In the last post in this series about AI agents and their upcoming role in healthcare, I want to start with the following table that shows how the capabilities of generative AI models has evolved over the last 2 years. As you can see, most of the foundation models by Open AI, Google, and others are gaining significant functionality and capable of handling increasingly complex tasks. This is very relevant for healthcare since workflows are more complex, involve processing different types of data (notes, images, labs, etc) and based on the meaning of all of this data taken together, an action has to be taken. For example, in ambient monitoring, AI agent can analyze signals about the patient’s facial expressions, words spoken, quality of their voice, movement patterns, gait quality, and more and notify the nurse when something needs their attention.
This could be an acute event such as distress as a result of pain, or something more chronic, like a deteriorating gait that can indicate a pending fall. All of this is done manually today and much of it goes unnoticed because of the shortage of resources. What are the implications for our healthcare system and economy? Instead of detecting costly issues before they happen, we end up having to deal with their complications. This puts strain on our system in terms of the required manpower to handle all of these issues and of course, it’s much more expensive. The biggest threat to the long-term solvency of our federal budget is the cost of providing healthcare for Medicare, Medicaid, and veterans.
Dr. Dennis Chornenky of UC Davis said in his talk at HIMSS 2025 that the rapid pace of change in AI capabilities are advancing at an exponential rate (as displayed in the chart above!) and predicted: “Over the next several years, we can expect, probably each of the years ahead, for AI’s cognitive reasoning capabilities to jump by not 10 times, but more like 100 times with each iterative cycle.” This, of course, could mean that many of the activities that we have not even imagined to be handed off to AI may actually happen soon. In a Healthcare IT News article about his talk, the differences between what gen AI is and agentic AI are described this way: “GenAI can create content, but agentic AI can take action – even learning and setting goals”. That could mean that many jobs involving analysis, creation of reports, patient communication and more can be done rapidly by AI agents. What does that mean for the healthcare workforce? or for that matter, our overall workforce? As I discussed in the first post in this series, it’s my opinion that we will not run out of things to do! Freeing up our human resources from doing these activities will mean that we will address the significant shortage of resources in our healthcare system, improve the lives of our burned out staff, and improve the experience for patients. Freeing up our staff to focus on higher value activities could also mean that we will move to a more proactive care that is far more cost effective and higher quality than the current reactive care.
To roll out all of these emerging capabilities, much has to be done before we can see the benefits. We need to create the infrastructure to manage all of these agents and strong governance to ensure that they are being used in a safe and effective manner. Much of that infrastructure does not exist yet today and will need to be created. Also, most medical centers don’t yet have the governance in place to use AI to its full capabilities, or the trained staff to onboard and monitor these agents on an ongoing basis. Another massive barrier is the fact that the healthcare workforce has not been trained in how to use AI in their jobs. If you are using AI agents to perform some of the activities that a nurse does today, the nurses will need to be trained in how to safely and effectively work with these agents. They will be the frontline staff who will ensure the output of the agents make sense, is not causing harm to the patient, and report any issues. The training challenge of preparing the healthcare workforce to work with AI is an enormous one. Healthcare is the number 1 employer in the country!
One more key area to mention in the context of agentic AI is the technology stack needed to maximize the potential benefits of these technologies. Current EHRs were designed decades ago and the primary objective was to create a record for billing purposes. The manner in which the information is collected is not AI-friendly. Also, given that each system has its own EHR, different parts of the patient’s information can be in different EHRs, not to mention that labs, genetics, and other types of data may be in other systems or in the EHR as a PDF. All of this will mean that the powerful functionalities that we envision for these AI agents will not fully materialize anytime soon for some very high-value use cases. Why? Providing high-quality, proactive care to patients requires complete information about them and any missing piece can alter the picture that may represent an inaccurate assessment of their condition.
All of this means that we’re at an inflection point. We now have new tools that can not only help in performing many complex healthcare activities but also soon it will perform those activities. While the optimism and hype around this is justified, it’s important to understand that many of the issues we’ve discussed such as training, technology infrastructure, safety and more will slow down the timely adoption of these tools. So, as I say in all of my talks or write in my book, there’s every reason for optimism for the long-term implications of AI in healthcare but making predictions about what will things look like in three or five years, is a fool’s errand. For now, agentic AI is on the horizon and the sooner we get busy addressing the infrastructure and the governance needed to accelerate its adoption, the sooner we will all see the benefits.
