Building the future of care delivery
August 24, 2026
Healthcare is undergoing a profound AI transformation. Just in the last decade, we’ve gone from “does it even work?” to “how can we scale it faster?”
Ambient documentation tools have already proven they can lift the administrative burden from clinicians and put hours back into a doctor's day. The harder, more consequential question facing health system executives, payers and technology leaders today is different: what does it actually take to turn a promising pilot into a durable, enterprise-wide advantage?
A recent conversation among four executives sitting at different points of the healthcare AI stack — Abhi Pathak, Chief Product Officer at Suki; Dr. Suchi Saria, Founder and CEO of Bayesian Health and Director of the AI and Healthcare Lab at Johns Hopkins; Bobby Sherwood, VP of Product Management at HealthEdge; and Dr. Ruben Amarasingham, Chief Medical Officer at Smarter Technologies — surfaced a striking consensus. The technology is rarely the bottleneck anymore. Strategy, governance, and adoption are.
From standalone strategy to core infrastructure
Two years ago, generating a clinical note automatically from a patient encounter was a novelty. Today it's table stakes, and the industry's attention has shifted to what comes after the note: reasoning over a patient's data, closing documentation gaps that affect billing, and orchestrating the administrative machinery, like prior authorizations and referrals, that consumes so much of healthcare's overhead.
AI is on a trajectory to disappear into the infrastructure of care delivery entirely.
Abhi Pathak
Suki Chief Product Officer
Pathak argues the very framing of "AI strategy" as a distinct discipline is temporary. "My prediction is that in two years we won't have the sentence people saying, 'Do we have any strategy?'" he said. "It sounds odd — as somebody who has seen the internet and then mobile — it's like saying, 'Do you have an internet strategy or do you have a digital strategy?'" In other words, AI is on a trajectory to disappear into the infrastructure of care delivery entirely, the same way the internet did for retail or mobile did for banking.
But getting there requires organizations to be honest about where they add unique value and where they don't. "The common theme I see in terms of companies who are getting this right... are very clear on what the differentiation is, what they own, what they don't own, where they're uniquely situated, and they're very disciplined about where they invest," Pathak noted.
Adoption is a design decision, not an afterthought
If there was one theme that cut across every panelist's remarks, it was this: the technology rarely fails on technical merit. It fails because nobody planned for the humans who were supposed to use it.
Dr. Saria's research at Johns Hopkins put hard numbers behind a problem the industry has quietly tolerated for years. "Best-in-class informatics teams... show something like 10 to 12% adoption," she said of typical clinical decision-support and alerting tools deployed inside major health systems. That statistic, she noted, is treated almost as a law of nature inside hospital IT departments — until it isn't. Her team published research showing 90% sustained adoption across 4,400 clinicians over two and a half years, a result she said was initially met with outright skepticism. Since then, she said, "we've been able to achieve repeatedly upwards of 85% adoption at every single system we've deployed... while the alternative still sees 10 to 12%."
The gap, in her telling, isn't a data science problem; it's a trust and workflow-design problem, built on continuously monitoring model drift, co-designing workflows with clinicians, and making the AI-assisted path measurably easier than the status quo.
The failure mode Pathak sees repeatedly is an organization that gets excited about a capability and builds it, "but then nobody thought about who's going to use it. How will you train your people? How do you manage the change management?" His prescription: "The biggest mistake is treating AI as a feature instead of a business decision." Teams that start with the clinician's actual problem, and instrument success against it from day one, are "built differently. They drive adoption faster because it was built in and not just bolted on."
The biggest mistake is treating AI as a feature instead of a business decision.
Abhi Pathak
Suki Chief Product Officer
Sherwood, whose team at HealthEdge evaluates and deploys AI capabilities on behalf of health plans, echoed a similar sentiment: "If users fall in love with the capability during a pilot, that health plan is going to have a hard time taking it away." Everything downstream, from contracting to pricing and scaling, becomes a formality once that emotional buy-in exists.
Financial ROI Is Now the Price of Entry, Not a Bonus Slide
Clinical benefit alone no longer justifies budget. With hospital operating margins hovering near 1%, every panelist agreed that health systems and payers are demanding a clear, defensible line from AI investment to dollars, and demanding it faster than in prior technology cycles.
Amarasingham described the shift plainly: organizations have moved past open-ended experimentation into something more disciplined. "Everything looks good in a demo," he cautioned. The real differentiator is whether an AI system's output is "accurate and safe and clinically authentic" once it leaves the sandbox and enters real clinical documentation and billing workflows, where nuance and fidelity to what actually happened in the encounter matter far more than a polished output. On the question of ongoing integration and maintenance costs eating into ROI over time, he was equally direct: health systems "need to see recurrent ROI," and technology vendors have an obligation to keep improving price and capability continuously rather than treating a signed contract as the finish line.
Sherwood echoed the same financial discipline from the payer side, where AI adoption still lags provider organizations, by his account. Every deployment, he said, starts with a testable hypothesis of value, typically anchored first in staff efficiency and time savings, the easiest outcome to measure, before organizations can credibly chase the harder, longer-tail prize of reduced cost of care and better revenue cycle management.
Build, partner, or do both
Perhaps the most practical guidance for executives came on the build-versus-buy question, an area where the panelists converged on a more nuanced answer than the binary the industry often assumes.
Sherwood reframed the decision entirely: "You can't expect the market to wait for your roadmap when there's that much to do." His team drew a hard line around the data and workflows unique to HealthEdge's platform, and committed to internal investment in those areas. From there, they treated "literally everything outside of that" as a partnership opportunity, selecting partners based as much on strategic and roadmap alignment as on immediate feature fit.
Amarasingham described an even more granular version of the same discipline at Smarter Technologies, where the company deliberately built certain components of its ambient documentation stack in-house while relying on partners for others, preserving what he called "product coherence" across the front, middle and back end of the revenue cycle rather than forcing an all-or-nothing vendor decision.
The next three years: an invisible intelligence layer
Asked to look ahead, the panel converged on a shared vision that has less to do with any single product and more to do with AI receding into the background of clinical work entirely. Pathak described a future where AI becomes "the intelligence layer that's invisible — not a tool that clinicians open or use," continuously staging orders, flagging documentation and care gaps, and closing financial and operational loops without requiring a clinician to seek it out.
Saria's framing was similar in spirit: high-quality transcription, high-quality clinical reasoning, and payer collaboration on reimbursement need to be stitched into "a very seamless experience" where clinicians can simply assess, agree or disagree, and act — with everything technically automatable actually automated.
For health system and payer leaders still treating AI adoption as a series of disconnected pilots, the message from this panel is unambiguous: the winners of the next three years will not be defined by which model they chose. They will be defined by whether they treated governance, adoption and financial accountability as part of the strategy from day one — not as problems to solve after the demo gets applause.
