By Pat Williams, CEO, iScribeHealth
LinkedIn: Pat Williams
LinkedIn: iScribeHealth
When healthcare leaders talk about ambient AI scribing, they almost always picture the same setting: a primary care physician, managing a panel of 15 to 20 patients, dictating notes between appointments. That’s the benchmark the industry has built its expectations around. But I spend most of my time working with a very different kind of physician, the orthopedic surgeon seeing 40, sometimes 50+ patients in a single day, moving room to room every twelve minutes, cycling through post-op checks, new injuries, and six-week follow-ups in rapid succession. There’s no time to pause and think through a note between rooms; the surgeon is already three doors down before the last patient’s chart closes. That physician operates in a different world entirely. And most ambient AI tools were not built for that world.
That gap is starting to catch up with the industry.
The Data Is Real. The Assumption Behind It Isn’t.
There is credible evidence that AI scribing delivers results. A landmark UCSF study found that AI scribe adoption was associated with a gain of 1.81 RVUs per week per physician, translating to roughly $3,044 annually at the 2025 Medicare Physician Fee Schedule. Those numbers matter. But that study drew from a health system dataset weighted toward internal medicine and general ambulatory visits. In high-volume orthopedic settings, where a surgeon may see two or three times as many patients in a single day, the productivity ceiling and the documentation complexity look very different, and a tool calibrated on a 20-patient day will be tested well past its design limits by hour six of a 45-patient one.
The technology is working. But “working for primary care” and “optimized for specialty practice” are not the same claim. And the market is beginning to feel the difference. A 2025 Menlo Ventures survey of more than 700 healthcare executives found that 67% of outpatient providers indicated they were likely to switch their current AI scribe vendor. Customers view scribing as becoming commoditized, and switching costs are low. While AI can transcribe, physicians are switching because their notes still require significant editing. A model that hasn’t internalized the HPI patterns, billing language, and return-visit cadence of orthopedics produces output that looks right on the surface, but it still needs a surgeon to fix it before it enters the chart. That editing time is exactly where the productivity gain disappears.
What Specialty Intelligence Actually Means
The difference between a generalist AI scribe and a specialty-trained one is the structure. An orthopedic encounter follows predictable patterns: mechanism of injury, prior conservative treatment, imaging correlation, functional status, and plan. Return visits have their own tight logic: weight-bearing progression, ROM benchmarks, hardware concerns, and clearance timelines. And when a model doesn’t know that logic, the surgeon steps in. I’ve watched orthopedic surgeons spend the final hour of their day correcting notes that captured the words but completely missed the shape of the encounter, rewriting sentences that were technically accurate but clinically useless to the next physician who reads that chart.
A model trained on millions of specialty-specific encounters doesn’t just transcribe what the physician said. It understands what a 10-week post-total knee replacement note is supposed to contain, flags when something is missing, and codes to the specificity that prevents downstream denials. That’s clinical intelligence built from the ground up for how subspecialists actually practice, not a general-purpose transcript with medical vocabulary layered on top. A decade of specialty-specific training data produces materially different output than 18 months of broad medical scraping. The market is beginning to feel that difference.
The EHR Mistake, Reloaded
I’ve seen this pattern before. During the EHR adoption era, physicians standardized on systems built for primary care workflows and then asked specialty physicians to adapt. The result was a generation of clinicians spending more time fighting their tools than using them. Burnout climbed, documentation quality suffered, and health systems spent years trying to retrofit solutions onto workflows they had never designed for. We are watching the early version of that same story play out again, just faster, and with a technology that promised to fix the very fatigue the EHR era created.
The next serious question the industry needs to ask is, “Does your AI scribe work for a 45-patient orthopedic day, at the pace those surgeons move and with the billing and coding precision their practice depends on?” The switching rates we’re already seeing tell you that generic answers to that question are not landing. Specialty physicians deserve tools built for the way they practice. The market will sort this out eventually. The question is whether health systems wait to feel the pain first or start asking better questions now.