Progress in artificial intelligence has been from machine learning models, which analyze data to algorithms that can act on it with minimal human involvement in clinical, administrative and addressed to patients.
Agentic AI, or autonomous agents, will compensate for standard generative artificial intelligence, because it is able to make autonomous decisions to achieve the goal, and not only producing production from the contribution to support human decisions or improvement.
The growing Agentic AI pace in healthcare was a scorching topic at the global conference and the Himss25 exhibition, from the show hall to educational sessions and speeches. But the whole discussion pointed to a key challenge for healthcare: the employ of agency technology to improve surgery is one thing, but careful navigating after the stern complexity of the employ of autonomous artificial intelligence in clinical care is completely different.
AI agents can lend a hand doctors meet the demand for care as accelerators and machines for ordinary tasks. But they are not a substitute for making decisions by a doctor. Since healthcare organizations are looking for AI agency to service more elaborate, multi -stage work flows in and outside EHR, be careful when implementing these tools in clinical conditions.
Management and health policy AI
The controversial law introduced in the American Chamber of Representatives in January and referred to the home and trade home committee, HR238 Will change the federal act of food, medicines and cosmetics to enable AI and ML, which were authorized by Food and Drug Administration to qualify as a doctor entitled to prescribing drugs.
But many technology programmers believe that autonomous artificial intelligence is not ready for employ for prescribing and other aspects of care.
Several clinical leaders and IT, with whom we talked before, during and after Himss25, mentioned the importance of building trust and ensuring the transparency of its functionality as the greatest obstacles to the adoption of care management systems.
And in Himss25, AI experts insisted on caution. For example in Himss AI on the healthcare forum at the beginning of this month, for example Dennis Chornenky, AI main adviser in UC Davis Health has noticed that management could not keep up with the progress of technology, and the current provisions regarding AI health were designed for machine learning models-not for autonomous AI decisions.
Move too quickly with AI agents, and the result may be biasing decisions and a potential risk for patients’ safety. Croatnky said that health care directors are under pressure from their acceptance, “but they must ensure safety”, preparing for AI and interference.
“How do they do it if they don’t have the right management mechanisms?”
Agentic AI and patient observation
The main driver of the agency artificial intelligence in all organizations is to alleviate the computer load at clinicians and administrative staff.
Epic integrates artificial intelligence between applications and positioning agents to improve clinical performance, said Seth Howard, Vice President for Research and Development at Epic.
“For example, we expect AI agents to help you prepare before a visit, talking to patients about their needs, identifying the missing tasks (such as the laboratory), helping in the schedule and performing these tasks and creating an easy-to-read summary,” he said.
Garrett Adams, Vice President of the Epic for Research and Development, used an example at the patient’s assistant posterium conference.
The multimodal continuation of the wrist surgery, which uses the voice and image, reaches the patient, “talking to them in conversations,” he said. He asks the patient about the recovery of the wrist and sharing her image or his video.
The agent can deduce the angle that the patient can bend the postoperative wrist and compare it with patients at the same recovery point at the Cosmos Epic database. Based on this point in recovery, the agent suggests that the control meeting may not be needed anymore, because the patient “incredibly good”.
The agent asks the patient if he wants to cancel the meeting and sends the patient’s request to cancel the care team to confirm.
“So still having this human verification for these further things, make sure the guy is right,” said Adams.
While the assistant assistant of the postoperative patient is still in the patient, he can also offer a visit to flu.
“Combining video contribution, patient communication, conversation tone, COSMOS data view, closing the gaps in the care and planning function in the aggression flow of work – this patient had quite smooth interaction,” he said.
Also in Himss25, Zoom Workplace for clinicians launched a public version of the beta of environment notes for employ during visits to teeth, as well as for personal meetings.
“When you finish at the meeting, as a doctor, you get a note quickly, and then browse and edit,” said Ritu Mukherjee, product head, acceleration and readiness in Zoom, in an interview at the Himss25 Show Floor hall. “And then pressing the button you just send it to the EHR system,” she said. “This is just a starting point for automation of clinical work flow.”
Developing AIC AI with caution
Meanwhile, in Himss25, the many years of interoperarative infrastructure and IT INTERSYSTYS infrastructure debuted by a modern EHR powered by artificial intelligence called Intelllicare. In addition to listening to the environment and generative AI functions, in order to simplify administrative tasks, the AI assistant can generate automatic summaries of patients’ history and aggressively prepare codes required for settlements in the revenue cycle management system.
The seller undertakes a cautious approach to the employ of Agentic AI in directly making clinical decisions, emphasizing human supervision and focusing preliminary efforts on administrative tasks. For example, adding to the diagnostic list.
“I have the impression that Agentic AI and Open Source models are two topics of the year in 2025, from the AI point of view,” said Don Woodlock, head of Global Healthcare Solutions, before Himss25.
There is a potential to raise the administrative burden of clinicians by automating tasks, such as sending letters, processing orders and servicing previous permits, enabling AI to think about the necessary set of steps and enabling them to collect, he said. He goes beyond the environment of the environment: “by performing subsequent actions and updates on the chart, and we will see how far we have gone.”
He added that in the case of higher -order work flows in EHR, developers may “not necessarily” push the AI agency. “What are the advantages and disadvantages of this? What are the dangers? Do we remove the level of medical supervision?”
Thanks to the automation of processes, such as sending prescriptions and orders, the general director of Eclinicalworks, Girish Navani, said he was more likely to pump a break from AI agent.
“There is time and place where agents will appear even with the support of clinical decisions,” said Navani in Himss25.
“But this is not a statement of a problem for today. We focus on managing the income cycle, taking administrative overload and work done today-who has no clinical decision making-and creating autonomous agents to work, while everyone is sleeping.”
Eclinicalworks showed practical applications of Agentic AI, such as the AI Intelligence Intelligence tool, which automatically separates the patient’s data from incoming documents – in PDF, fax or ccda formats. The company said that it can accurately adjust the data to patients for 75-85% of these documents. There is also an agent of Genie Healow, which can provide patients with immediate answers to joint queries 24/7 in a voice, text or chat.
However, in the case of elaborate questions, Genie escalates questions to human agents.
Athenahealth is also gradual in how it integrates advanced functions with his Products supported by AI. Recently, the company offered its clients to listen to the environment, a summary tool and Salesforce agentswhich offers aggressive flows of work for summaries of hospital visits, unsuccessful visits, care gaps and pharmaceutical orders.
In clinical work flows, AI agents automate the supplier’s tasks, such as observing laboratory results and preliminary conditions.
But Alicia Bassolino, vice president of Athenahealth for analytics and AI, also emphasized the importance of human supervision.
“I look at AI, aggressive or not, as the ability to remove some mechanically oriented thinking, creating tasks, things that must enter the system, and thinking, which is their decision making and applying medical knowledge,” said Himss25.
RCM Driving Agentic AI
Now that AI Healthcare technology has moved to the place where it can make autonomous decisions, several EHR suppliers are bolder in the implementation of administrative benefits in RCM.
“You can really orientate generative artificial intelligence around the task or purpose, and it can move around this autonomous or pseudo-autonomous parameter set that you give him,” said Bassolino.
Automation in a broader spectrum of payers, suppliers, specialties and other permutations of the revenue cycle is now possible for artificial intelligence, and suppliers want to deal with this friction.
“It allows us to work with payers or automatize some types of tasks in which there is a set of options that are not perfectly linear and do not necessarily flow,” she said.
Navani said that ECW is also focused on the implementation of Agentic AI in its RCM products.
“We have a large group of RCM ourselves, a thousand employees,” he said. “We came across seven or eight agents who, in principle, could start doing work, what we have and then run them. The products are absurd outside the letters.
He said that in addition to the autonomous processing of electronic data and papers based, the potential for verification of claims coding and compliance control is huge, he notes that the company will have more data on effectiveness this year.