The loudest voices in the AI revolution belong to technologists and executives. But the people who may be best suited to lead AI are the ones least likely to be handed the reins: in healthcare, that’s practicing clinicians.

At Presbyterian Healthcare Services, the largest integrated health system in New Mexico, that unlikely choice wasn’t an afterthought, it was the strategy. Instead of elevating a software expert or a physician‑executive to run its AI transformation, the system asked a passionate nurse practitioner with two decades of frontline experience to take charge.

Lori Walker, who serves as Chief Medical Information Officer, still spends part of her week seeing patients. She shares the same constraints her colleagues face daily: an overstuffed schedule, long drives between rural clinics, patients whose medical needs are tangled with social stressors, and electronic records that demand more attention than anyone can reasonably give. When she evaluates technology, she evaluates it from inside that pressure cooker, not from a conference room.

That perspective has shaped one of the most ambitious clinical AI deployments in the country. Presbyterian is rolling out a precision‑care AI platform across roughly 200 primary‑care clinicians. Unlike the pilot programs that often generate more press releases than progress, this rollout sits inside real workflows: identifying overlooked conditions, surfacing crucial patient history and collapsing hours of data review into minutes.

But the technology alone doesn’t explain why it’s working. The difference is that the effort is being led by someone who knows, from lived experience, how fragile the care environment can be.

The Human Layer of AI Adoption

In medicine, technology succeeds only when three things align: it solves a real problem, it fits into real workflows, and it is championed by people clinicians trust.

Walker understands the tiny failure points that don’t show up on dashboards: the lab buried in a scanned PDF, the medication change hidden in a note from an outside clinic, the payer rule nobody remembers until a claim is denied. She knows what it means to walk into an exam room and realize the patient has more problems than can possibly fit into the allotted time. And because her colleagues know she understands this, they trust her judgment about whether AI will help or hurt.

Clinicians Are Already Using AI

That trust matters more than ever, because a very different trend is unfolding quietly across American hospitals. A recent national survey found that roughly 40% of healthcare professionals have seen colleagues using AI tools their organizations never approved. Nearly 20% admit they’ve used unapproved tools themselves. And 10% acknowledge using them in direct patient care, shaping decisions that affect diagnosis, treatment or follow‑up.

Clinicians are not doing this to be reckless. They are doing it because they are drowning in documentation, in inbox messages, in prior‑authorization puzzles, in the sheer volume of information a single visit now requires. When official tools are too slow, too clunky or simply nonexistent, they reach for whatever helps.

But shadow AI introduces risk that is almost invisible: confident but wrong answers, inconsistent reasoning, unprotected patient data, no audit trail. The danger isn’t just the mistakes the tools might make, it’s the fact that nobody knows those mistakes are being made.

Presbyterian as a Model

Presbyterian’s approach offers a way out of that spiral, and it starts with who leads the work.

Practicing clinicians see real patient needs before anyone else. They are often the first to notice a workflow that’s buckling or a patient who might fall through the cracks. They know where clinical judgment depends on small details, and where AI must tread carefully.

Trust flows through people, not technology. In a system where nearly half the workforce is experimenting with unauthorized AI, the only way to bring clinicians back into a safe, sanctioned environment is through leadership they recognize and believe.

The official tool must actually be better than the unofficial one. At Presbyterian, AI isn’t a separate portal or a shiny demo. It’s inside the electronic record clinicians already use. It brings together clinical history, guidelines, social‑needs data and payer criteria in a single view. It saves time. It reduces friction. It answers the questions clinicians actually ask.

And it creates accountability where shadow AI cannot. Every recommendation is traceable, reviewable and explainable.

The Real Question Healthcare Must Answer

This is not “innovation theater.” It’s not hype. It’s the kind of quiet, operational transformation that rarely makes headlines but actually changes how healthcare works.

The stakes are not limited to one health system. Across the country, clinician burnout is deepening. Primary care access is shrinking. Administrative tasks are swallowing hours of every provider’s day. And the flood of new AI tools, good, bad and untested, is accelerating.

We can either let AI seep into the system through the back door, through desperate clinicians and consumer chatbots, or we can decide to build a front door worth walking through.

Presbyterian shows what that front door can look like: leadership grounded in the realities of patient care, not just the promise of technology.

The country has plenty of AI strategies. What it needs now are leaders who understand the work, the messy, nonlinear, deeply human work, that AI is meant to support. And the people who often understand that work best are the ones the system has too often overlooked.

If we want AI to make healthcare safer, more reliable and more humane, we should stop asking who builds the best model. We should start asking who leads it.

Deepthi Bathina is the Founder and CEO of GW RhythmX, an AI-native company defining the category of Enterprise Precision Care AI and building the large-scale foundation for the next generation of intelligent, connected Smart Hospitals. Formed through the merger of Get Well and RhythmX AI, the platform is deployed across more than 150 health systems, reaching over 85 million patients including 8 million U.S. military veterans and is powered by one of the industry’s deepest healthcare datasets spanning 300 million patient records and 4.4 billion annual claims.