Abridge: the eaves and the house, built by a cardiac surgeon who was still writing notes.

Abridge used its time under Epic’s eaves to build a four-layer house underneath. On February 5, 2026, the eaves went free. What happens next depends on whether the house still stands on its own.

1. A $5.3B price tag, and the day the eaves opened up

On June 24, 2025, Abridge, a medical AI company based in Pittsburgh and San Francisco, raised a Series E of $300M at a $5.3B valuation led by Andreessen Horowitz and Khosla Ventures. Just four months earlier, its Series D had come in at $2.75B (Fortune coverage). Valuation doubled in four months.

Consider the numbers concretely. Abridge is contracted with 150+ US health systems (large networks that bundle multiple hospitals and clinics), processed over 50 million medical conversations in 2025, covers all 12,000 clinicians at UPMC, and is deployed across 40 hospitals, 600 clinics, and 24,600 physicians at Kaiser Permanente. Market share sits at 30% of the ambient AI scribe category per 2025-2026 KLAS-linked data. Second-place Ambience Healthcare holds 13%. Abridge leads by more than 2x.

Put plainly, the largest academic medical centers in the US are standardizing on a single AI vendor for the exam room.

Then on February 5, 2026, the ground shifted. Epic Systems, the largest US EHR vendor covering 42.3% of acute-care hospitals, released an AI Charting feature natively inside its product that does the same thing Abridge does.

It was as if Epic had opened up the very eaves that Abridge had been building its house under. This article asks one question. Now that the eaves are free, does the house Abridge built underneath them still stand? The short answer up front. Abridge’s bet was a timing play: use the months under the eaves to stack four layers underneath. February 5, 2026 is the day that timing bet started to be graded.

2. A cardiac surgeon who was still writing notes built it out of his own frustration

There is one person to introduce first. Founder and CEO Shiv Rao, MD (formally Shivdev Rao).

Rao is a practicing cardiologist at UPMC (University of Pittsburgh Medical Center). Practicing in the literal sense. He kept seeing patients after founding the company, and takes a weekend on-call shift roughly once a month. He told Fortune:

“I’m an end user myself. Just last night, I was on call at the hospital.”— Shiv Rao (Fortune, Feb 2025)

That fact shapes every product decision at Abridge. Abridge is the tool a note-writing doctor built to fix his own frustration, not a tool shaped by customer interviews. The founder touches the product every week as one of its users.

The mission Rao repeats on No Priors and in Fortune comes out in one sentence.

“We’re not going to fully automate doctors. We’re going to force-multiply them.”— Shiv Rao (Fortune)

Force-multiply is a military term. In this context it means giving one doctor more patient capacity in the same hour. Less paperwork, more time facing the patient. It is also the reason Abridge has stayed stubbornly listen-only. No AI that talks to the patient. An AI that sits next to the doctor, listens quietly, and writes the note. A single, deep lane.

There are three co-founders. Zack Lipton (CTO) is an associate professor of machine learning at Carnegie Mellon University, director of ACMI lab, and co-author of the open-source textbook Dive Into Deep Learning. Lipton’s research stance reduces to one line: “not every problem is a prediction problem”. That stance is the exact line Abridge drew between “AI that predicts disease” and “AI that removes administrative load from doctors.” Florian Metze (CSO) is a speech-recognition researcher trained at the CMU Language Technologies Institute, and the reason Abridge’s medical ASR (automatic speech recognition, the technology that turns audio into text) runs on its own trained model rather than public Whisper. Sandeep Konam was co-founder and original CTO.

In a line, the founding team is a four-way junction: a practicing physician CEO, CMU machine learning theory and education, the CMU speech technology lineage, and the original CTO. It is close to impossible to replicate that junction through outside hiring.

3. What the bet is — the eaves, and the four-layer house underneath

Here is the article’s thesis in one sentence.

Abridge’s bet was to use the 30 months as Epic’s first Pal, under the eaves, to finish building a four-layer house of its own underneath. When the eaves opened up, the house would still be standing.

The eaves-and-house metaphor, spelled out. The eaves are the embedded loop inside Epic, with audio captured in Epic Haiku (mobile) and notes reviewed in Epic Hyperdrive (desktop). In August 2023, Abridge became the first participant in the Epic Partners and Pals program as first Pal. That position meant every Epic-using hospital in the US considered Abridge first when shopping for an ambient scribe.

The problem: the eaves were borrowed on Epic’s goodwill. The moment Epic shipped the same feature itself, that borrowed value would vanish. Abridge knew this from the start and used the time under the eaves to build four layers of its own. Here they are.

Point 1 — The stack of paperwork (compliance layer)

HIPAA BAAs (Business Associate Agreements, the contracts that bind vendors handling medical data), SOC 2 audits, data-handling maps, and per-hospital legal and security reviews. Put plainly, this is the stack of documents a hospital needs to sign a new AI vendor. It is worked out hospital-by-hospital across 150+ networks. Epic shipping a native feature on February 5, 2026, and 150 BAAs being re-signed on February 6, 2026, are structurally different events. This friction layer is underestimated in software-migration stories and is Abridge’s deepest moat.

Point 2 — A cap table of investor-customers

Kaiser Permanente Ventures, CVS Health Ventures, Mayo Clinic, American College of Cardiology, California Healthcare Foundation, UC Investments. Each is an institutional actor that is both an investor and a customer, or something close to a customer. The cap table itself doubles as sales collateral. Pair this with the Pittsburgh origin (the research provenance of the Pittsburgh Health Data Alliance) and it is a structure new entrants find hard to copy.

Point 3 — A medical-specific ASR corpus (data moat)

50 million medical conversations a year feed Abridge’s in-house speech-recognition model. Coverage spans 55 medical specialties and 28 languages, with vocabulary, accents, and jargon distributions that are highly specific to clinical conversation. A text-first AI foundation vendor would need the same customer footprint for years to match the density. This is one of the few cases where “data moat” is doing real work.

Point 4 — Linked evidence and multi-EHR support (product differentiation)

Linked evidence is the feature that ties every sentence of a generated note back to the second in the source audio it came from. It lets compliance and medico-legal teams trace every AI-written line to an audio timestamp for audits or litigation. As of April 2026, the public spec for Epic AI Charting does not include this feature. Abridge also runs on Oracle Health, Athena, and Meditech. Multi-EHR reach takes surface that Epic AI Charting structurally cannot.

Chris Carmody, Chief Technology Officer at UPMC, says this.

“Abridge provides accurate notes, freeing our clinicians to fully connect with patients.”— Chris Carmody, UPMC CTO (Abridge press release)

What customers are actually paying for is not the eaves (Point 1, Epic integration) but Point 3 (note accuracy, a function of ASR corpus quality) and Point 4 (workflow embedding). The eaves are the entrance. The value has already moved inside the house.

4. Four peers, and the shape of Abridge

The ambient AI scribe category has four main players. To see Abridge’s outline from outside, place the other three next to it.

Ambience Healthcare (13% share) layers coding, after-visit summary, referral letters, and pre-visit chart prep on top of the scribe, a “scribe plus surrounding automation” bundle. Wide and thin by design. Suki AI (10%) leans on voice commands and can drive lab orders and referrals, a voice-command- centric environment. Nabla (4%) has the largest simultaneous-launch footprint, but Epic integration depth does not reach Abridge’s.

Put plainly, the four companies sit on different differentiation axes. The corner Abridge occupies (Epic integration depth crossed with a medical ASR data moat) is not one the others can fill simultaneously. Each has its own lane.

Desiree Gandrup-Dupre, senior vice president of care delivery technology services at Kaiser Permanente, describes the 40-hospital rollout this way.

“The largest implementation to date of ambient listening technology.”— Desiree Gandrup-Dupre, Kaiser Permanente SVP (Fierce Healthcare)

On the No Priors podcast, Rao offers one more interesting observation.

“Doctors are using Abridge as a verb.”— Shiv Rao (Fortune / No Priors)

Doctors saying “let’s Abridge today’s conversation.” Verbing (Google, Xerox, Uber) is not a sign that a category has commoditized. It is a sign that the brand has fused with the act. Switching cost is starting to be measured not in software but in habit.

5. The strongest counter-argument — the day the eaves went free

Take the strongest counter-argument against the thesis head-on.

“Epic covers 42.3% of US acute-care hospitals and 54.9% of beds. On February 5, 2026, it shipped AI Charting natively. 85% of Epic customers are already using some generative AI feature. HIT Consultant named the moment the Platform Squeeze, and MedCity News asked whether the ambient scribe startups have a future. If the eaves went free, does the house not come down with them?”

There are three responses.

Response 1 — For existing customers, switching is human behavior change, not software migration

Epic AI Charting will take hold fastest with new customers yet to choose. Replacing 12,000 UPMC clinicians and 24,600 Kaiser clinicians who already use Abridge is a different problem. Each of those clinicians takes months of change management (training plus habit rewrite) to trust a new tool. Enterprise-wide clinician adoption is orders of magnitude slower than software migration. Slower means, from Abridge’s side, harder to lose.

Response 2 — At launch, four product advantages remain with Abridge

(a) Linked evidence is not in Epic AI Charting’s public spec. (b) Multi-EHR reach: Abridge runs on Oracle Health, Athena, and Meditech; Epic AI Charting is structurally Epic-only. (c) 55 specialties by 28 languages is wider than Epic AI Charting at launch. (d) The 50-million-conversation medical ASR corpus. Matching that signal density requires redoing the customer rollout. What Epic’s native launch compressed is the distribution premium. Product premium stays where it was.

Response 3 — “Landlord ships a native feature” is more survivable than it sounds

MailChimp on the Shopify App Store. AppDynamics before Datadog rose. Independent ISVs on Salesforce AppExchange. Each lived through “landlord shipped something similar natively” and survived on product depth and installed-base trust. At a customer town hall in August 2025, Rao put it this way.

“All boats can rise here.”— Shiv Rao, Aug 2025 (StatNews)

That framing reads the native launch itself as external validation that “ambient documentation has become a baseline EHR capability.” The floor of the category rises. A higher floor forces partners to differentiate at the next layer. For Abridge, that next layer is Points 1-4 from Section 3.

Three pressures are worth naming honestly. Price compression, because Epic AI Charting is bundled into existing Epic contracts. Data access asymmetry, because Epic has native access to the full longitudinal patient record, Abridge via API. Regulatory exposure, because HIPAA, BIPA, and GDPR Art. 9 all land on the same recording event; the consent and licensing article in this series goes into detail. Against all three, Abridge holds structural answers: $778M cumulative runway (the largest in the category), multi-EHR portability, and a 150+ BAA regulatory-readiness posture.

Put plainly, on February 5, 2026, Epic and Abridge drew level at Point 0 (the eaves). They did not draw level on Points 1-4.

6. Landscape — the next chapter of medical voice AI is full-duplex

Abridge is not part of the speech-to-speech (STS, audio-in and audio-out AI) mainline that this series covers. The product is listen-only. The AI does not talk. Abridge does not fit any of the four full-duplex STS families mapped in the pipeline-to-integrated article.

That said, Abridge’s trajectory is a reference for other vertical voice-AI companies. Sierra to Salesforce Agentforce, Decagon to Salesforce Service Cloud, Parloa to Genesys and NICE. Each is running the same “borrow the eaves, build the house underneath” arc. Moveworks sold to ServiceNow for $2.85B in early 2025 and chose a different playbook (integrator, early exit). Abridge is running the opposite playbook: stay independent, compound installed base, hold price even when the eaves go free. Which playbook pays off better is an open question for vertical voice AI as a whole.

The next chapter of medical voice AI is full-duplex (telephone-like two-way real-time conversation). Hippocratic AI launched its AI Front Door and Nurse Co-Pilot on April 16, 2026, at Cincinnati Children’s, OhioHealth, Cleveland Clinic, and Wellspan. It is the first production deployment of a full-duplex patient-facing voice AI at top-tier US academic medical centers. Abridge is clinician-facing listen-only. Hippocratic is patient-facing full-duplex. Which side crosses the other’s boundary first will decide the next phase of medical voice AI.

At Epic UGM 2025, CEO Judy Faulkner spoke about internal AI work.

“Somewhere between 160 and 200 AI projects.”— Judy Faulkner, Epic UGM 2025 (Advisory)

With Epic having turned decisively toward being an AI-native platform, partner integrators face a binary: differentiate on Points 1-4, or exit via the Moveworks route. Abridge is the leading example of the first choice. The next Series F will grade it.

7. Two signals, and a note on getting in touch

To close, two signals that will let readers judge whether the bet lands.

Signal 1: does the next round re-price at $5.3B or above? Sacra is tracking the Series F as the leading indicator. Flat to up means the market credited Points 1-4 with carrying through the distribution compression. A down round confirms, after the fact, that Point 0 (the eaves) was load-bearing.

Signal 2: do UPMC and Kaiser continue their enterprise-wide contracts through 2026 without being replaced by Epic AI Charting? These two systems are the clearest measurement points for BAA and installed-base switching cost.

Abridge’s bet was the claim that “a vertical medical AI house can be built as an independent company, one a landlord cannot fully replicate even by shipping a native feature.” The evidence as of 2026 (150+ BAAs, 50 million conversations per year, four product advantages at the Epic AI Charting launch moment, and a CMU + UPMC + ACMI founding team) is consistent with that claim still holding. The two signals above will resolve the rest over the back half of 2026 and into 2027.