Welcome to the Era of AI Agents in MedTech
Have you ever dealt with a medical condition only to be referred to a string of specialists, each requiring you to repeat your history, each forming an opinion without the full picture, each applying the tools and techniques they have within their specialty, but never quite finding the full picture or a true diagnosis of the problem? While Electronic Health Records (EHRs) were meant to improve this, they often fall short: data is scattered, unstructured, and not easily synthesized to provide a truly holistic patient view and understanding of a condition. This can lead to missed insights, frustrated patients, excess cost/waste and suboptimal clinical outcomes.
By 2025, the global data sphere is projected to exceed 180 zettabytes, with healthcare contributing a substantial portion, estimated at over one-third. This massive data deluge presents both an unprecedented opportunity and a significant challenge. Yet, a mere 3% of healthcare data is effectively leveraged for patient care.
This underutilization stems largely from limitations within existing healthcare IT systems. These systems often struggle to process multi-modal data (e.g. genomic, imaging, electronic health records) at the scale and speed required for timely and optimized clinical decision-making. As a result, clinicians are burdened with manually sifting through vast amounts of information, hindering their ability to derive crucial insights for patient care.
Furthermore, the rate of medical knowledge growth is accelerating exponentially. The National Institutes of Health (NIH) estimates that medical knowledge doubles approximately every 73 days, particularly in rapidly evolving fields such as oncology, cardiology, and neurology.
This rapid expansion of medical knowledge further exacerbates the need for innovative solutions that can efficiently process and analyze massive datasets to support evidence-based clinical practice.
Welcome to the Era of AI Agents
If 2024 was characterized by an influx in Large Language Models (LLMs) with many applications within the development of MedTech and the devices themselves 2025 has the promise of agentic AI in MedTech.
Agentic AI refers to artificial intelligence systems that exhibit agency, meaning they can:
- Act autonomously: They can take actions and make decisions independently that don’t require constant human intervention.
- Pursue goals: They are designed to achieve specific objectives and can adapt their behavior to reach those goals.
- Learn and adapt: They can learn from their experiences and adjust their actions based on changing conditions.
The systems are goal oriented and can formulate a plan, have defined objectives and strive to achieve them efficiently. Agentic AI adapts its behavior and strategies in response to its interactions with the environment, new information, changing circumstances and its progress toward its goals.
Being action oriented, Agentic AI is designed to do things in the real world or a simulated environment. This might involve controlling a robot, making API calls, sending emails, controlling a medical device, or any other action that can have significant effect. It can also run in a loop or continuously or over a period of time vs. chat-oriented large-language-models which typically run an output based on a prompt.
LLMs comprehend the nuances of written text and language to generate informed responses. Multi-modal foundation models (FMs) combine more than a single type of media, like images, videos, text and/or sound, to make a decision.
In a healthcare setting, other types of inputs could include genomic, radiology or pathology data.
Tremendous Future Potential
One of many use case for agentic AI in healthcare is to support the clinical professional and act as a robust healthcare support specialist, integrating data, prioritizing patients and recommending care. Unlike an LLM where you go through a traditional prompt/response, the AI agent can always be working as new data is gathered around a patient’s condition. It is working around the clock, 24 hours a day, 7 days a week. It has a temporal context measured in nano seconds as new data comes in vs. a human that might take in new data across longer time spans (hours, days, weeks or months).
Agentic AI can correlate patient, clinical results, genetic, pathology, etc., acting as a super-intelligent assistant to the physician. Perhaps the AI has a goal to diagnose a patient and recommend the optimal treatment, making informed recommendations to physicians or informing the patient of lifestyle and nutrition optimizations.
Agentic AI does not care about information overload and isn’t encumbered by limited minutes in a day. Agentic AI can also take data in real time as it become available and alert clinicians that something should change or be trending in a certain way.
Imagine a patient with a complex autoimmune disease. The agentic AI could integrate information from their rheumatologist, dermatologist, and gastroenterologist, identifying patterns that might not be apparent to individual specialists. This could lead to a faster and more accurate diagnosis, as well as a more coordinated treatment plan. The patient could also report symptoms frequently, and monitoring devices could facilitate the collection of real-world data to continuously provide feedback on treatment plans which can be then reviewed by a clinician.
Agentic AI and Impact to MedTech
This vision for the future is exciting, and it definitely offers huge potential to improve patient outcomes and allow us to live longer healthspans.
If you have been keeping up with our MedTech trends blog series, we have already covered many of the foundational ideas that MedTech companies can embrace to help make this a reality. This includes collecting real world data from an effective IoMT platform, embedding AI (and even adaptive AI inside the device), building a digital eco-system to commercialize beyond the device and building a comprehensive cybersecurity approach in alignment with new guidance from the FDA and other regulatory bodies.
Companies that embrace agentic AI can build on many of these trends to further improve the efficacy and safety of medical care and also monetize new services. Industry leaders, like GE, are already working to integrate agentic AI into workflows for clinics.
The integration of digital health is already enabling MedTech companies to adopt new and more sustainable revenue models, moving away from the traditional "boom-and-bust" cycles associated with sporadic device sales. A significant shift is occurring towards recurring revenue models, which include subscription-based services, software add-ons that enhance device capabilities.
The integration of agentic AI in healthcare represents a further transformative leap forward, addressing critical challenges such as data underutilization and the exponential growth of medical knowledge. By autonomously processing and analyzing vast, multi-modal datasets, agentic AI can significantly enhance clinical decision-making, reduce the burden on healthcare professionals, and should dramatically improve patient outcomes. This technology can seamlessly integrate diverse data sources, provide real-time insights, and facilitate a more holistic and coordinated approach to patient care. It can also serve as another way to cement recurring revenue for the industry.
As the MedTech industry advances through 2025 and beyond, it stands at the forefront of a paradigm shift towards AI-driven, connected digital health, where intelligent automation and agentic AI will almost certainly play a pivotal role in always-on data orchestration and continuous improvement.
One (perhaps ambitious) predictive model, synthesized from analyses by leading industry analysts and consulting organizations using an advanced AI model (ChatGPT 40 Deep-Research Mode), projects a potential tripling of global MedTech revenues, from a base of one trillion USD to over three trillion USD by 2035. The Deep-Research model is impressive and shows it’s work, lists the research sources and a basis for reasoning and projecting the result. Figure 1 below illustrates that while "base business" will exceed one trillion USD by about 2035 based on a commonly used 6% CAGR, companies embracing digital health, including agentic AI, stand to unlock a significantly larger market, potentially reaching over three trillion USD in that same time period. While some of these revenues will be claimed by non MedTech entities like hospitals, clinics and big tech organizations (who are increasingly entering the healthcare space), the MedTech industry stands to increasingly gain substantial ground here. The size of this prize is clearly enormous.
The industry is poised to deliver innovative solutions that enhance the efficacy and safety of medical care while driving new opportunities for growth. By developing robust clinical workflows, personalized treatment plans and secure IoMT platforms, MedTech companies can capitalize on this exciting era of always-on AI-driven healthcare, ultimately leading to longer, healthier lives for patients worldwide.
Relevance to the Digital Thread
For stakeholders in the MedTech industry, these findings underscore the critical need to strategically invest in the digital thread and digital health capabilities. This includes building a digital foundation for product design, development, manufacture, distribution, monitoring and developing connected devices, leveraging AI (including AgenticAI) for enhanced functionality, and exploring telehealth platforms to expand reach and improve patient care.
Companies that supercharge their value chain by implementing an automated and integrated digital thread, supporting devices with embedded software and AI models through modern platforms and tools such as: Systems Engineering, Application Lifecycle Management (integrated with DevOps and MLOps), Product Lifecycle Management (including change, configuration, and variant management), and secure, maintainable Internet-of-Medical-Things (device connectivity), alongside closed-loop workflows and Electronic Health Record (EHR) integration, will undoubtedly be best positioned to capitalize on the paradigm shift in healthcare.
Of course, all of this must be validated against clinical outcomes and implemented with a clear understanding of use models and access rights: who is permitted to run agents, receive data, and act on recommendations. Patient privacy (in accordance with HIPAA), human-in-the-loop considerations, cybersecurity, and regulatory guidance for AI, embedded software, and Software as a Medical Device (SaMD) must be embedded from the start. This is not trivial, coordinating the many systems involved requires orchestration, funding, and expertise. But given the size of the prize, this is within reach for any well-executed digital program built on strong data governance, agile development, and continuous stakeholder engagement.
Developing a digital thread strategy to attain access to the digital health market which includes agentic AI will vary based on the type of device, the application of it and the clinical use case and the condition being treated. Here is a three-phased approach a MedTech company might take:
Phase 1: Establish the Digital Foundation
- Establish a Robust Digital Thread Foundation: Lay the groundwork with fully integrated PLM, ALM, DevOps, MLOps, and Systems Engineering, connected to MES. This digital backbone enables precise variant control, effective change management, and rapid quality feedback—essential for delivering software and AI enabled devices. Without this, intelligent automation will stall before it starts.
- Deploy a Secure IoMT Platform: Deploy a secure and maintainable IoMT platform that facilitates remote monitoring and seamless integration with patient EHRs. Ensure the platform supports remote software updates and AI model deployment by leveraging the digital thread. This is done with a robust security framework and must also make patient privacy and data usage rights a top priority.
Phase 2: Enable Intelligent Workflow Integration
- Collaborate on Clinical Workflows: Concurrently, collaborate with customers to define clinical workflows and patient services that leverage the IoMT platform. This might include AI-powered diagnostics, with persona-based role-based access for patients and clinical roles.
- Develop Intelligent Automation Models: Progressively develop models that interact with the IoMT platform, the medical device, and EHR systems. This foundational capability lays the groundwork for more autonomous agentic AI. Over time, these models can be trained to learn from patient and device data, improving accuracy and responsiveness. This adaptive AI approach can be supported through a Predetermined Change Control Plan (PCCP)—as outlined in FDA regulatory guidance on adaptive AI and discussed in a previous blog entry.
- Define and Implement Specific Agentic AI Applications: Implement specific applications of Agentic AI to enhance clinical workflows and patient experiences. Examples might include:
- Personalized Treatment Optimization: Implement AI agents that dynamically recommend adjustments to treatment plans based on real-time patient data collected from the medical device but also incorporates other clinical context or data gathered.
- Remote Diagnostic Assistance: Develop AI agents that analyze patient-generated data (e.g., wearable sensor data, home monitoring results) to provide preliminary diagnostic insights to clinicians.
- Predictive risk assessment: Implement agents that can predict patient risk for adverse events and provide alerts to care teams.
In most cases, at least for now, we recommend a human-in-the-loop approach where the agent is providing evidence-based recommendations but ultimately leaving it to the clinician to make final decisions on diagnosis and treatment.
Phase 3: Drive Clinical & Business Impact and Continuous Improvement
- Focus on Clinical Outcomes and Reimbursement: Maintain a strong focus on improving clinical outcomes, which also supports reimbursement strategies.
- Enable Data-Driven Feedback Loops: Implement mechanisms to send data back to relevant points throughout the digital thread. This includes:
- Verifying device behavior aligns with risk profiles established in Systems Engineering.
- Detecting agent behaviors that may indicate failures or unintended use.
- Monitoring software behavioral characteristics to identify potential improvements.
- Iteratively Enhance Features: Through an agile-driven process, continuously enhance device and consider offering new types of agents with different use cases. Each agent provides value and may be monetized individually or collectively.
- Optimize Device Performance: Continuously improve device performance and deliver remote updates and AI models to enhance clinical outcomes.
Throughout the journey it should go without saying that it’s critical to adhere to validation and regulatory guidance.
Best in class companies are already well on their way. Not every component is needed to take advantage of agentic AI so prioritization and strategy here are key, with data privacy and security woven into every stage of implementation.
Authors
