As buzz around conversational AI reached a fever pitch last year, a new term began gaining traction: “AI agents.”
Over the past twelve months, Google searches for AI agents have increased nearly tenfold. While the category of “agentic AI” may seem to have appeared from nowhere, it has quickly become one of the most talked-about trends in tech.
Yet, descriptions of agentic AI are still largely abstract, making it difficult to grasp what sets it apart or how it might apply in healthcare.
One tells us an AI agent is: “A system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilizing available tools,” and another offers that an agent: “…perceives its environment, takes actions autonomously in order to achieve goals, and may improve its performance with learning or acquiring knowledge.”
While technically accurate, these definitions aren’t especially evocative or helpful. Worse, some blur the distinction between AI agents and traditional chatbots.
Is this just another case of vendors drumming up new terminology to generate excitement? Or window dressing – giving a new moniker to the same old chatbots? Rebranding what’s really just an incremental step in the evolution of Interactive Voice Response (IVR)?
The answer is an emphatic no. Agentic AI is not marketing hype. It represents a fundamental shift in how automation works — especially in patient access — and holds promise to transform the experience of receiving care.
Understanding and Solving the Patient’s Goal
To understand Agentic AI, it helps to take a step back and consider the scenario of a patient calling a health system to either learn something or to get something done. These technologies are attempting to solve two distinct aspects of that contact: determining what the patient needs (the goal), and achieving it.
The traditional IVR in a call center gives a simple example of this. A patient hears a menu (“Press 1 for general information, 2 for scheduling, 3 for billing, …”), and follows the breadcrumbs until they get to the hold music that (eventually) becomes the right agent with (hopefully) the tools and knowledge to achieve their goal.
Traditional conversational AI, voicebots, chatbots, and Intelligent Virtual Assistants (IVAs) have improved this experience through natural language understanding. The patient states their intent directly to the bot (e.g. “Do you take Aetna insurance?” or “I need to cancel my appointment”) versus navigating voice menus.
These bots behave like a railyard switching station, using intent recognition — often powered by keyword matching or machine learning classifiers like Deep Neural Networks (DNNs) trained with a large set of utterances and paraphrases — to route the patient to a predefined FAQ answer or automation script.
These automation scripts are indeed like railroad tracks – static, deterministic – and don’t allow for variance, say the impatient patient that tries to give all the information up front; or complexity such as combinations of functions to be done together. They can’t respond empathically to the patient’s specific articulated situation. Solving new use cases is cumbersome and resource-intensive, requiring conversational design, training data, intent modeling: deep technical know-how.
To use another travel analogy, chatbots developed using traditional conversation AI are like AAA TripTik booklets or Mapquest directions printed out from the Internet. They are an extremely useful improvement from the dog-eared fold-out maps we used before. They are also static and impersonal.
Agentic AI however is like turn-by-turn GPS built right into your dashboard. It’s interactive, dynamic, and constantly adapting to new conditions in real time.
How Agentic AI Works
Agentic AI is generative, the patient’s goal and the steps to achieve it are determined live during the conversation. There are no switches, classifiers, scripted branches or hard-coded flows. Instead, the system dynamically plans the optimal next step based on what the patient says and what it has access to.
Think of it as an autonomous rail-laying machine: it builds a personalized track on the fly.
This is possible thanks to the capabilities of modern large language models (LLMs). These models understand language context, can follow complex instructions, can reason through multi-step processes and are fast enough for real time voice conversation.
To function effectively, agentic AI needs:
- Instructions and SOPs: Clear guidance on policies, workflows, and decision logic.
- Tools: Access to systems like EHRs for scheduling, data retrieval, and authentication.
- Knowledge: A corpus of documents, FAQs, protocols, and resources to consult when answering questions.
- Escalation logic: A graceful fallback to human agents when confidence is low
Let’s look at two scheduling examples.
Simple Task: Reschedule an Appointment
With traditional automation, the system follows rigid steps to authenticate, locate the correct appointment, and then asks multiple time-consuming questions to attempt to identify the optimal slot.
Quite often, the frustrated patient interrupts the process with “Agent! Agent!” or repeatedly presses “0.”
With agentic, the patient might simply say “I’d like to reschedule my upcoming dermatology appointment, ideally during lunchtime sometime next month.”
The system understands the request, authenticates the patient using a single input, identifies the relevant appointment, finds matching slots, and confirms the new booking — all in a natural, single conversation turn. Or if there are no matching slots, the bot suggests applicable alternatives. Just as a skilled human agent would.
Complex Task: Coordinating Multiple Diagnostics
Consider a patient with COPD who needs a pulmonary function test, 6-minute walk test, chest X-ray, blood draw, and a pulmonologist visit — ideally all in the same morning.
But there’s also additional complexity. Diagnostics at the same office should be scheduled back to back. There needs to be extra time allocated to navigate the facility, particularly considering the patient’s mobility challenges. The pulmonologist appointment should happen after the diagnostics in order to review with the patient the results.
This scenario is beyond the reach of traditional conversational AI. The scripting would be too complex, and the scenario too niche to justify the development costs. This would be accomplished by a human scheduler who is trained on how to understand these interdependencies and assess the available timeslots accordingly.
With agentic, that’s essentially how it works with the AI. The LLM is “trained” through instructions that clearly articulate in English these considerations. The AI evaluates scheduling options and makes a decision: book it all on one day if possible, or split across two if necessary — just like a human scheduler.
These are just a few examples of the impact that agentic AI can have on the myriad of administrative events that encompass a patient’s care journey.
Aligning Agentic with the Quadruple Aim
The ‘Quadruple Aim’ is a framework established a decade ago to address systemic challenges in US healthcare. It seeks to improve the quality of care, reduce per capita costs, enhance the patient experience, and improve the work life of care providers
Much attention has been paid to how AI can enhance clinical quality — from predictive analytics to imaging interpretation to ambient documentation.
Agentic AI, especially in patient access, targets the other three aims: to lower costs by reducing staffing demands in high-attrition contact centers, improve the employee experience by automating rote admin work, and transform the patient experience by removing friction from every interaction – scheduling, prescriptions, billing, referrals, eligibility, prior auth, claims
These are the everyday administrative burdens patients face. Until recently, most were only solvable with human effort. But now, LLMs have crossed a threshold: capable, contextual, and fast.
This journey is just beginning — there will be challenges, missteps, and learning curves. But the direction is clear. Healthcare organizations ready to embrace this new paradigm will not only reduce friction and cost, but also transform the care journey into something far more humane, responsive, and effective.
Soon, when patients say “Agent! Agent!”, they just might be asking for the bot — not the human.
About Chris Ingersoll
