Next-generation AI agents are headed for healthcare. What will they do once they get here?

Healthcare AI agents can be classified as one of four models. In increasing order of autonomy and clinical integration, these are: foundation, assistant, partner and pioneer.

What all four have in common is the potential to transform healthcare by advancing clinical decision support, automating workflows and personalizing patient care. 

That’s according to researchers with the Icahn School of Medicine and Mount Sinai Health System in New York City. 

Success with the technology in medicine will depend on “continued technological innovation, robust validation frameworks and effective collaboration between technical and healthcare stakeholders,” corresponding author Dhavalkumar Patel and colleagues write in a study awaiting peer review and journal publication. 

“As the field evolves,” the authors add, “maintaining focus on patient outcomes while addressing privacy, security and accessibility concerns will be paramount for sustainable adoption and meaningful impact in healthcare delivery.”

In a section on future directions for agentic AI in healthcare, Patel and co-authors note the current trajectory “indicates a shift toward more integrated, efficient and patient-centered systems.” They consider the developing field by making six granular forecasts, as follows. 

1. Emerging technologies: Technical advancements. 

Next-generation AI agents will harness cutting-edge architectures such as transformers and generative AI models to achieve unprecedented levels of accuracy and efficiency, Patel and colleagues write. For example, hybrid models combining reinforcement learning and supervised learning are “anticipated to excel in real-time decision-making scenarios, such as emergency care and surgical assistance.”

Integration innovations, including edge computing and federated learning, will enable AI systems to process data locally, enhancing privacy and reducing latency, the authors add. More: 

‘Performance improvements in graphics processing units and tensor processing units will further accelerate AI training and inference capabilities, allowing for faster and more scalable deployments.’

2. Emerging technologies: Healthcare applications. 

Emerging technologies will introduce novel clinical applications, such as AI-driven molecular diagnostics, enabling precise detection of rare genetic disorders, the authors predict. “Workflow evolution will focus on automating complex processes like multidisciplinary treatment planning, reducing clinician workload,” they write before adding: 

‘Advances in patient care are expected through personalized medicine, powered by AI agents capable of tailoring therapies based on genetic and lifestyle data.’

3. Research directions: Technical research. 

Future research will prioritize the development of more robust and interpretable algorithms. Explainable AI (XAI) models “will address the black-box nature of current systems, increasing clinician trust and facilitating regulatory approval,” the authors write. “Advances in system architecture will focus on modular, plug-and-play designs to simplify integration with existing healthcare IT systems.” More: 

‘Research into integration methods will emphasize standardized APIs and interoperability frameworks to enhance data flow between disparate systems.’

4. Research directions: Healthcare research. 

Clinical validation will remain a critical focus, with large-scale randomized controlled trials assessing the efficacy and safety of AI interventions, the authors expect. Meanwhile implementation studies “will investigate the best practices for deploying AI agents in diverse healthcare environments, from rural clinics to urban hospitals.”

‘Outcome research will explore long-term impacts on patient health, provider efficiency and healthcare costs.’

5. Implementation roadmap: Development path. 

Short-term goals (1 to 2 years) include refinement of existing models for improved accuracy and reliability, Patel and co-authors point out. “Key initiatives will focus on integrating AI into high-impact areas such as radiology and pathology. Medium-term objectives (3 to 5 years) emphasize expanding AI applications into complex domains like personalized medicine and predictive analytics.”

‘Long-term vision (5+ years) aims to create next-generation AI agents that seamlessly integrate into all facets of healthcare.’

6. Implementation roadmap: Success indicators. 

“Success will be measured by clinical outcomes, such as reductions in diagnostic errors and treatment delays,” Patel et al. note. “Operational metrics, including workflow efficiency and cost savings, will also serve as critical benchmarks.” 

‘User satisfaction among healthcare providers and patients will be key to assessing the real-world impact of AI agents.’

Download the full preprint paper here.