Reality Check: Agentic AI in healthcare IT is more Walkman than Spotify (for now)

This article is part of our ongoing coverage of the recent ViVE 2025 event, co-hosted by HLTH and CHIME in Nashville in February.

Agentic AI has captured the imagination of healthcare IT leaders with its promise of intelligent, autonomous decision-making. But while visions of fully self-governing systems are compelling, today’s agentic AI tools are more like the Sony Walkman — innovative, yes, but still limited in functionality — than the seamless intelligence of Spotify. The truth is, healthcare organizations are only beginning to experiment with agentic AI, and even the most advanced implementations require considerable oversight.

It’s important to clarify that agentic AI is designed to act with human-like autonomy to complete a specific task. It can execute on instructions according to its own idea of best practices for getting to the right answers and can adapt and improve its approach over time by learning from experience. Not all “AI agents” are agentic. Some are relatively simple, like a remote process automation (RPA) bot that automates billing tasks based on pre-defined rules. These systems can efficiently follow instruction but may struggle with unexpected scenarios and complex tasks. In contrast, agentic AI represents a more advanced form of AI agent, designed to operate with greater independence and adapt dynamically to changing conditions.

Rudimentary agentic AI tools in play

The gap between current capabilities and the futuristic vision of autonomous agents is significant. In a ViVE 2025 session called Agentic & Robotic Automation: A New Era (recording; registration required), Jason Warrelman noted that most organizations are still in the exploratory phase, piloting tools that combine sophisticated automation with limited adaptive capabilities. These early tools may excel at predefined tasks — such as threat detection, basic pattern recognition, or automated alert generation — but they aren’t yet capable of fully independent, strategic decision-making.

To understand the difference between agentic AI and traditional automation, consider this explanation from Yan Chow, Global Healthcare Strategist at Automation Anywhere, in his ViVE session called The AI Agentic Revolution in Healthcare (recording; registration required): “So the idea is that the data input is the sensing function of an AI agent. The LLM is actually the ‘thinking’ piece. And then the automation we previously called RPA is the ‘doing’ piece. So, thinking, taking data in, thinking and doing, that’s the AI agent.”

Warrelman described current agentic AI systems as “highly guided” rather than truly autonomous. Much like a Walkman requires the user to select a tape, flip it over, and manually adjust settings, today’s agentic AI systems require significant prompt engineering and ongoing refinement. These systems follow defined paths but lack the nuanced flexibility of fully agentic AI tools.

Emerging agentic AI use cases in healthcare IT

Despite these limitations, rudimentary agentic AI tools are already supporting healthcare organizations in various ways. Key examples from the transcripts include:

• Threat intelligence augmentation: Early AI agents are helping identify suspicious network behavior faster, using pattern recognition to highlight anomalies that might otherwise be missed. However, these tools still require human security analysts to validate and act on insights, Warrelmann advised.
• Automated response systems: In some cases, basic AI agents are being trained to initiate containment protocols, such as isolating compromised devices or restricting network access. While effective for predictable threats, these systems struggle with more complex, adaptive attacks.
• Enhanced vulnerability management: Agentic AI tools can automate patch prioritization by correlating threat intelligence with asset inventory data, improving response times. But these systems still depend heavily on predefined rules rather than true strategic decision-making.
• Care coordination and post-discharge monitoring: AI agents are being used to assist in care coordination and post-discharge monitoring, helping to standardize care and proactively intervene in patient care, according to Chow.
• Revenue cycle management: Agentic AI is being applied to revenue cycle management, including automating tasks such as denial appeals and status checks, as well as analyzing and categorizing denials, Warrelmann assured.

Building an AI agent

When considering the development of AI agents, it’s important to move beyond traditional use cases. As Warrelman pointed out in his ViVE talk, “Agents aren’t built like use cases, they’re built on intent because humans work on intent. So, when you build an agent, you want to build it on intent, not use case.” This intent-driven approach is crucial for creating effective and adaptable AI agents.

The Spotify future: Fully autonomous agents

Healthcare IT leaders shouldn’t mistake these early wins for fully realized agentic AI. As Yan Chow warned in his ViVE session, the real promise—AI systems that autonomously detect, analyze, and respond to complex threats with minimal human involvement — is still a long way off.

It’s also important to remember that agentic AI is different from generative AI, which is focused on creating new content based on what it knows about its training data. AI agents can harness generative AI capabilities to complete their assigned tasks, because sometimes the agent needs to generate something new as a step toward solving a problem or meeting an objective.

Developing systems capable of continuous learning, context-aware decision-making, and adaptive response remains an ambitious goal. Agentic AI typically leverages a combination of established AI strategies, including large language models (LLMs), natural language processing (NLP), computer vision, machine learning, deep learning, neural networking, and automation technologies. Bringing all of these approaches together in the right way can create tools that require little in the way of active intervention by their human creators.

In a ViVE session titled You Say You Want an Ambient Evolution, Joe Petro, CVP, Health and Life Sciences, Microsoft, spoke about the trajectory of AI in healthcare. Petro noted that “over the last 18 to 24 months, what’s really happened is the novelty of capturing speech and doing the listening … It has, more or less, become table stakes … ambient intelligence is really the direction that it’s actually going in.” He urged the audience to “just imagine for a minute the pervasive infusion of intelligence into every single workflow for every single physician, for every single nurse, for every single radiologist, every episode of care that touches every single patient.” This, Petro said, “is actually what’s coming when you hear people like me bloviate about agentic AI.”

The path forward will require:

• Improved Data Integration: Today’s agentic AI systems struggle to process fragmented data from disparate healthcare IT environments. More effective integration will be critical to improving their adaptability.
• Better Guardrails and Governance: As agentic AI capabilities expand, healthcare organizations will need robust oversight frameworks to ensure systems operate ethically and securely.
• Enhanced Explainability: Trust in agentic AI tools will depend on their ability to provide clear, understandable rationales for their decisions.

Addressing challenges and risks to move forward

Implementing AI agents in healthcare settings presents unique challenges and risks. During his presentation, Chow highlighted several concerns: “I think that, you know, for healthcare, the biggest risk is to patient safety. There’s also risk to organization. There’s a risk to regulatory compliance. And of course, one of the big concerns is data privacy and data security.” Addressing these risks is crucial for the responsible adoption of agentic AI in healthcare.

For now, healthcare IT leaders should manage expectations. While agentic AI offers exciting possibilities, its role in healthcare IT today is more about augmenting human expertise than replacing it. Much like the Walkman marked an important step toward portable music, these early agentic AI tools represent meaningful progress — but a Spotify-level revolution is still on the horizon.