Skip to content
Back to publications
Nature Medicine Journal Comment

Initial lessons from real-world implementation of an AI-agent eye clinic in China

Tao Yan, Di Zhang, Luxiao Chen, Taizhangtian Ma, Chunyang Tang, Zhe Pan, Ziyao Xia, Yiming Qin, Zehua Jiang, Haoyang Liu, Mengda Li, Wan Lu, Junyi Wang, Chendi Li, Chen Xin, Siyong Lin, Chun Zhang, Peng Liu, Jiamin Wu, Ya Xing Wang, Qionghai Dai, Tien Yin Wong

Early lessons from AI-TEC on data quality, clinical workflows, and clinician feedback when bringing AI agents into ophthalmic care.

Read paper (opens in a new tab)

My role

I contributed to data collection and processing and was a main contributor to the development and deployment of the AI-TEC system.

Coverage: ScienceAlert (opens in a new tab) All coverage

Comparison of conventional ophthalmic care with the proposed AI-TEC pathway, connecting agents before, during, and after visits while retaining ophthalmologists' final clinical decisions.
The proposed AI-TEC care pathway connects specialized agents across the patient journey, with ophthalmologists responsible for final clinical decisions. Yan et al., Nature Medicine (2026).

Overview

This Comment discusses early experience with the AI-Agent Augmented Tsinghua Eye Clinic (AI-TEC). Its central concern is how AI can fit into clinical care through coordinated workflows, clinician engagement, and evaluation of clinical value.

The care pathway and its implementation

The proposed framework connects agents for pre-consultation, triage, diagnosis, decision support, and patient follow-up. A prototype was integrated into Beijing Tsinghua Changgung Hospital in November 2025, with capabilities including pre-consultation and fundus image analysis. The broader pathway illustrated above describes the framework’s intended coordination across care.

Three themes emerge from the implementation experience:

  • Data quality: expert review of clinical data matters for model performance.
  • Workflow integration: usability and the operational burden on clinicians affect adoption.
  • Continuous feedback: close collaboration between clinicians and developers supports ongoing improvement.

These are early implementation lessons. The article calls for evaluation of effects on care delivery and patient outcomes; strong image-analysis performance alone does not establish those benefits.

Media coverage

DOI: 10.1038/s41591-026-04631-z (opens in a new tab)