The artificial intelligence applications most people have grown familiar with in recent years usually take the form of assistants: chatbots answer questions and respond to prompts, but a human instigates each step.
An AI agent, on the other hand, can take action on behalf of a user. Models with agency have existed for a few years, but recent advances in natural language processing and memory structure have made them much more powerful. An AI agent can conduct an entire task – starting an action and completing it.
An agentic AI system is a network of agents that can work together to complete an entire workflow, not just an individual task. In the healthcare context, this could mean reviewing and processing a complex insurance claim. Using natural language, an agentic AI system can plan, collaborate and connect.
Virtual co-workers
For individual AI agents and for systems, people intervene only to take care of problems as they crop up and to ensure oversight. AI agents, in a sense, are virtual co-workers – and very efficient ones.
"There are use cases across the healthcare value chain and from the beginning to end of many individual tasks," said Jessica Lamb, a partner in the social, healthcare and public entities practice in the New York office of McKinsey & Company, a research and consulting firm. "Think of a hospital, for example, that has to order a wide range of supplies, from gowns and gloves to high-end medical equipment.
"AI agents can be deployed to flag when items are running short, source the best supplier, initiate the purchase order and make the payment," she explained. "Agents also can be used to support patients, helping them prepare for upcoming appointments, streamlining discharges and coordinating case management."
Agentic AI is a good fit for healthcare for several reasons, she added.
An easier transition
"For starters, healthcare already is heavily dependent on information technology, and agentic AI can work with existing software tools and platforms," Lamb said. "That can make the transition somewhat easier.
"Another example is how the healthcare industry continues to rely on many manual processes, based on legacy technology and practices," she continued. "As the examples I used indicate, AI agents can perform a wide range of complex but repetitive tasks that, for a variety of reasons, have not yet been automated."
And the benefits could be substantial. The American Hospital Association estimates administration accounts for 40% of hospital expenses.
For hospitals and health systems looking to deploy agentic AI, getting it right can be challenging. So where is the best place to start? Lamb said, at the beginning, of course. But what is the beginning, according to her? Follow the need.
"Figure out where the highest potential for implementation is, such as the complex processes and workflows that involve a lot of touchpoints and handoffs and that are more manual than they have to be," she suggested. "Once that is done, set actionable goals – agentic AI is not just a shiny new technology, it needs to be results-driven.
"It's important not to get stuck in pilot purgatory – an application in one department, an experiment in another," she continued. "Unfortunately, that is where many organizations are, and it can be difficult to escape. The better approach is to pick a whole domain or work area, rather than a specific task."
Changing how an organization functions
The greatest benefit of agentic AI comes when it is scaled and changes how an organization functions, not when it is used simply to cut costs here and there, Lamb believes.
"Think of it this way: A use case is generating an appeal letter – a domain change is reimagining revenue cycle management," she said.
Finally, embed change management from the beginning – if agentic AI is implemented well, it will transform the way the whole organization functions, and the way people work, she added.
"It's important to remember the human element," she said. "Agentic AI is not just about tech. It is – or should be – about improving the job experience, creating higher-value work and democratizing technology in a way that fosters human connection."
Deployment requires careful planning
Managing risks and ensuring governance certainly is essential for building trust in agentic AI systems. Hospitals and health systems implementing the technology must plan carefully.
"For ethical, security and management reasons, trust is essential," Lamb stated. "And that means organizations need to take the principles of responsible AI to heart. Here is something that may not be well known: Investing in trust pays off. As my colleague Roger Roberts recently put it, 'When implemented well, responsible AI leads to real ROI.' The reason is simple: Without trust in agentic AI, people will not use it.
"It is up to people to define the parameters of agentic AI autonomy," she continued. "That means structuring the workflow of each agent so it is deliberate about what it can and cannot do."
Another concern, bolstered by well-publicized instances of AI glitches, is the possibility of biases in inputs and outputs. That requires ensuring data accuracy and security.
"Keeping humans in the loop is a critical safeguard," Lamb advised. "From the start, people need to be at the center of agentic AI development. That means education, in the form of a clear communications strategy that informs everyone about how and why the organization is planning to use agentic AI. To ease concerns about the possibilities of job losses, there also should be a clear strategy to re-skill existing employees.
"A transformation of this breadth and depth is a general responsibility," she continued. "Good governance therefore requires mechanisms to report problems and cross-functional collaboration to ensure all perspectives are included."
The principle is simple: Just do it – but do it thoughtfully, she added.
The future of agentic AI in healthcare
So, what does the future of agentic AI look like in healthcare? What is next for hospitals and health systems? Lamb puts it simply.
"The future of agentic AI is readily expressed: There will be more and more of it," she predicted. "Older, manual and slow systems are on their way out – and that is a good thing. The use of AI agents will ease some of healthcare's most annoying pain points while allowing people to focus on more interesting and novel work.
"The process will not be straightforward," she cautioned. "Some hospitals and health systems will be faster and defter than others. But the trend is clear: Agentic AI is the future."
Follow Bill's HIT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.
WATCH NOW: Helpful tips on smart rooms from OhioHealth
