Medical artificial intelligence is moving beyond the era of general-purpose models into a new phase where AI agents autonomously handle complex research tasks, from drug discovery to clinical diagnosis. This shift represents a fundamental change in how healthcare organizations will deploy AI, moving from building individual models for each disease to creating intelligent systems that can adapt and work independently across thousands of medical applications.

What Is Driving Medical AI Into the AI Agent Phase?

On September 2, Hong Kong Polytechnic University and Diagens Tech unveiled the PolyU-DIAGENS Joint Laboratory for Artificial General Intelligence and Medical Applications, marking a concrete step toward the next generation of medical AI. The partnership reflects a growing recognition that healthcare AI has reached an inflection point, where the technology must evolve to solve real-world problems at scale.

The journey to this moment has unfolded in distinct stages. The first stage involved building small, specialized models for individual diseases or medical tasks, a process that typically took years and required teams to collect targeted data and train dedicated systems. While these models worked, they were expensive and slow to develop. The second stage introduced large foundational models that could be adapted for different diseases and datasets, cutting development cycles from years to months. However, even this approach still required substantial manual work for data preparation, model configuration, and result analysis.

"AI for Science is reshaping the global medical AI landscape. AI presents challenges and opportunities comparable to the Apollo Program in helping humans decode life and health, and advance diagnosis, prevention and prediction of complex diseases," said Dr. Song Ning, Founder and Chairman of Diagens Tech.

Dr. Song Ning, Founder and Chairman of Diagens Tech

Now, the field is entering stage three: the age of AI agents. In this phase, autonomous AI systems will handle the manual work that currently slows down medical research, including data curation, parameter configuration, model training, result analysis, and iterative refinement. Rather than requiring human experts to build specialty-specific models from scratch, AI agents will adapt general medical models to new diseases and tasks with minimal human intervention.

How Are Universities and Companies Building This New Generation of Medical AI?

The PolyU-DIAGENS Joint Laboratory will combine academic research strengths with industrial deployment capabilities to accelerate the transition into the AI agent era. The collaboration will focus on three key areas:

• Medical Image Analysis: Developing AI systems that can interpret medical images across multiple organs and disease types without requiring separate models for each condition.
• Medical Foundational Models: Creating large-scale AI models that serve as a foundation for thousands of specialized applications, similar to how general language models power diverse text-based tasks.
• Automation of Research and Development Workflows: Building AI agents that autonomously handle the repetitive, time-consuming tasks that currently slow down drug discovery and clinical research.

Diagens Tech has already demonstrated the feasibility of this approach. As of mid-2026, the company has collaborated with 99 hospitals to train 158 vertical models spanning 43 human organs and 61 disease areas, validating the technical pathway for batch model training enabled by reuse of foundational capabilities. The company developed the world's first foundational medical imaging model called iMedImage, along with intelligent image annotation platform iMedStudio and a dedicated model training and delivery platform called iMedMaaS, creating an end-to-end value chain covering data generation, model development, and deployment optimization.

"Medical AI may well become the highest-value vertical industry for AI deployment in the future. AI-healthcare integration is now at a critical inflection point of technological paradigm shift," explained Dr. Li Yongqi, Project Lead of the Joint Lab.

Dr. Li Yongqi, Project Lead of the Joint Laboratory

Why Does This Matter for Healthcare Systems and Patients?

The transition to AI agents addresses a fundamental challenge in medical AI: the sheer scale of unmet need. There are over 5,000 medical imaging detection tasks globally awaiting solutions, far more than traditional small-model approaches could ever address. By automating the workflow for developing and deploying medical AI models, the field can accelerate the pace of innovation and make advanced diagnostic and research tools available to hospitals and clinicians worldwide.

However, experts caution that technology alone is not enough. Healthcare organizations face a broader architectural challenge: many are attempting to layer new AI tools onto outdated, inefficient workflows rather than redesigning care delivery from the ground up. When AI is simply bolted onto legacy systems, it often shifts burden to clinicians rather than relieving it. Less than half of new AI tools deployed over the past two years have actually made healthcare providers more productive, according to research cited in industry commentary.

The real promise of AI agents in healthcare lies not just in their technical capabilities, but in their potential to fundamentally reimagine how medical research and clinical care are organized. By automating routine tasks and enabling rapid model development, AI agents could free clinicians and researchers to focus on higher-value work: patient care, complex decision-making, and scientific discovery.

What Regulatory Pathways Are Emerging for AI Medical Devices?

As medical AI technology advances, regulatory frameworks are evolving to keep pace. The FDA has launched the TEMPO pilot program, which allows developers to release certain generative AI medical devices on the market without full marketing authorization, providing regulators hands-on experience with cutting-edge technology in real-world settings. Products from companies like Cadence and Limbic have been accepted into the program, which is intended to expand the availability of technologies for the Medicare ACCESS model, an experiment in paying for technology to help beneficiaries manage chronic conditions.

This regulatory flexibility reflects recognition that traditional approval pathways may not be well-suited to rapidly evolving AI systems. By allowing real-world deployment alongside regulatory oversight, the FDA is creating space for innovation while maintaining safety standards. As medical AI agents become more autonomous and capable, this adaptive regulatory approach will likely become increasingly important.

The convergence of academic research, industrial-scale deployment, and regulatory innovation suggests that medical AI is poised for significant acceleration. The question is no longer whether AI will transform healthcare, but how quickly healthcare organizations can adapt their workflows and infrastructure to harness its full potential.