The Double-Edged Sword of AI Double-Edged Sword
( continued from page 9)
What“ AI in Healthcare” Actually Means Today
One reason AI conversations often feel muddled is that we talk about it as a single thing. In practice, health care AI today falls into three distinct categories, each with different risks and rewards.
1. Generative AI: The Language Layer
Generative AI is the most visible shift clinicians feel today. At our organization, ambient documentation has been the first AI capability that many physicians describe as“ finally worth it.” During an outpatient visit, the conversation is captured, structured and transformed into a draft note inside Epic( our electronic health record). The clinician reviews, edits and signs, but the cognitive burden of starting from a blank page disappears.
The real benefit isn’ t just time saved. It’ s attention restored. When clinicians aren’ t typing, clicking and hunting through templates, patients notice. Eye contact improves. Visits feel more human.
Beyond notes, generative AI now assists with summarizing charts, drafting replies to inbox messages and hospitalization summaries. It’ s quietly becoming the connective tissue of health care communication.
2. Predictive and Diagnostic AI: The Pattern Layer
This category includes models that detect risk, classify images, predict deterioration or surface patients who need attention sooner. In imaging-heavy workflows, such as in the emergency department or in oncology, AI is already helping prioritize worklists. Studies with concerning features rise to the top. Clinicians still make the diagnosis, but they do so faster and with fewer blind spots.
In inpatient care, predictive models are increasingly used to identify patients at risk for ICU transfer or delayed discharge. These tools can help hospitalists prioritize rounding and coordinate care earlier in the day, directly impacting a patient’ s length of stay.
These models tend to be narrower and more measurable than generative AI, but they still require careful monitoring. Drift, bias and changing patient populations can erode performance if governance is weak and these changes go untracked.
3. Agentic AI: The Action Layer
The most consequential shift is just beginning. Agentic AI doesn’ t just generate text or scores: it completes tasks. This includes assembling prior authorization packets, drafting referral documentation, routing inbox messages, closing care gaps, reconciling medications and preparing discharge summaries with follow-up already scheduled.
Early pilots in some organizations have shown that these systems can dramatically reduce administrative friction, but only when tightly constrained. Give an agent too much autonomy too soon, and risk rises quickly. This is where AI stops being a tool and starts becoming part of the operational fabric. And it’ s where governance matters most.
How AI Is Changing Patient Care
The clinical impact of AI is less about replacement and more about amplification. In radiology, pathology, dermatology and cardiology, AI increasingly acts as a second set of eyes, flagging findings, prioritizing urgency and reducing the chance that subtle signals are missed during high-volume periods.
In primary care, AI-supported workflows help surface patients overdue for screening, identify rising risk earlier and prepare clinicians with concise summaries before visits even begin.
The broader shift is from reactive care to earlier intervention. When systems can recognize patterns across thousands of patients, long before symptoms escalate, care teams gain time: time to intervene, to educate, to prevent hospital admissions rather than respond to them.
And when documentation becomes a byproduct of care rather than its center, clinicians can focus on what patients experience: presence, clarity and trust.
How AI Is Changing the Patient Experience
Despite the focus on diagnostics, the most immediate AI benefits for patients are logistical. Ask patients what frustrates them most, and you’ ll hear about access: scheduling delays, unanswered messages, confusing bills and care coordination that feels fragmented.
AI is beginning to smooth these friction points. Already deployed at some organizations are conversational scheduling tools, automated reminders and smarter triage pathways that reduce wait times and unnecessary phone calls. Patients receive clearer instructions, faster responses and fewer dead ends. An automated wait list at our organization has allowed for patients of our behavioral health providers to be seen three to four weeks earlier than their originally scheduled appointment.
But convenience alone isn’ t enough. Patients deserve transparency. They want to know when AI is involved, why it’ s used and who remains accountable. Simply disclosing“ AI was used” doesn’ t build trust. Explaining how decisions are made, and ensuring that humans remain responsible, does.
10 LOUISVILLE MEDICINE