The pace of artificial intelligence adoption in healthcare is enough to make even experienced technology leaders rethink how they do their jobs. New tools are arriving at a remarkable clip, clinicians are eager to put them to work, and health systems are under pressure to find efficiencies, expand access and improve patient care. At the same time, questions about safety, validation, cybersecurity and accountability remain very real.

That tension was at the heart of a recent Harvard Clinical Informatics Lecture Series webinar I attended, Leading Change as CIDO in the Age of AI, featuring Daniel Nigrin, M.D., chief information and digital officer at MaineHealth and associate professor at Tufts School of Medicine.

Nigrin brought a valuable perspective to the discussion. A pediatric endocrinologist by training, he has spent much of his career bringing together clinical care, informatics and health system leadership. His experience developing clinical tools long before today's AI boom shapes how he evaluates the opportunities and challenges healthcare organizations face today.

One point in particular resonated with me: Leading AI adoption is not simply about deciding which technologies to implement. It is about making sound decisions for an organization, helping people navigate change and ensuring that innovation ultimately serves patients.

As Nigrin put it, AI is changing a job he thought he knew how to do, requiring him to learn how to do things differently. For healthcare leaders across the industry, that may be one of the most important realities to acknowledge.

The pace of AI adoption is changing the leadership equation

Nigrin illustrated the speed of change by comparing the adoption of AI-powered ambient documentation with the rollout of electronic health records. EHR implementations took years to scale across hospitals and physician practices. Ambient documentation tools, by contrast, have gained traction over weeks and months.

The difference is more than a matter of speed. Healthcare organizations now face a growing stream of AI capabilities, each competing for attention, funding, implementation resources and the time of clinicians and staff.

The applications are broad. AI is supporting clinical documentation, revenue cycle operations, capacity management and clinical decision-making. Patients are also using publicly available AI tools to investigate symptoms, check information they receive from clinicians and explore possible treatments.

The challenge for leadership is deciding how to respond when the possibilities seem almost limitless, but an organization's ability to evaluate and implement them is not.

As Nigrin explained, health systems cannot deploy everything their clinicians and administrative teams request. They have finite budgets, limited implementation capacity and a workforce that can absorb only so much change at once.

There is also a more fundamental concern. Many AI products are being promoted before their claimed benefits have been adequately validated. Will a tool improve clinical outcomes? Will it make a workflow more efficient? Will it generate meaningful financial returns? Those are not interchangeable measures of success, and enthusiasm for a technology is not a substitute for evidence.

For leaders, this means resisting the temptation to equate rapid adoption with progress. The more consequential question is whether a particular application addresses a genuine need and delivers measurable value without introducing unacceptable risk.

A governance model that balances enterprise priorities with frontline needs

At MaineHealth, Nigrin described an AI governance approach that combines top-down priorities with bottom-up proposals. The executive leadership team has identified three organizational priorities for AI: financial benefit, capacity and access.

Financial benefit includes opportunities such as revenue cycle optimization. Capacity initiatives focus on making better use of constrained resources, including operating rooms and inpatient space. Access initiatives seek to help patients reach the right clinician in the right setting, whether through virtual care, urgent care, ambulatory services or other appropriate venues.

These priorities give the organization a framework for directing investment toward its most pressing needs. But Nigrin recognizes that not every worthwhile AI application will fit neatly into an enterprise-wide initiative. Individual departments may identify opportunities that are important to their own work. He offered the example of a legal team looking to use AI to review contracts more efficiently and apply established templates to new documents. Such a project may affect fewer people than an enterprise-wide clinical initiative, but it can still deliver meaningful value.

MaineHealth therefore also evaluates proposals from the bottom up, considering their merits, the number of people affected, potential return on investment and the organization's capacity to implement them.

That balance offers a practical lesson for other health systems. A governance structure should provide strategic direction without shutting out the people closest to the problems that need solving. It also creates a way to say not now, or not this particular tool, without dismissing the underlying need.

For healthcare leaders, that distinction matters. Good governance is not about slowing innovation for its own sake. It is about establishing a disciplined way to determine which opportunities deserve attention first.

Understand what the AI is optimizing for

One of the more thought-provoking parts of Nigrin's presentation concerned the values embedded in AI systems. He described a scenario in which a large language model is given a patient's history and laboratory results and asked to recommend a course of treatment from the perspective of a pediatric endocrinologist. The same patient information could produce a different rationale if the model were instead asked to evaluate the case from the perspective of an insurer deciding whether to approve that treatment.

The clinical facts have not changed. The context and objective have. That distinction raises an important leadership question: Do healthcare organizations understand the assumptions and priorities that shape the recommendations their AI tools produce?

As these systems become more deeply integrated into clinical workflows, it will not be enough to assess whether an application is fast, convenient or technically sophisticated. Leaders must also consider whether its recommendations align with the organization's clinical standards and patient-care priorities. This is particularly important when a tool's output may influence a clinician's judgment, a patient's next steps or a financial decision.

There are no easy answers to every question of AI governance, Nigrin acknowledged. Continued discussion and scrutiny are essential. That makes governance an ongoing responsibility, not a one-time approval process.

Moving faster on cybersecurity while managing new risks

AI also complicates the cybersecurity equation.

Healthcare organizations have spent years strengthening their defenses against ransomware and other attacks. Now, as AI capabilities advance, security teams must contend with the potential for vulnerabilities to be identified and exploited at a pace that outstrips traditional response processes.

Nigrin described the urgency of applying security patches and fixes to healthcare systems, including Epic, as vulnerabilities emerge. For MaineHealth, keeping pace requires a commitment to responding quickly and dedicating resources to the work.

He also raised an issue that deserves attention as organizations evaluate AI vendors: Every additional system, interface and integration can introduce complexity.

Embedded AI tools and third-party applications both carry risk. Integrating tools into existing workflows can make them more useful for clinicians, but adding more moving parts can create additional opportunities for misconfiguration and vulnerabilities.

Vendor stability is another consideration. The AI marketplace is changing so quickly that companies and products that appear promising today may not be around in a few years. Meanwhile, established technology vendors are expanding their own AI offerings, forcing health systems to reconsider existing investments and future platform decisions.

Nigrin also emphasized the importance of contractual protections, including vendor indemnification and appropriate insurance coverage, as health systems work through questions of medical liability.

These considerations may not be as exciting as a new AI capability, but they are essential elements of responsible adoption. Leaders need to consider not only what a technology can do, but who will support it, how it will be secured and who bears responsibility if something goes wrong.

Preparing clinicians for an AI-enabled future

Perhaps the most human dimension of Nigrin's presentation was his discussion of deskilling and clinical education. AI can help clinicians work more efficiently, but what happens if reliance on these tools weakens the knowledge and judgment they need to practice independently?

At MaineHealth, the organization has chosen to restrict ambient documentation tools for clinicians in the early stages of training. The goal is to ensure that medical students, interns and residents first develop a grounding in writing clinical notes themselves before relying on AI assistance. It is a measured approach to a complicated issue. The objective is not to deny emerging clinicians access to useful technology indefinitely. It is to preserve the foundational skills they need to understand, evaluate and ultimately take responsibility for their work.

Nigrin suggested that the same question could arise across cognitive specialties as AI increasingly assists with clinical interpretation and decision-making. For health system leaders, this has implications well beyond training programs. AI adoption requires workforce planning, education and thoughtful decisions about when human expertise must be developed before technology takes on more of the work. It also requires acknowledging that people adapt at different speeds. Some clinicians may be eager to adopt new tools, while others need more time, support and opportunities to build confidence.

During the Q&A, Nigrin offered an observation that applies to nearly every major technology initiative: "The capacity for change is often the rate-limiting step that we need to adhere to."

That is worth remembering in an environment where the technology can advance much faster than the people expected to use it.

Knowing when to buy and when to build

Nigrin also offered a useful perspective on build-versus-buy decisions. For large-scale applications, such as ambient clinical documentation, he sees little reason for most health systems to develop their own solutions when established vendors already offer them. For smaller, more specialized needs, however, AI-assisted development may create new opportunities.

He described a project at MaineHealth involving a hoteling application that allows employees to reserve workspaces when they come into the office. The organization had been paying approximately $60,000 annually for an existing tool.

Nigrin challenged the internal development team to see whether it could build an alternative using AI-assisted coding tools. The team produced a proof of concept within 48 hours and had a functional platform ready for rollout in about a week.

The example illustrates how AI may allow health systems to address targeted operational needs without committing to expensive, off-the-shelf solutions. It also reinforces the importance of context in technology decisions. Building a specialized tool may make sense when the need is well defined and the solution is manageable. Other applications require the scale, expertise, testing and ongoing support that established vendors can provide.

The lesson is not that health systems should build more or buy less. It is that leaders have new options to consider, and those decisions should be driven by organizational needs, risk and the resources required to maintain a solution over time.

Keeping patients at the center of change

As the discussion drew to a close, Nigrin returned to a principle that has guided his leadership throughout his career. "We're clinicians at heart," he said, explaining that even in an executive role, he still considers himself a clinician first and foremost.

His advice was straightforward: Use what is best for the patient to guide decisions, whether the discussion takes place at the operational level or in the executive boardroom. That principle sounds simple, but it provides a useful compass when leaders face competing demands from clinicians, executives, legal teams, risk managers and technology vendors. It can help determine whether an AI application is ready for deployment, whether a proposed efficiency gain is worth the potential trade-offs, and whether a new capability should be introduced now or after additional safeguards are in place.

MaineHealth is already using internally focused AI agents for administrative tasks and preparing to introduce patient-facing voice agents for selected applications, including billing inquiries, appointment requests, prescription refills and care reminders. Nigrin emphasized that the organization intends to proceed cautiously as it expands into these patient-facing uses. That approach reflects the larger challenge facing healthcare leaders. AI's potential is significant, but the responsibility to protect patients does not diminish as the technology becomes more capable.

My biggest takeaway from the Harvard Clinical Informatics Lecture Series discussion is that successful AI leadership requires more than technical expertise or a willingness to experiment. It requires clear priorities, thoughtful governance, an understanding of workforce readiness, attention to security and liability, and a willingness to question whether a promising tool is delivering the value it claims.

The technology will continue to change. The leadership responsibility remains the same: Make decisions that serve patients, support the people who care for them and strengthen the health system's ability to deliver safe, effective care.

For healthcare organizations navigating the AI transition, that is a sensible place to start.

Melinda Taschetta-Millane is Market Content Director of Healthcare Editorial, and Head of Content for Healthcare Innovation.