There’s a new type of AI bot on the block—and this time, it’s completely autonomous.
AI agents are steadily making their way into the public consciousness as more companies release them. Last year, generative AI was all the rage, producing ambient scribes that could transcribe a conversation into clinical notes and in-box bots that could draft responses to patients’ MyChart messages. This year, however, the spotlight has shifted to agentic AI, which can initiate a task and complete it—start to finish—without human intervention or oversight.
It’s widely considered that this technology could change the way health care organizations function. Earlier iterations of AI could make humans’ work easier, more efficient or more accurate, but AI agents can work independently of us. While generative AI can answer your questions, agentic AI can pose its own and even reason through them.
AI agents were a hot topic at the Healthcare Information and Management Systems Society (HIMSS) conference in early March, and the buzz has only grown since then. On April 23, Nvidia launched a new platform to help companies build AI agents, which it calls “AI teammates.” The company estimates that the agentic AI market is worth $1 trillion, according to The Wall Street Journal. Other projections anticipate rapid market growth, from $7.8 billion in 2025 to $56.2 billion in 2030.
Photo-illustration by Newsweek/Getty
But health systems have taken a more cautious approach to AI than their counterparts in business and tech. Although many health care organizations are experimenting with AI, only 30 percent of their pilots and proof-of-concept projects make it to the development phase, according to a recent report from Bessemer Venture Partners, Amazon Web Services and Bain & Company.
As agentic AI introduces even more capabilities and risks, it can further complicate the gradual rollouts underway at health care organizations: so Newsweek connected with 10 agentic AI developers to learn exactly what leaders can expect from the technology.
How are health systems using agentic AI?
AI agents have been deployed in various departments across hospitals and health systems, from the revenue cycle to the clinical decision-making process. They’re yielding strong results—according to the technology companies that create them and the researchers who are examining them.
Cedar, a patient financial platform for health care providers, launched an AI voice agent on April 29 to automate patient billing calls. Two days later, Zocdoc announced an agent to automate scheduling calls. Both companies said that their agents could speak conversationally and answer phones 24/7, freeing up operators to focus on more complex requests.
Care providers are also using agentic AI. Google Cloud collaborated with more than 50 health care providers at Seattle Children’s to develop Pathway Assistant, an AI agent that can synthesize information from clinical standard pathways. The tool is expected to increase compliance with standard care processes and make it easier for physicians to access the information they need, the companies said. It would take 15 minutes for a physician to conduct this search manually, but the agent can do it in seconds.
AI agents are improving system-level efficiency too. Take Kontakt.io for example: It is designed to orchestrate the chaotic reality of hospital operations by continuously gathering, reasoning through and acting on real-time data. The platform starts by graphing a comprehensive map of how patients, staff and clinical spaces interrelate, using live feeds from electronic health records, staffing schedules and inventory management systems (supplemented by Bluetooth technology). This allows it to predict problems, like equipment shortages or staffing bottlenecks, and initiate efforts to prevent them.
Kontakt.io uses a team of AI agents, each assigned a narrow task to focus on constantly. While one agent monitors the availability of clean equipment, another might keep watch over a predictive model that calculates future equipment demands. When the robot squad senses trouble, it will ping another agent to call the biomedical department and relay the message to a human coordinator: for example, “We need to move six pumps to the ICU in the next 45 minutes, and there are five broken pumps in the cardio unit. Maybe you should go take care of the broken pumps, then grab another from room one and take them to the ICU.”
Since AI agents can converse, the human coordinator can respond with questions about the inventory or tell the agent that they’re busy and should call back later. The agentic team is constantly triaging issues to flag the most pertinent problems first, based on both real-time data and the predictive model’s concerns.
Historically, humans have made these small decisions themselves, but they didn’t have the time or bandwidth to coordinate with one another, Rom Eizenberg, Kontakt.io’s chief revenue officer, told Newsweek. Agentic AI can serve as the middleman, eliminating guesswork that causes accidental clogs.
“It’s a jungle out there,” Eizenberg said. “The 1,400 vendors, the unstructured data, the siloed behavior, the millions of phone calls to make everything work: that’s the root cause for all the evils in health care.”
Can AI agents make call centers more efficient?
Communication is a major hiccup in the health care industry; as of 2019, 70 percent of health care providers still used fax machines. Patients, providers, payers and pharmacies frequently swap info by phone (to all parties’ dismay).
About half of patients are satisfied with the service at their health care provider’s call center, according to a 2023 survey of 200 senior leaders. The average hold time at these organizations was 4.4 minutes, well above the HFMA’s recommendation of 50 seconds.
When patients can’t reach their health care providers, they may seek instant answers elsewhere: turning to social media and search engines, which are rife with misinformation.
“I wish I had thousands of doctors that could do every single phone call to every patient, every outreach, every follow-up,” Dr. Jackie Gerhart, a family medicine physician and the chief medical officer and vice president of clinical informatics at Epic, told Newsweek. AI agents can help fill the gap: calling patients to check in after missed appointments, scheduling upcoming labs and even talking through top concerns so a patient’s doctor can prepare for their visit before they enter the exam room.
AI agents can also handle the back-office phone calls, which present their own pricey challenges. During Medicare Advantage reverification season—from January 1 to March 31—health care organizations staff seasonal workers to handle heightened call volumes as they confirm patients’ insurance plans.
Enter machines, which have a higher tolerance for hold music than the average human being. Tech company Infinitus deploys AI agents to help alleviate the pressure on health care call centers, especially during busy seasons.
Last year, the company’s AI agents spent more than 1.4 million minutes waiting on hold, CEO Ankit Jain told Newsweek. This January alone, they spent more than 2 million minutes navigating interactive voice response systems (the more archaic version of a robotic call assistant, known for asking callers to “press one if you are an existing patient, press two if you are a pharmacist, press three if …”).
Agentic AI is far savvier, according to Jain and other solutions developers. While researching for this article, Newsweek‘s health care editor spoke with two AI agents and did not feel inclined to yell, “I need to speak to a representative!”
“The conversational AI voice platform that we have built is extremely natural, extremely conversational, and it’s akin to the ones that you would have if a human picked up right away,” Jain said.
Does agentic AI hallucinate?
If agentic AI is going to be carrying conversations and informing doctor’s decisions, it needs to meet the same standards as a call center representative or a board-certified physician. Although numerous studies have examined generative AI for hallucination, there isn’t extensive data on agentic AI.
On April 24, Infinitus launched AI agents that it “guarantees” are hallucination-free. Newsweek asked the company’s technology lead, Shyam Rajagopalan, how he could be sure.
The company’s AI agents are confined to hyper-specific sets of data, according to Rajagopalan. For example, if it’s calling to verify a patient’s information, it will access that individual patient’s information—not information from every patient within the health care system.
“Because I’ve constrained the space to only be relevant for this particular patient, [the agent] will never be able to tell you a different patient’s birthday or a different patient’s diagnosis,” he said.
Color Health—a health tech company focused on cancer care solutions—is also working to reduce AI hallucinations by using agentic models. It developed a “large language expert” that merges the strengths of an LLM with the structure of an expert system. Unlike traditional LLMs, which can invent plausible-sounding (but incorrect) outputs when faced with ambiguity, the LLE forces reasoning through structured clinical decision factors (the individual yes/no questions that an AI system parses from clinical guidelines) and Boolean formulas (the strict rules for combining the answers to those yes/no questions into a final recommendation). Since the LLM’s role is limited to answering specific questions rather than generating broad narratives, errors are easier for the model to catch and correct, according to a recent study from the company.
“Agentic AI goes beyond generative AI with the proactive performance of tasks,” Othman Laraki, CEO of Color Health, told Newsweek. “While generative AI creates content by learning from different data sources and patterns, agentic AI is an autonomous, decision-making technology that takes action based upon its learnings.”
Is agentic AI going to replace generative AI?
The future will include both generative and agentic AI, according to Gerhart and her colleague Sean McGunigal, Epic’s director of AI.
“There are going to be cases where the simpler forms of AI make sense, especially if we look at it from a compute-saving or cost-saving perspective,” McGunigal told Newsweek. “If we don’t need the heavier firepower of an agent, we won’t go that route—but I think you will see more and more automation in the form of agents.”
We shouldn’t think of agentic AI as an evolution of generative AI, per Kontakt.io’s Eizenberg. It’s something entirely different—not a system upgrade, but a stand-alone invention, set apart by the agents’ ability to connect with one another.
“An LLM is a transformer, the enabling tool to talk to people or reason or make decisions,” Eizenberg said. “But it isn’t software architecture. Agentic AI gives us ways to build that we never had before.”
Is agentic AI going to replace humans?
Yes and no. AI agents may cut down on call center staff, but it’s unlikely that they’ll ever stand in as your doctor.
Three health system and AI researchers—including Dr. Eric Topol of the Scripps Research Translational Institute—explored the question in an April comment for the academic journal Nature Biomedical Engineering.
“As AI continues to advance, physician-independent workflows are likely to emerge in certain areas of health care,” the authors said. “These workflows may be driven primarily by collaborations between clinical and operational AI agents and may streamline processes, optimize resource utilization and improve patient outcomes.”
But these physician-independent workflows “will not be suitable” for other areas of health care, such as complex cases and rare diseases, according to the authors. In these instances, AI agents can still support physicians by offering insights and optimizing workflows.
Doctors’ roles have evolved as the health care industry has grown more complex. Some physicians, like Gerhart, are optimistic that AI agents could assume some of that work to help provide more thorough, comprehensive patient care.
“When I think about what it means to be a doctor in the future, it’s not only doing individual workflows and only knowing medical knowledge,” Gerhart said. “It’s knowing how to manage and give the best care. So I’m hopeful that my team of care coordinators and AI agents can work together to make sure that the patient actually gets everything they need done at the appointment, their population health is taken care of, their family dynamics are considered.”
“It’s really this opportunity to reimagine what medicine can be, and the extent of medicine that you can do, with the new tools that you have.”
