Tairex Announces Multidisciplinary Human-Agent Collaboration (MHAC)

On October 21, 2026, Tairex held the Agent Hospital Multi-Agent Collaborative Ecosystem Launch Event, themed “Coexistence · Collective Intelligence · Shared Future,” at the China National Convention Center in Beijing, officially unveiling its Multidisciplinary Human-Agent Collaboration (MHAC) framework. MHAC is a new approach to diagnosis and treatment in which human physicians from different specialties work collaboratively with agents to deliver medical services with earlier intervention, broader applicability, wider accessibility, lower supporting costs, higher-quality decisions, and stronger closed-loop feedback.

MHAC is closely related to multidisciplinary teams (MDTs), which bring together specialists from different departments to discuss a patient’s condition and develop a diagnosis and treatment plan. For example, for a patient with a complex cancer, the lead department may convene specialists in oncology, surgery, radiation oncology, radiology, pathology, and other relevant fields to review images and test results and jointly determine the most appropriate treatment. MDTs offer patients several benefits: more comprehensive treatment plans that avoid the limitations of a single department; integrated, one-stop assessments that reduce repeated appointments and hospital visits; individualized treatment plans tailored by specialists across disciplines; fewer biases in treatment decisions and lower risks of misdiagnosis or missed diagnosis; and consideration not only of immediate treatment but also of rehabilitation, nutrition, pain management, follow-up, and other subsequent care.

Nevertheless, MDTs face six challenges. First, intervention often comes too late: patients may be considered for MDT consultation only after visiting multiple departments without resolving their problems, potentially missing the optimal treatment window. Second, applicability is limited: MDTs generally rely on centralized, in-person meetings. Because gathering experts from multiple departments is difficult, a single meeting often covers several patients, leaving little time for each. Third, access is narrow: MDT services are generally offered by large top-tier tertiary hospitals to their own patients, reaching only a small number of selected complex cases while excluding many early-stage, medium- or high-risk cases that could benefit. Fourth, service costs are high: the involvement of experts from multiple departments makes MDTs much more expensive than ordinary outpatient visits, restricting their frequency regardless of whether hospitals or patients pay. Fifth, decision quality is constrained: conventional MDTs mainly involve experts from the same hospital. Where certain departments are weaker, the quality of their input may be limited, while bringing leading external experts to an in-person meeting is prohibitively expensive and often impractical. Sixth, closed-loop feedback is inadequate: MDTs typically produce recommendations without mandatory implementation and follow-up mechanisms. Important information about changes to treatment plans, adverse reactions, and treatment outcomes may not flow back to the MDT for review, leaving decisions disconnected from clinical practice.

The key to overcoming these challenges lies in doctor avatar agents. If every physician has a dedicated, personalized avatar agent, all of the above MDT limitations could potentially be addressed. To intervene earlier, a physician can dynamically call on avatar agents from relevant specialties during an outpatient consultation, bringing the right expert perspective into the process at the right time. This would remove the restriction of MDTs to centralized in-person meetings and substantially expand access to medium- and high-risk early-stage patients who could benefit. Because specialists need not necessarily attend at a fixed place and time, overall MDT costs would fall significantly. Doctor avatar agents could also help transcend institutional boundaries: subject to regulatory compliance and agreement between the hospitals concerned, avatars of leading experts at other institutions could participate in an internal MDT. This would greatly expand the expert pool and raise the potential ceiling for decision quality. Finally, doctor avatar agents could connect directly to a patient’s full-cycle health-management loop, promptly returning follow-up outcomes to the MDT team and thereby improving service quality.

Doctor avatar agents fundamentally redefine the relationship between human physicians and artificial intelligence. They are an important means of helping physicians apply medical knowledge and expertise at scale beyond constraints of time and location. We believe doctor avatar agents should follow three core principles:

1. Compliance with established constraints. Strictly follow recognized medical guidelines, regulatory requirements, and ethical constraints.

2. Replication of physician characteristics. Reproduce the individual characteristics of the human physician, including but not limited to consultation style, clinical reasoning, and risk preferences.

3. Deference to physician decisions. If the first two principles conflict, explicitly alert the physician and leave the decision to the human physician.

In addition, doctor avatar agents should possess eight key attributes:

1. Attributable. Be linked to a specific human physician and constitute that physician’s exclusive digital asset.

2. Evolvable. Continuously develop automatically through collaboration with the human physician.

3. Assessable. Allow the degree of the avatar’s evolution to be measured quantitatively.

4. Explainable. Explain the reasoning behind its decision recommendations.

5. Auditable. Maintain records of all actions for audit at any time, enabling causes to be traced when problems arise.

6. Calibratable. Allow the human physician to manually correct deviations in the avatar’s behavior.

7. Authorizable. Once a specified evolution threshold is reached, allow the human physician, subject to regulatory and ethical requirements, to authorize the avatar to make designated categories of decisions on the physician’s behalf.

8. Shareable. Be listed in Agent Square and invoked by other users.

Building doctor avatar agents is highly complex and difficult. Learning from physicians’ publications and medical records alone is insufficient. Avatar development must be closely integrated with physicians’ decision-making throughout their daily workflows, enabling physicians to teach their avatars through both instruction and example. To achieve this, Tairex has worked in three areas and developed the capabilities needed to build doctor avatar agents:

1. Technology. Tairex has developed a dual-world agent evolution system. In the virtual world, AI doctor agents interact with AI patient agents and acquire medical knowledge and skills through high-throughput evolution. A doctor avatar agent is initialized from an AI doctor agent, then collaborates with a human physician and interacts with human patients in the real world. Through symbiotic evolution, the avatar achieves exceptionally close alignment with its human physician.

2. System. Tairex has developed an AI-native clinical operating system that deeply integrates AI into clinical workflows and reshapes those workflows. During routine outpatient consultations, physicians collaborate with their avatars and provide supervision through an “AI recommendation–human approval” process. Physicians do not need to label data themselves; their avatars can evolve automatically. Behind the user interface, Tairex has also undertaken extensive infrastructure work to ensure data security and privacy protection.

3. Operation. Tairex has established deep partnerships with physical hospitals. Its Agent Hospital system is closely integrated with daily hospital operations, spanning administrative departments such as medical affairs and information technology, outpatient specialties such as pulmonology and neurology, and care stages including pre-consultation, outpatient visits, follow-up visits, inpatient care, and MDTs. Tairex has explored and established efficient workflows, communication mechanisms, and iterative feedback loops that are essential to the continued evolution of doctor avatar agents.

Building on the foundation provided by doctor avatar agents, MHAC is a dynamic framework that enables human clinicians, personalized doctor avatars, and specialized AI agents to collaboratively solve medical problems through adaptive recruitment, communication, and coordination of heterogeneous clinical intelligence. The essence of MHAC is not to add more AI tools, but to establish a new paradigm of medical collaboration that dynamically organizes, coordinates, and integrates diverse forms of clinical intelligence. MHAC supports three basic modes:

1. Single-human–single-agent mode. One human physician collaborates with their own avatar agent. This was the standard operating mode of Tairex Agent Hospital before MHAC; MHAC maintains backward compatibility with it.

2. Single-human–multiple-agent mode. One human physician collaborates with multiple agents, including their own avatar, avatars of physicians in other departments, and other relevant medical agents.

3. Multiple-human–multiple-agent mode. Multiple human physicians collaborate with multiple agents. This resembles conventional consultations involving several physicians, but also includes multiple doctor avatar agents and other medical agents.

Importantly, in all three modes, human physicians remain indispensable decision-makers, while agents primarily play supporting roles. Accordingly, the Tairex Agent Hospital system prohibits decisions made entirely by agents without human physician participation.

The single-human–multiple-agent mode is MHAC’s most distinctive new form. It brings multidisciplinary collaboration forward to the outpatient stage and significantly expands the population that can benefit. In this mode, in addition to using their own avatar by default, the attending physician can manually select multiple doctor avatar agents or AI doctor agents from Tairex Agent Hospital’s Agent Square to serve as assisting specialist agents. These agents may come from the same specialty as the attending physician or from other relevant specialties. In the first case, avatars of senior physicians can help junior physicians in the same discipline improve their clinical practice. In the second, they enable timely access to opinions from specialists in other departments. Assisting specialist agents can participate in two ways:

1. Passive triggering. Assisting specialist agents participate throughout the clinical encounter and issue timely alerts concerning decisions made by the attending physician and their avatar, based on information available at that point. Alerts may support those decisions or offer suggestions from other perspectives. Such alerts are advisory only; the attending physician makes the final decision.

2. Active triggering. When the attending physician considers the information sufficient and believes multidisciplinary collaboration is necessary, they can select assisting specialist agents and initiate a consultation during the outpatient visit. Each assisting agent offers its opinion and discusses the case with the attending physician, who makes the final decision.

Assisting specialist agents can be selected manually or automatically. In manual selection, the attending physician enters Agent Square, reviews the available doctor avatar agents and AI doctor agents, and selects agents from relevant specialties. Each agent’s profile includes information such as the physician’s name, professional title, hospital, department, avatar level, and areas of disease expertise. In automatic selection, the system dynamically recommends suitable agents from Agent Square based on information available during the clinical encounter. For example, if a patient displays typical symptoms of type 2 diabetes during history-taking, the system can automatically recommend, in the background, an avatar agent of a type 2 diabetes specialist to join the consultation. Manual and automatic selection can be used together and configured flexibly.

The multiple-human–multiple-agent mode is a significant extension of conventional MDTs, with agents providing the crucial new capability. Consultations in this mode can take place entirely offline, entirely online, or in a hybrid format. Before a consultation, the system automatically prepares meeting materials. During the consultation, each physician participates alongside their own avatar and other agents. Based on the meeting materials and the contributions of other experts, avatars prepare case analyses, decision recommendations, and speaking points for their physicians. After the consultation, the system produces meeting minutes and uses Tairex Agent Hospital’s automated follow-up function to relay the patient’s subsequent condition to participating physicians. Importantly, avatars can evolve automatically based on physicians’ actual decisions and statements.

Tairex Agent Hospital provides each physician with an avatar management interface to ensure complete control over their own avatar agent. Key functions include:

1. View. Review basic information, including the avatar’s overall level, skill levels, and areas of disease expertise.

2. Train. Accelerate avatar evolution by uploading de-identified data or directly imparting experience.

3. Publish. Decide whether to publish the avatar in Agent Square once its level meets the admission threshold.

4. Authorize. Authorize the avatar to perform specified actions, subject to regulatory and ethical requirements.

5. Audit. Audit the avatar’s complete activity history, including participation in other physicians’ clinical encounters.

6. Calibrate. Manually correct avatar behavior when it deviates from the physician’s expectations.

The newly released MHAC and Agent Square, together with the previously introduced doctor avatar agents, form a tightly integrated whole. Doctor avatar agents are physicians’ exclusive digital assets, the fundamental units of medical intelligence resources, and a bridge between the virtual and real medical worlds. Agent Square is the infrastructure for aggregating, organizing, and distributing medical intelligence resources. MHAC is the key collaboration mechanism for organizing these resources effectively to solve clinical problems. Put simply, doctor avatar agents create resources, Agent Square manages resources, and MHAC uses resources.

Tairex Agent Hospital’s virtual consultation rooms currently support 27 clinical specialties and include 54 AI doctor agents, covering the vast majority of common diseases, major diseases, and rare diseases. Furthermore, during testing at physical hospitals in Beijing, Guangxi, Wuxi, and other locations, the Tairex Agent Hospital system has made initial progress in evolving high-level avatar agents for human physicians. These AI doctor agents and doctor avatar agents constitute the initial expert pool of Tairex Agent Hospital’s Agent Square, laying the foundation for multidisciplinary human-agent consultations.

In the future, subject to regulatory and ethical requirements, Tairex will progressively conduct functional testing of multidisciplinary human-agent collaboration in real-world settings at partner hospitals. Tairex looks forward to partnering with more hospitals and physicians to explore the boundless possibilities of healthcare services in the age of artificial intelligence.