Q&A: How AI Is Reshaping the Pharmacist's Role in Medication Safety

Artificial intelligence (AI) is moving through pharmacy practice faster than the profession's ability to validate it, from ambient documentation tools already deployed across hundreds of health systems to medication-intelligence platforms flagging risk across millions of patients. The question facing pharmacists is no longer whether AI will reshape the job but which parts of it AI can safely absorb and which parts demand a human who is accountable for the outcome.

Diyaben Patel, PharmD, is a clinical pharmacist at Union Health in Terre Haute, Indiana, where she practices in an underserved rural community. In the following Q&A, Patel discusses where AI is already delivering genuine value in clinical pharmacy, where it still falls short, and why she believes pharmacists need to move from being end users of these tools to becoming their evaluators and builders. She also addresses how pharmacy schools should be preparing students, where she draws the ethical line between AI supporting a decision and AI making one, and why patient trust remains, in her view, something only a pharmacist can earn.

Drug Topics®: How do you see AI changing the role of the pharmacist over the next decade, and do you think that's a threat or an opportunity?

Diyaben Patel, PharmD: It's an opportunity, and honestly, the transformation is already underway—just unevenly. Look at what happened with ambient AI documentation: tools like Abridge and Microsoft's Dragon Copilot went from pilots to deployment across hundreds of health systems, including places like Mayo Clinic and Johns Hopkins, in about three years. A study published in JAMA this year across five academic medical centers found these tools measurably cut clinicians' [electronic health record (EHR)] and documentation time. That same wave is now arriving at the pharmacy: prior authorization automation, medication reconciliation, [and] prescription verification support.

What that tells me about the next decade is this: AI will absorb the information-processing layer of pharmacy work, and the pharmacist's role will consolidate around what AI cannot do, which is accountability, clinical judgment applied to a specific patient, and, critically, the evaluation of the tools themselves. The FDA has authorized over [1400] AI-enabled medical devices, yet as of early this year it had not authorized a single generative AI device. The regulatory science is still being built, and pharmacists need to be part of building it. The threat isn't AI taking pharmacist jobs. The threat is AI tools being designed, validated, and deployed in medication-use systems without pharmacists at the table. My research at Purdue evaluating AI performance on pharmacy tasks convinced me that our profession's expertise is exactly what these systems are missing.

Drug Topics: In your experience, where does AI add the most genuine value in clinical pharmacy today, and where does it still fall short?

Patel: The genuine value today is in the administrative and operational layer. Ambient documentation is the clearest success story in all of healthcare AI, and pharmacy-adjacent applications are following. AI-assisted prior authorization, benefit verification, refill triage, and 340B compliance workflows. There's also a maturing category of medication-intelligence platforms—Arine is a good example, notably founded and led by a PharmD—that use AI to flag patients at risk of medication-related problems across large populations, which is population-level medication management that no human team could do manually at that scale.

Where it still falls short is patient-specific clinical judgment. This is consistent across the literature and my own research. In a study I coauthored, only 17.5% of pharmacy students judged an AI chatbot's drug information response clinically usable as-is. And the broader evidence agrees: a 2026 study in JACCP testing [large language models (LLMs)] on real internal medicine pharmacy cases concluded that inaccurate and incomplete responses still limit their utility in clinical practice, even with retrieval-augmented systems and prompt engineering. Interestingly, some narrow benchmark studies show top models scoring impressively on structured prescription-review tasks, sometimes rivaling pharmacists. But a benchmark is not a bedside. The gap between performing on curated vignettes and being trustworthy for the complex, ambiguous, incompletely documented patient in front of you is exactly the gap pharmacist-led evaluation research needs to map. That's the work I want to keep doing.

Drug Topics: Patient trust is central to pharmacy practice. How do you think patients feel about AI being involved in their medication decisions, and how should pharmacists address that?

Patel: I practice in a rural community in Indiana, and what I've learned there is that trust is personal before it's technological. Patients trust the pharmacist who knows their name, remembers their kidney function, and answers the phone. If AI gives that pharmacist more time for those interactions—and the ambient AI data suggests it can, with health systems reporting clinicians spending measurably more face time with patients—patients ultimately benefit and rarely object. If AI replaces those interactions, trust erodes fast. And in underserved communities, where the pharmacist may be the most accessible clinician a patient has, that's genuinely dangerous.

Two things matter here. First, transparency: patients should always know when they're interacting with a tool, and there should always be a clear path to a human. The FDA's own draft guidance on AI-enabled devices now emphasizes exactly this kind of transparency and labeling, and I think that principle should extend informally to every patient-facing tool, regulated or not. Second, design intent. I built a prototype chatbot to support patients recovering after chemotherapy, and my governing principle was that it should extend the care team between visits: answer the 2 AM question, reinforce counseling that already happened, never impersonate the clinician, or make a decision. There's also encouraging research suggesting AI tools can genuinely improve access in rural and low-resource settings, which is the population I serve. But access without trust is worthless, and the pharmacist is the trust layer.

Drug Topics: How should pharmacy schools be preparing students for a world where AI handles many of the tasks they're currently trained to do manually?

Patel: The instinct to protect students from AI is understandable and, I believe, wrong. Students are already using these tools; the only question is whether they use them with judgment. And the profession they're graduating into has already changed: a third of providers now have access to ambient AI, health systems are deploying medication-intelligence platforms, and pharmacists at institutions like Jefferson Health are being pulled into AI vendor evaluation and EHR implementation as the resident medication experts. That last part is telling: at the [American Pharmacists Association (APhA) annual meeting] this year, one of the emerging themes was pharmacists serving as the "human in the loop" for AI adoption. Schools should be preparing students for that role deliberately, not accidentally.

But—and this sounds contradictory, [but] it isn’t—schools cannot stop teaching the manual fundamentals. You cannot verify an AI-generated dosing recommendation if you never learned to calculate the dose yourself. Verification without foundational knowledge is just rubber-stamping. The pharmacist of the next decade needs both: the knowledge to recognize what right looks like and structured, repeated practice evaluating AI output against that knowledge. Our research at Purdue was built on that premise. Rather than asking whether students should use AI, we asked how well AI actually performs on pharmacy tasks and had students assess it critically. I'd argue that evaluative skill is now a core clinical competency, on par with literature evaluation, which, notably, we've always taught.

Drug Topics: What does AI literacy actually mean for a pharmacy student, and is the current curriculum keeping up?

Patel: AI literacy is not knowing how to write a clever prompt. It's knowing how these systems fail. It means understanding that a large language model can produce a fluent, confident, wrong answer; knowing the difference between a general-purpose chatbot and a domain-specific system grounded in vetted drug information, because the research shows those perform very differently; understanding what retrieval-augmented generation is and why it reduces, but doesn't eliminate, hallucination; and having a verification habit that engages automatically before anything AI-generated touches patient care.

The encouraging finding from my own research is that students can develop this discernment when given a framework. In our chatbot evaluation study, students didn't naively accept the AI's output; most correctly identified that the response wasn't clinically usable without revision. That's AI literacy in action. But it happened because faculty built a structured exercise around it. Is the curriculum broadly keeping up? No. The research community has moved fast—there are now published benchmarking suites specifically for evaluating LLM safety on medication-related tasks—but most pharmacy curricula still treat AI as either an academic-integrity problem or a novelty elective. The programs that formalize AI evaluation into required coursework will produce the graduates health systems are already asking for.

Drug Topics: Where do you draw the ethical line between AI supporting a pharmacist's decision and AI making the decision?

Patel: The line is accountability. AI supports a decision when a qualified pharmacist reviews its output, applies it to the specific patient, and owns the outcome. AI makes the decision when its output reaches the patient without meaningful human review. That's the line I don't think we should cross, no matter how good the models get. And interestingly, it's a line embedded in U.S. regulation: clinical decision support software is largely exempt from FDA device oversight, precisely on the condition that clinicians can independently review and understand the basis for its recommendations. The moment a tool becomes an opaque decision-maker rather than a reviewable advisor, both the ethics and the regulatory status change.

The subtle danger isn't AI openly making decisions; it's automation bias: the pharmacist who technically reviews the AI's recommendation but rubber-stamps it because it's usually right. When a tool is right 95% of the time, staying vigilant for the 5% is genuinely hard, and vendors' own materials acknowledge the clinician remains the legal author of every AI-drafted note and the owner of every AI-suggested intervention. That's a human-factors problem the profession needs to solve with workflow design and training, not just policy statements. My quality assurance training during an internship at Novartis shaped how I think about this: safety is not a declaration, it's a system, and the system has to be designed assuming the human will sometimes be tired, rushed, and trusting.

Drug Topics: What's one area of pharmacy practice you believe AI will genuinely transform in the next five years and one area you think it will never replace human judgment?

Patel: Transform: the medication-access bureaucracy—prior authorization, benefit verification, documentation, and population-level risk flagging. This is already happening. Companies are extending ambient AI into real-time prior authorization; medication-intelligence platforms are running across tens of millions of covered lives; specialty pharmacy startups are automating the entire intake-to-dispense workflow so independent clinics can operate in-house pharmacies. Within five years, I expect the hours pharmacists currently lose to paperwork to shrink dramatically, and in understaffed rural settings like mine, that's not a convenience; it's access. Fewer administrative hours means more time for the counseling, monitoring, and interventions that only a pharmacist can do.

Never replace: the counseling conversation with a scared patient. When someone has just been discharged after a heart failure admission with six new medications or is starting chemotherapy and terrified of what's coming, what they need isn't information—information is abundant now. They need a clinician who can read what's unsaid, gauge what they can absorb today, and earn enough trust to be honest with them. I built an AI tool for post-chemotherapy support precisely because I saw the gap between visits, and even as the builder, my conclusion was that the tool's job is to hold the space until the human gets there. The models will get better at sounding empathetic. Being accountable to a patient is not a sound.

Drug Topics: Is there anything else you would like to add?

Patel: Only that the stakes of this conversation are higher than they look. AI is entering U.S. medication-use systems faster than pharmacist-led evaluation research can validate it: the FDA has authorized over [1400] AI-enabled devices, ambient AI has reached hundreds of health systems in three years, and venture capital is pouring into medication-management AI. Meanwhile, the peer-reviewed evidence on AI's clinical pharmacy performance remains mixed at best. That gap, between deployment speed and validation speed, is where patient harm lives, and pharmacists are the profession best positioned to close it. We are the medication experts and the last checkpoint before a drug reaches a patient.

My own path runs from training in Kenya to Purdue to clinical practice in a rural Indiana community, and every stop taught me the same lesson: technology only improves care when the people closest to the patient shape it. I'd like to see far more pharmacists move from being end users of AI to being its evaluators, critics, and builders, and I'd like to see the profession claim that role loudly. It's encouraging that one of the most prominent medication-AI companies was founded by a pharmacist. That should be the norm, not the notable exception. The seat at the table isn't guaranteed. We have to take it.