do the avatars acquire their capabilities, how do they understand ever-increasing patient information, how do they adapt to doctors' working habits, and how do they connect to hospitals' existing workflows…… All these interlocking factors affect the final results of use.
This is why Zijing AI Healthcare built theZijing AI Hospital ecosystem.
Similar medical agent platforms exist, such as Microsoft's Healthcare Agent Orchestrator, which organizes different agents to process information and builds workflows around complex tasks like tumor consultations.
What makes Zijing AI Hospital different is that its avatars continue to grow through real outpatient practice.
This core technical route originates from the SEAL framework proposed by Professor Liu Yang's team at AIR when Agent Hospital was first established, namely Simulatable Evolvable Agent Learning.This is a dual-layer evolutionary architecture of"virtual world training + real-world validation and fine-tuning"
.In the virtual environment, roles such as patients, doctors, and nurses are all played byagents
.AI doctors conduct high-intensity simulated diagnosis and treatment in it every day, with millions of highly realistic AI patients seeking medical care, forming a closed-loop medical virtual world.Successful cases
can enter the case library for reference in subsequent tasks;
when problems arise, the system reflects on failures, distills lessons, and gradually forms reusable rules through organization.In just a few days, an AI doctor agent can complete hundreds of thousands or even millions of cases, running through all rare diseases, complex diseases, and multi-course disease combinations.Analogous to AlphaGo's self-play, AI doctors evolve themselves in the virtual hospital. While capabilities steadily improve,
compute consumption is far lower than traditional large model training
.However, early AI trained in the virtual world also has an obvious problem: it tends to consider all possibilities。
excessively thoroughly.。
Qiao Yuchen, Chief Product Officer of Zijing AI Hospital, revealed such an interesting little detail:
The AI recommends the most effective drug to the doctor, but the doctor, considering the patient's financial situation, finally switched to a relatively cheap drug covered by medical insurance.
Thusreal-world diagnosis and treatmentcomes into play.
By introducing real-world validation and companion growth, the system lets the digital avatar work alongside human doctors, then feeds the doctors' operations back into internal fine-tuning.
Agents must not only act like doctors, but also think in sync with them.
When it comes to practical applications, Zijing AI Hospital serves three sides: doctors, patients, and hospitals:
Patient portalResponsible for pre-consultation screening, health records, and follow-up management;Doctor sideTakes on the AI avatar, medical records, and decision support;Hospital sideSupports scheduling, permissions, and data management.
The three-sided product brings the underlying capabilities to different roles. Continuing to integrate specialty tools and partners is the next step built on this foundation.
At present, this system has already delivered a report card in real hospitals.
As of the 7th month of 2026, the system had built 54 AI doctors and several million ultra-realistic AI patients, covering thousands of diseases.
Beijing Tsinghua Changgung Hospital has launched outpatient testing in 18 clinical departments, with more than 40 doctors using it on a regular basis, and doctors' overall diagnostic and treatment efficiency can be improved by30%。
In benchmark testing, the AI doctor achieved an accuracy of over96%on the MedQA dataset, and achieved the highest scores in multiple international medical evaluation datasets.
The next question to answer is how to bring this professional support into more hospitals and benefit more patients.
The AI hospital's next step: reaching more people
Over the years, medical AI has tended to work in isolation.
Whether vendors or hospitals, all are accustomed tostarting from a specific tasksuch as imaging recognition, medical record generation, or assisted consultation.
But when it comes to actually seeing a doctor, it is certainly a continuous process.
The industry has long lacked a production-grade AI-native systemthat can connect different roles and run through the entire diagnostic and treatment process.。
Is it really that nobody wants to do it? Not necessarily — the difficulty is that very few people can tie together cutting-edge technology, clinical practice, and engineering.
This is an awkward situation faced by both academia and industry:
Harvard, MIT, Stanford and othertop universitiesindeed possess the most complete medical AI innovation ecosystems. But going from experiments to clinical use still lacks systematic engineering validation.
Top medical institutions(such as Mayo Clinic, Toronto General Hospital, and Cleveland Clinic) sit on deep clinical resources and abundant patient needs.
But they lack top AI talent and technology, and cannot independently develop AI-native systems.
The AI tech giants of Silicon Valleyhave gone somewhat further, but medicine is only one of the many fields in their portfolios; they lack system-level innovation and also need to understand and adapt to medical workflows step by step in actual diagnosis and treatment.
Deploying production-grade AI-native medical systems requires bringing these capabilities together, so at presentZijing Zhikang is T0-level。
Zijing Zhikang is quite a new player, incubated only two years ago byTsinghua's Institute for AI Industry Research(AIR).
In the medical field, where a single research project often takes a decade or more, it does indeed seem green behind the ears.
But the team's research accumulation goes back much earlier, and its members have long been rooted in smart healthcare:
Multiple papers have successively appeared in top medical AI journals such as Nature Medicine and NEJM AI.
Among them, the Agent Hospital paper that went viral at home and abroad in 2024 became an instant hit, being called"the world's first AI hospital"。
The related practice also receivedNature'sclose attention, and was listed by it as a core reference case for deploying AI virtual avatars in the medical field.
Zijing Zhikang is a very typical success story oftranslating academic results into practice.
Starting from Tsinghua's AIR, to publishing in top journals, and then entering real clinics, this team is step by step turning academic accumulation into products that both doctors and patients can actually use.
One More Thing
At today's launch event, there was one key quality of Zijing Zhikang that moved me deeply.
A gentleman should carry the world with virtue.Inheriting the inclusive character of Tsinghua's AIR, Zijing Zhikang, while advancing its technology, has never neglected the humanistic care of medicine.
Zijing Zhikang advocates that, throughvalue alignment, medical humanities concepts are translated into algorithmically understandable and executable constraints and preferences.
In other words, the system firmly locks in the boundary of collaboration between AI and humans at the design stage, following ahuman-centeredorientation and clearly defining AI's auxiliary role.
In building its agents, it alsodoes not pursue an all-capable, unified super AI, but instead preserves the distinctiveness and diversity of each agent, drawing on the strengths of all.
Concretely, this means the devolution of professional-grade capabilities, enabling more grassroots doctors, barefoot doctors, and even township institutions to access front-line resources, delivering real benefits to the broad patient population.
Admittedly, this road is not easy; every compliance hurdle and every doctor's habit change must be polished through concrete practice.
The path Zijing Zhikang has chosen—from the virtual world to real clinics—is slower but also steadier.
Yet a foreseeable glimmer of hope is that every additional piece of effective support for doctors helps patients take fewer detours.
Only when life is sustained can the value of technology be truly revealed.
Official website link: https://www.tairex.cn/