The Scribe That Launched a Thousand Takes

With AI Charting, Epic absorbs another threat while actual disruption hides in plain sight

Sing, LinkedIn, the takes,
the furious, proliferating takes,
that cost the market countless hours,
hurling into threads and timelines so many confident claims,
analysts’ claims, founders’ claims, executives’ claims,
made carrion for comment sections and quote-tweets,
while the will of platforms moved steadily toward its end.

Begin when the players first broke and clashed:
Epic, lord of workflow, and the brilliant upstarts,
Abridge, Ambience, Nabla,
with Microsoft watching from Olympus,
cloud and model in hand,
and founders and CMOs drawn into the quarrel,
each certain the war was theirs.

Epic’s long-anticipated ambient scribe has finally arrived, and with it, the ships have been launched yet again: a flood of commentary about power, defaults, and who gets to define the future of clinical work. From Brittany Trang at STAT:

The nation’s largest electronic health record company, Epic Systems, announced on
Wednesday more details around its long-anticipated AI Charting feature, which could disrupt the market for ambient scribes. Several health systems are piloting the tools now, the company said.

But don’t call it a scribe!

“I almost bristle a little when people are like, ‘Oh, it’s an AI scribe,’ and I’m like, ‘No, no, it’s a scribe, but that’s passive. We really want this to be active,’” said Jackie Gerhart, chief medical officer at Epic, in a call with STAT ahead of the announcement. “We’re just going to keep adding more besides just notes. So it’s going to be orders, it’s going to be diagnoses.”

Well, yeah! The end state for Art (and all scribes worth their strategic salt) is a clinical copilot that minimizes or eliminates every keystroke and mouse click a provider makes today. As mentioned in “One Copilot to Rule Them All”:

The various clinical jobs to be done (ambient, visit prep, HCC coding, clinical trial identification, etc) are all pieces of a single clinical copilot product…our siloed product categories are early beachheads on the same battlefield, B2B always moves to the bundle, and further convergence is ahead of us.

The arena is heating up, and there’s room for more convergence ahead, where more use cases sit unbundled:

In a world where the marginal cost of writing software is approaching zero, differentiation no longer comes from building a feature. It comes from assembling all the functions a business depends on. That dynamic pushes vendors to expand laterally. It’s now a race to build out all the functions a business might need, as Ben Thompson articulated this past week:

That means the growth story for all of these companies is in serious question — the industry-wide re-rating seems completely justified to me — which means the most optimal application of that new AI coding capability will be to start attacking adjacencies, justifying both your existence and also presenting the opportunity to raise prices. In other words, for the last decade the SaaS story has been about growing the pie: the next decade is going to be about fighting for it, and the model makers will be the arms dealers.

That context clarifies precisely how exactly Dragon (and Microsoft) fits into the picture, which had been in question ever since Epic’s Ambient Bombshell in August. Epic is simply using Azure to access different LLMs behind the scenes, but not Dragon/DAX:

Epic generally usesdifferent models for different specialties and tasks, and accesses those through Microsoft’sAzure platform…When asked whether Epic’s new AI Charting feature shares technology with Microsoft’s DAX, Miller declined to affirm or deny. “This is a separate product than DAX. We are working with Microsoft, partnering with them closely on how we’re creating this,” said Miller.

Our discussion in Epic’s Ambient Bombshell and elaborated in “A First Look at Art” is thus validated:

We see now that it’s not one or the other, but both. When word leaked that they were releasing a scribe, Microsoft, as a Cornerstone partner, put pressure on them, leading to the vague announcement.

Epic customers do need defaults, and that’s what’s provided here - barring a customer opting for any of the other 36 ambient products in Epic’s Showroom, Microsoft’s models are what power Art. They are almost certainly using the varied Blueprints available to the Ambient Toolbox, so that an Epic customer can switch to the vendor of their choice to undergird the experience if they want. Rather than kingmaking, it’s further commoditizing their complements.

Make no mistake - Microsoft should be delighted with this arrangement. Defaults matter, meaning they’ll get a lot of free wins and traffic towards their Azure LLM offerings baked in. Competitive ambient products will have tremendous pressure and need to differentiate on economics, clinical performance, and next-level workflow. But what matters (and what courts are increasingly signaling under information blocking jurisprudence) is that the default cannot be the only viable path. Equal access to the same integration interfaces is what keeps this from being exclusive dealing, and that constraint is exactly what forces innovation, not stagnation.

There is plenty more to unpack in Brittany Trang’s excellent reporting, and much of the commentary will understandably fixate on Abridge, the supposed demise of third-party scribes, and hints of anticompetitive behavior. I’m less persuaded by the more alarmist takes. Third-party scribes are not going away. Given the regulatory and litigation climate, Epic would be reckless to self-preference (or price) in a way that meaningfully forecloses competition. The continued collaboration between Epic and Abridge on DANbient notes (arguably one of Art’s most compelling features) is an explicit signal in the opposite direction.

There are many viable paths to differentiation, and most of them are operational rather than cosmetic. Josh Liu nails it as always:

Don’t count the Ambient AI vendors out just yet! To do it well, you need to support multiple specialties, multiple languages, provide strong change management, etc. It’s unclear how much of this Epic has done and is willing to do - and until we see real success delivered, the competition is far from over.

I couldn’t help myself. Eric Bana and Shiv are basically the same person.

Stepping back, the implication of Epic’s entry (as well as every other EHR’s) is straightforward: AI is being absorbed into the EHR in exactly the way incumbent systems of record have historically survived threatening point solutions - by internalizing the new capability, setting defaults, and preserving control of the system of record. This is classic sustaining innovation - it improves performance along dimensions mainstream customers already value without challenging the EHR’s central role. The ultimate beneficiary is the provider: lower prices, broader choice, faster iteration, and sustained workflow innovation.So where is the disruptive change we were promised?

AI and Disruption

True shakeups in tech markets come almost invariably as a result of platform shifts. AI feels like a platform shift, so it’s not surprising that you can hardly shake a tree on LinkedIn lest a hundred “AI will disrupt EHRs” takes falling out. Can we craft one sui generis here? We certainly must try.

Clayton Christensen’s “The Innovator’s Dilemma” provides some clear guidance, which we can utilize:

Most new technologies foster improved product performance. I call these sustaining technologies. Some sustaining technologies can be discontinuous or radical in character, while others are of an incremental nature. What all sustaining technologies have in common is that they improve the performance of established products, along the dimensions of performance that mainstream customers in major markets have historically valued. Most technological advances in a given industry are sustaining in character. An important finding revealed in this book is that rarely have even the most radically difficult sustaining technologies precipitated the failure of leading firms.

Occasionally, however, disruptive technologies emerge: innovations that result in worse product performance, at least in the near-term. Ironically, in each of the instances studied in this book, it was disruptive technology that precipitated the leading firms’ failure. Disruptive technologies bring to a market a very different value proposition than had been available previously. Generally, disruptive technologies underperform established products in mainstream markets. But they have other features that a few fringe (and generally new) customers value. Products based on disruptive technologies are typically cheaper, simpler, smaller, and, frequently, more convenient to use.

We can apply the four core questions drawn from Christensen’s framework to AI copilots to assess their effects on Epic and traditional EHRs.

1. Does AI improve the performance of established products?

AI provider tools, especially with the current focus on copilots, are all geared toward enhancing clinician productivity within the existing EHR workflow. Artificial intelligence is being integrated into existing EHR platforms to make them faster, more efficient, and less burdensome to use. Tools like ambient documentation (e.g., Nuance DAX or Abridge), predictive alerts, and intelligent task automation are all being layered onto systems like Epic and Cerner to enhance their utility. These improvements serve to make traditional EHRs better at the jobs they already do, such as capturing billing codes, managing clinical documentation, or surfacing clinical decision support. So by this metric, AI is functioning as a sustaining innovation, enhancing the core capabilities of EHRs rather than challenging their relevance.

2. Does AI reinforce the performance dimensions that mainstream customers have historically valued?

Yes. The mainstream customers in this context (hospital systems, health systems, and large provider groups) have long prioritized qualities like auditability, billing precision, compliance, integration with enterprise IT, and risk adjustment. Current AI deployments often serve to reinforce these values (to the point that we see AI scribes shifting their value propositions to focus on ROI and revenue cycle more strongly)

For example, health systems are investing in AI to reduce physician burnout without compromising revenue cycle integrity or documentation compliance. By improving productivity while preserving (or enhancing) the priorities of large-scale enterprise buyers, AI is being shaped to align with the values of the incumbent market. This again supports the conclusion that AI is, at least in its present mainstream applications, a sustaining innovation.

3. Does AI result in worse product performance (in mainstream metrics), at least initially?

Generally, no. AI copilots are designed to integrate seamlessly with existing EHR systems and maintain compliance standards. Unlike standalone AI tools, copilots work within established workflows and documentation requirements. They may occasionally produce errors or require oversight, but they’re built to meet the same auditability and billing precision standards that mainstream customers demand. This reinforces their sustaining rather than disruptive nature.

4. Does AI offer a very different value proposition than has been available previously?

Not fundamentally. While AI copilots introduce new interaction modalities like voice commands or intelligent suggestions, their core value proposition remains making existing EHR functions more efficient. They don’t challenge the centrality of the EHR or propose alternative care delivery models. Instead, they make the current system work better for current users doing current tasks.

Disruption One Layer Removed

So what’s the takeaway? For me, it’s that copilots and AI provider tools are sustaining innovations that accrue more value to the EHR (if they are deft enough to seize it), much like the advent of many prior generations of workflow extensions and novel softwares, such as business intelligence platforms, patient portals, population health systems, CRM software, and “digital front door” tools. These additions didn’t displace the EHR; they expanded its orbit. AI copilots do the same: they improve usability and reduce friction for clinicians, but ultimately reinforce the EHR’s position as the central system of record and workflow hub.

Where the threat from AI really lies is one layer removed: not the EHR, but the provider organizations themselves. When you apply “The Innovator’s Dilemma” framework to traditional hospitals and clinics, the analysis yields very different results. These institutions are built around complex, capital-intensive service delivery models that optimize for inpatient care, compliance, reimbursement, and high-acuity interventions. In other words, they serve mainstream markets with established expectations (a.k.a exactly the kind of setting where sustaining innovations thrive).

AI opens the door to disruptive care models that are cheaper, more scalable, and often good enough for lower-acuity needs. Virtual-first clinics, asynchronous care platforms, AI-driven triage and diagnostic tools, and even autonomous agents that can handle care coordination or routine follow-ups all serve different jobs, for different customers, at a radically different cost structure. They often underperform on traditional measures, such as billing complexity or continuity of care, but they offer speed, convenience, and affordability to populations that are historically underserved or outside the reach of traditional systems.

Another framing for this that I like to understand this is the via Ben Thompson’s “Idea Propagation Value Chain”. In that article, he describes how technology has changed the bottleneck of idea propagation:

The evolution of human communication has been about removing whatever bottleneck is in this value chain. Before humans could write, information could only be conveyed orally; that meant that the creation, vocalization, delivery, and consumption of an idea were all one-and-the-same. Writing, though, unbundled consumption, increasing the number of people who could consume an idea.

Now the new bottleneck was duplication: to reach more people whatever was written had to be painstakingly duplicated by hand, which dramatically limited what ideas were recorded and preserved. The printing press removed this bottleneck, dramatically increasing the number of ideas that could be economically distributed:

The new bottleneck was distribution, which is to say this was the new place to make money; thus the aforementioned profitability of newspapers. That bottleneck, though, was removed by the Internet, which made distribution free and available to anyone.

What remains is one final bundle: the creation and substantiation of an idea. To use myself as an example, I have plenty of ideas, and thanks to the Internet, the ability to distribute them around the globe; however, I still need to write them down, just as an artist needs to create an image, or a musician needs to write a song. What is becoming increasingly clear, though, is that this too is a bottleneck that is on the verge of being removed.

We can crudely map the pre-digital healthcare process in a a similar fashion.

In the pre-digital healthcare era, medical knowledge creation was slow and localized. New ideas had to be substantiated through peer-reviewed journals and manual protocol development. These protocols were then duplicated by reimplementing at each institution and slowly spreading through medical conferences and organizational networks. Distribution was limited to physical hospitals and clinics with geographic constraints. Finally, consumption required in-person care delivery with direct doctor-patient interaction.

The telehealth era maintained the same creation and substantiation processes - medical knowledge still originated from research and was encoded into peer-reviewed protocols through human expertise. Duplication remained largely the same, with providers using established organizational protocols via clinical decision support systems. However, distribution was revolutionized by telehealth platforms that broke geographic boundaries, allowing remote care delivery. Consumption evolved to video calls and remote consultations, though still requiring synchronous interaction with human providers.

The AI era represents the most dramatic shift. Creation and substantiation processes now benefit from AI assistance, with algorithms helping generate and refine medical protocols. Most significantly, duplication is transformed by AI agents that can follow codified protocols with infinite capacity and 24/7 reliability, removing human bottlenecks. Distribution continues through telehealth infrastructure, but consumption becomes asynchronous and AI-mediated, allowing for on-demand, personalized care delivery without requiring real-time human provider availability.

Indeed, we’ve seen a slew of these released on the past few months:

• Counsel Health’s raise

• Verily Me’s launch

• General Medicine’s AI chat release

• Lotus Health’s model

These are all cut from the same cloth, even if the tailoring is different. They will not be the last - the model is inherently copyable with limited moat, so one could expect a mad stampede of near-identical entrants, similar to the DTC goldrush. For now, so long as licensure requirements still exist, there’s the promise (and perhaps illusion) of some real providers somewhere, pulling the levers behind the scene.

But we may have it backwards. The question isn’t whether there’s a provider behind the curtain supervising a model. It’s whether the model is now the thing pulling the levers, with the provider reduced to a legitimizing prop. Indeed, “pay no attention to the man behind the curtain” would be oddly apt for this phase of care delivery in the era of Oz.

The Cultural Disruption Challenge

So hold up - with that in mind, do agents represent the end of the EHR vis a via the replacement of their customer base by a nimbler, more cost-effective set of care organizations? The major problem I see isn’t that these AI-enabled care models underperform on traditional clinical or operational metrics. It’s that they underperform on societal expectations and cultural norms.

Two main factors have kept the telehealth era from having the same disruptive impact as the internet had on traditional media. These will similarly impede the AI era of the medical care propagation flow:

1. Physical nature of medicine: Unlike information or media content that can be fully digitized and transmitted, healthcare often requires physical examination, hands-on procedures, lab work, imaging, and direct patient contact. Many medical conditions cannot be properly diagnosed or treated through a screen. Critical care, surgery, emergency medicine, and complex diagnostics still fundamentally require physical presence and infrastructure. This creates an inherent limitation that doesn’t exist in purely informational industries like journalism or entertainment.

2. Social mores and regulatory regimes: healthcare is governed by social norms and professional gatekeeping. Patients, providers, and regulators share deeply embedded beliefs about legitimate care: credentialed clinicians, synchronous encounters, human empathy, continuity, and clear lines of accountability. These norms are enforced not just culturally, but through licensure, malpractice regimes, reimbursement policy, and medical board oversight. Telehealth only advanced as far as it did because a global emergency forced temporary exceptions and many of those are now continually stuck in limbo or being actively unwound .

What we’re seeing now is not skepticism about AI per se, but systemic pushback against care models that threaten incumbent assumptions and power structures. That pushback shows up as:

• Regulatory agencies waffling on emergency telehealth flexibilities that were implemented during COVID-19

• Insurance companies pushing back on reimbursement rates and coverage scope for virtual visits

• Medical boards maintaining strict licensing requirements that limit cross-state practice

• Cultural expectations from both patients and providers who still view virtual care as “lesser than” traditional face-to-face medicine

• Organized resistance from incumbent providers facing price pressure from lower-cost alternatives

This cultural dimension exposes a gap in Christensen’s framework. In most technology markets, disruption occurs when a cheaper or simpler product becomes “good enough” for mainstream users. In healthcare, “good enough” is not determined solely by clinical or financial performance - and often not by the consumer at all. Payment structures, licensure, liability, and professional norms intervene.

Consider how other industries overcame similar cultural resistance. Uber required people to get into cars with strangers (violating fundamental safety norms parents had drilled into us since childhood), yet convenience and cost eventually won out. Online banking initially felt unsafe and impersonal, yet efficiency overcame concerns about human tellers.

Healthcare’s barriers suggest a slower disruption that may require generational change, not just technological improvement. The populations most likely to accept AI-first care, like digital natives comfortable with app-based services, underserved communities with limited healthcare access, and cost-conscious consumers frustrated with traditional care, may need to grow large enough to force cultural change from the bottom up. This suggests a longer disruption timeline than typical technology adoption curves would predict.

While I do think this disruption is coming (particularly in primary care, behavioral health, and other low-acuity, protocolized care settings), the cultural inertias (and physical form factors) above will protect larger incumbent healthcare organizations longer than economic factors alone would suggest. While AI makes new care models technically feasible and economically attractive, widespread adoption may wait until we see the erosion of the many cultural barriers to non-human care delivery. The time horizon is almost certainly longer than we hope or imagine. As with Helen, the scribe may be the cause of the war, but the siege thereafter? That will unfold slowly, constrained by custom, institutions, and time.

Point Solutions’ Pyrrhic Victory

A parting thought on this section is to be careful what you wish for. The coup de grâce of AI, presuming the trajectory charted above is accurate, is that the same forces poised to topple Epic and incumbent EHRs will simultaneously render irrelevant the horde of third-party applications that clamor for precisely such disruption. This is the disruption paradox in a beautiful nutshell: everyone’s fighting the last war while the real transformation happens at a different layer entirely. The third-party vendors cheering for Epic’s downfall will discover they were just rearranging deck chairs on the Titanic. If AI-powered care delivery models emerge outside the traditional hospital system, they won’t just bypass the EHR. They’ll bypass the entire ecosystem of bolt-on tools, too.