The Future of AI in Healthcare Will Be Won in the Workflow

Last week, Ben Scharfe’s interview offered a grounded view of AI’s role in healthcare today. Instead of treating AI as a monolithic solution, he frames it as a set of targeted, specialty-aware tools designed to enhance, not replace, clinician performance. The most compelling examples are those embedded in existing workflows: ambient listening for documentation, pre-visit patient history summarization, and automation of low-risk administrative tasks.

Workflow integration is the differentiator

Healthcare leaders have seen too many promising technologies fail because they were bolted on rather than built in. AI is no exception. Tools that require clinicians to leave their EHR environment, manually import data, or navigate additional interfaces will struggle to gain traction.

Scholarly research supports this. A JAMA Network Open study found that digital tools integrated directly into clinical workflows saw adoption rates nearly three times higher than standalone applications. For AI, integration is not just a convenience feature, but it is also a precondition for utility.

Multi-specialty practices offer a strong proving ground for domain-specific AI models. Scharfe notes that tailoring AI training to the depth of clinical knowledge needed in each specialty can reduce the irrelevant or inaccurate outputs common with general-use models. This specialization mitigates risk while delivering more relevant recommendations at the point of care.

Trust is the critical adoption metric

If integration is the technical requirement, trust is the cultural one. Clinician resistance can derail even the most sophisticated deployment. The concern is not unfounded: AI models have been documented producing errors that could harm patients if left unchecked.

Building trust involves transparency, oversight, and incremental adoption. AI systems should make their reasoning visible, provide confidence scores where possible, and remain under clinician control. As Scharfe emphasizes, AI must assist, not replace, the human decision-maker.

Patients must also be considered. Clear communication about when and how AI is used in their care can improve comfort levels and prevent misconceptions. According to KFF, patients are more accepting of AI when it is framed as a tool used in collaboration with, rather than instead of, their clinician.

Regulation could accelerate, or slow, innovation

Scharfe’s warning about fragmented regulation is well founded. Maintaining compliance across differing state and federal rules is expensive and time-consuming, particularly for developers serving multi-state health systems.

A unified, federally led framework could help standardize requirements for safety, transparency, and accountability. This is not without precedent. The ONC’s interoperability and information-blocking rules created a nationwide baseline that accelerated certain forms of data exchange. Similar clarity for AI could reduce compliance burden and enable faster scaling of effective solutions.

Measuring impact beyond efficiency

While efficiency gains are an attractive ROI measure, healthcare leaders should resist the urge to define AI’s value solely by minutes saved or costs reduced. AI’s more profound impact may be in improving clinician-patient interaction, enhancing diagnostic accuracy, and supporting care plan adherence.

Ambient listening, for example, not only reduces documentation time but can improve patient satisfaction by allowing physicians to focus fully on the conversation. A Health Affairs study found that patient perceptions of attentiveness and empathy improved when clinicians were less distracted by screens during visits.

Similarly, AI-driven pre-visit summaries can ensure that critical data is not overlooked, potentially leading to earlier interventions and better outcomes. These benefits are harder to quantify but no less important for strategic decision-making.

The road ahead

Over the next five years, AI’s trajectory in healthcare will likely be determined by its ability to deliver measurable, trusted value within existing workflows. Leaders who pilot AI tools with strong change management plans, engage end-users early, and align technology investments with high-impact use cases will be best positioned to succeed.

The challenge is not in imagining what AI can do, rather it is in executing deployments that are safe, relevant, and embraced by those they are meant to serve. That means balancing innovation with caution, efficiency with empathy, and automation with human judgment.

If Scharfe is right, the promise of AI in healthcare is not about replacing people. It is about enabling them to work at the top of their license, with more time for patients and fewer administrative burdens. In that future, AI is not the star of the show, but an indispensable member of the cast.