AI in the Clinic: Three Ways It’s Showing Up in Healthcare Today

In a modern clinic, a physician listens to a patient with her full attention, while an AI scribe transcribes the conversation and drafts clinical notes for review. Down the hall, a radiologist opens a scan that AI has flagged for signs of stroke and pushed to the top of the queue. In another exam room, a clinician working through a puzzling set of symptoms opens ChatGPT to think it through.

None of this is on the horizon. It's happening now, in clinics and hospitals across the country. Artificial intelligence has already arrived in healthcare, not as a single breakthrough but as a presence woven through the clinician's day, in applications both approved and informal.

AI Is Already in Healthcare, at Scale

Scenes like these are no longer unusual. AI has moved into everyday clinical practice faster than almost anyone predicted. More than 80% of physicians now use AI in their practices, more than double the share just three years ago, most commonly for tasks like clinical documentation. More than 1,500 AI-enabled medical devices have been cleared by the FDA, roughly three-quarters of them in radiology. And beyond the tools that are formally approved or deployed, many clinicians are quietly turning to general-purpose AI to help think through their toughest cases.

In current practice, clinical AI adoption falls into three broad categories:

• Clinical workflow assistance (AI in the background): Tools that lift administrative load. They don't make clinical decisions; they handle the documentation, coding and note summarization that pull clinicians away from patients. Ambient scribes that listen to a visit and draft a structured clinical note for the patient EHR are the fastest-moving example, and the most invisible, because they touch clinical judgment the least.
• AI-enabled medical devices (AI in the instruments): Systems built into medical devices that analyze images, waveforms and other clinical data for specific tasks, like flagging a suspicious cluster of cells on a mammogram or detecting a heart arrhythmia in ECG data. They're the most established in terms of FDA oversight, tied to defined clinical uses and reviewed through device pathways.
• Clinical decision support (AI in the loop): Tools that help clinicians make decisions at the point of care. This is the broadest and least settled category, spanning familiar EHR alerts, predictive tools that flag patients at risk of sepsis or deterioration, and the informal use of general-purpose AI to check information or work through a diagnosis. It sits closest to the clinical decision itself, which makes it both the most promising and the most scrutinized. The recently launched ChatGPT for Clinicians signals a shift from informal use of general-purpose LLMs to more clinically-focused models.

The Path Forward for Clinical AI

AI is advancing quickly, in some cases faster than the regulatory frameworks and clinical purchasing systems around it can adapt. That gap doesn't make these tools any less promising, but it does mean the path forward looks different for each of the three categories. What's a short hop to adoption for one can be a long climb for another.

Across all three, though, the same three questions determine how far and how fast a tool will go:

• Regulatory pathway. Where a tool sits on the spectrum from administrative support to clinical decision-making largely shapes how it's regulated. Workflow tools often fall outside FDA device oversight entirely, which is part of why they've spread so fast. AI-enabled medical devices sit at the other end, tied to defined clinical uses and reviewed through established device pathways. Clinical decision support is the moving target, where the level of oversight depends on how transparent and verifiable the tool's recommendations are.
• Ease of adoption. Clearance or low regulatory burden doesn't guarantee uptake. Tools have to fit real clinical workflows to earn a place in them. Alert fatigue, fragmented patient data and uncertain reimbursement can all slow adoption even when a tool works as intended, and the friction looks different for an ambient scribe than for a diagnostic algorithm.
• Trust. Ultimately, every one of these tools has to earn clinician confidence. That means is must perform reliably across different patients, settings and data, and hold up in practice rather than only in testing. Trust also raises still-unsettled questions, like who bears responsibility when a clinician acts on an AI recommendation and the outcome is poor.

The regulatory picture, at least, is becoming clearer. Clinical workflow tools that simply handle documentation generally fall outside device regulation, while AI-enabled medical devices that analyze a medical image or physiological signal are regulated through established device pathways. Clinical decision support is the hard middle, and it's where the FDA moved most recently. Updated final guidance on clinical decision support software issued in January 2026 sharpened the line: a CDS tool can sit outside device regulation when it supports rather than drives a decision and stays transparent enough for a clinician to review the basis for its recommendation, but the more it works as a "black box" or edges toward time-critical predictions, the more likely it is to be regulated as a device. AI guidance is expected to continue to evolve alongside the tools themselves.

Ultimately, adoption comes down to the clinicians themselves. A tool can clear every regulatory hurdle and still go unused if it doesn't earn a place in real clinical work. Does the tool save time or add a step? Does it surface something the clinician would have missed, or just add noise? And can they trust its output enough to act on it, across the range of patients and conditions they actually see?

From Promise to Practice

Building AI that genuinely works in clinical care takes more than a strong algorithm. It means engineering it into a device or workflow that clinicians can actually use and navigating a regulatory path appropriate for the tool. That's the kind of work Battelle is built for. We bring together device engineering, regulatory strategy, data science, human factors and clinical insight to help developers move AI-enabled tools from promising idea to real-world use, safely and effectively.

AI is rapidly transforming clinical care—and it’s just getting started. AI-based tools won’t replace physicians, nurses or other clinicians. But they do have the potential to reduce administrative burden, free clinicians to spend more meaningful time with patients, speed the time to accurate diagnosis, and identify dangerous conditions sooner. Getting more of these tools into the clinic is a win for both patients and clinicians.

Learn more about our medical device development capabilities.