Knowtex is a 2022 women-founded company led by Stanford AI scientists that is headquartered in San Francisco, building ambient clinical intelligence to transform how clinicians capture and use medical information. Designed to be EHR-agnostic, specialty-specific, and deeply integrated into workflows, Knowtex enables providers to generate complete, accurate notes, codes, and orders in real time. By combining clinical-grade AI with enterprise-grade security and speed, Knowtex helps health systems and providers reclaim time, reduce burnout, and deliver comprehensive patient care.Knowtex is backed by Y Combinator, Amazon Web Services (AWS), the UCSF Rosenman Institute, and MedTech Innovators among others. To learn more, visit www.knowtex.ai
Recent Press Release on their rollout in Kansas below as well!
Knowtex Successfully Launches Ambient Clinical AI at VA Kansas City Site, Advancing Multi-Phase Rollout Across VA Primary Care
This is a founder story that announces itself through a single anecdote. Caroline Zhang, CEO of Knowtex, tells hers about a 75-year-old oncologist who had spent his entire career resisting technology.
“He had never used a scribe,” Zhang told me during a recent conversation at JPM Healthcare Conference. “He would print out twenty pages of paper every day for chart review and pre-charting. He liked to write up the notes too. He saw notes as his legacy.”
His verdict on ambient AI? “I’ll have to die before this technology can help me.”
Six months later, that same physician was using Knowtex for every patient visit—going from fifteen patients daily to over twenty—and spending his reclaimed hours reading clinical research papers rather than documenting encounters. His new position: “Do not take it away from me.”
The conversion of a single skeptic might seem like thin evidence. But it captures something essential about where clinical AI has landed in early 2026: the technology has crossed from “promising but unproven” to “infrastructure I cannot practice without.” And Zhang believes we’ve only glimpsed the beginning of what ambient technology can become.
Enjoy this article and podcast? Consider becoming a free or paid subscriber to Techy Surgeon!
The Operating System Thesis
Knowtex occupies an increasingly crowded market. Ambient scribes—tools that listen to clinical encounters and generate documentation—have become healthcare’s fastest-adopted AI category. Abridge recently raised $300 million at a $5.3 billion valuation. Nuance DAX, now owned by Microsoft, has deployed across thousands of sites. Epic announced native ambient capabilities at its 2025 User Group Meeting, sending a shiver through every startup in the space.
Against this backdrop, Zhang makes a distinction that shapes how she thinks about the market: scribing is a feature; ambient is a platform.
“When it comes down to it, I think workflows stem from the doctor-patient visit,” she explained. “Even intake and everything else…we want to work in the center of everything.”
If you enjoy reading founder stories and staying on the cutting edge of where clinical AI and health policy are going, consider becoming a paid or free subscriber to Techy Surgeon.
The argument is that whoever captures the clinical conversation captures the most valuable data in healthcare. Not claims data, which is downstream and filtered for billing. Not chart data, which is shaped by documentation requirements and time constraints. The actual conversation—what the patient said, how the physician responded, the nuances that never make it into a structured note.
“We have over a million clinical, real clinical conversations in our dataset,” Zhang noted. “That’s a novel dataset, and that informs the rest of the product and workflow design.”
The vision extends beyond documentation. Zhang describes a “voice AI operating system”—a command center that can surface fifteen years of patient history in a face sheet, alert clinicians to clinical trial eligibility mid-visit, complete coding and orders within thirty seconds of encounter end, and push everything back into the medical record for downstream billing and quality teams.
“It really becomes the central focal point for upstream and downstream workflows,” she said. “That can eliminate the reliance on EHRs that historically have been the data repository, the billing system, the reminder, the alert—all those types of things. Ambient can abstract all that away.”
The VA Breakthrough
The scale of Zhang’s ambition became tangible in October 2025, when Knowtex was selected alongside Abridge as one of two ambient AI vendors for deployment across the Department of Veterans Affairs—170 medical centers and 1,193 outpatient clinics serving 9.1 million veterans.
The contract, valued at $15 million for Knowtex’s portion, represents more than revenue. It represents federal validation for a company that emerged from Y Combinator’s Summer 2022 batch with modest seed funding. And it required solving a problem that has defeated larger, better-funded competitors: integrating with CPRS, the VA’s notoriously antiquated electronic health record.
“The VA and the government side—they’ve allocated CTO resources, so their team has been a true partner in that sense,” Zhang said. “That’s also how we approach integration as a whole. If we can work with the provider and we can work with the health system and they are willing to think beyond EHR limitations, then we can build out API connectivity, FHIR API connectivity, HL7—all the things.”
The subtext is important. CPRS, the legacy VistA system, is not known for its interoperability. That Knowtex achieved integration where others have struggled speaks either to unusual technical capability or unusual government partnership—likely both.
Zhang frames it as a product philosophy rather than a technical achievement. “I would recommend to other folks—still makes sense to be in EHR marketplaces. We’ve worked with Epic, Cerner, Athena, OncoEMR, Flatiron. We definitely have earned our stripes in EHR integration. But we believe there’s more that you can do with innovating directly with the provider.”
The Metrics That Matter
When I asked Zhang about the KPIs she tracks internally and shares with health system buyers, the list was specific:
Technology performance: 97.3% medical accuracy for audio processing, voice activity detection, and custom speech recognition. Over 99% accuracy for medical entity extraction—the translation of spoken information into notes, coding, and orders.
Clinician outcomes: Two hours saved per day for adopters. A 29% decrease in “pajama time”—the evening and weekend hours physicians spend completing documentation—documented through internal Epic Signal data.
Financial impact: Over $92,000 per month in additional revenue capture at implemented sites, primarily through surfacing ICD-10 codes and evaluation and management suggestions that busy clinicians would otherwise miss.
Adoption: Above 80-90% utilization across specialties, which Zhang considers the leading indicator. “Initially, when you roll out AI, it’s a lot to tell the hospital stakeholders that you’re going to get XYZ results for your system—that’s just not something that we can guarantee for an organization. But when you see good utilization, then we hear all the happiness and the time saved and the burnout reduction.”
The University of Rochester Medical Center case study, published in February 2025, offers independent verification of similar claims. URMC physicians reported reducing documentation time from ten hours weekly to two hours weekly. The revenue capture calculation—16,000 appointments monthly, 10% undercoding rate, approximately $58 revenue leakage per appointment—arrives at figures consistent with Zhang’s claims.
The First Customer
The conversion of the 75-year-old oncologist took six months. Zhang’s team was developing the product in his clinic, iterating on feedback from a user who represented the hardest possible case.
“Someone who delivered such quality of care but in his lifetime told us that the EHR had only increased his workload,” Zhang recalled. “He didn’t feel like he was operating at his competency level. He was spending two to three hours, four hours every night, just working on documentation.”
The breakthrough wasn’t a feature or a demo. It was sustained exposure to technology that actually worked—that didn’t require him to change his workflow but rather eliminated the parts of his workflow he had always resented.
“I think that’s the biggest win we can have with anybody,” Zhang said. “Now our technology is something that is a daily reliance in your life and that is adding value and that is making the joy of medicine possible for you again.”
The observation echoes across ambient scribe implementations: physicians often don’t realize how dependent they’ve become until the tool is unavailable. The ultimate test of product-market fit may be whether users panic when it goes down.
The Patient-Facing Question
Our conversation turned to a topic generating considerable anxiety in clinical circles: patient-facing AI. OpenAI launched ChatGPT Health in early January 2026, offering medical record integration through b.well and enterprise rollouts to HCA Healthcare, Cedars-Sinai, and Stanford Medicine. Anthropic followed days later with Claude for Healthcare, emphasizing HIPAA-ready infrastructure and partnerships with Banner Health, Sanofi, and Novo Nordisk.
Both products target patients directly, positioning AI as a first-line resource for health questions, a development that evokes memories of “Dr. Google” but with vastly more sophisticated capabilities.
Zhang’s response was measured but firm.
“I’m going through some of a journey here too,” she acknowledged, describing a personal shoulder injury that led her to run her own MRI report through GPT and Claude. “Ultimately when my decision-making process comes down to it, I’m still trusting the expert. I’m still trusting the orthopedic surgeon and the PT as the person quarterbacking my medical care.”
The concern isn’t that AI can’t provide useful information. It clearly can. The concern is about accountability, confidence calibration, and the subtle ways AI can speak authoritatively about matters where uncertainty should dominate.
“When we’re in healthcare, we believe people still want humans being the face of the interaction,” Zhang said. “We believe that people want expertise. It’s really about supercharging the clinician, not replacing them. We believe that it’s about making the experience more human and not less.”
This framing positions ambient scribes as fundamentally different from patient-facing AI. The clinician remains in the loop, signing off on notes, reviewing coded suggestions, approving orders. The AI augments rather than replaces clinical judgment.
But the lines may blur faster than anyone anticipates. If ambient technology evolves into an “operating system” that surfaces recommendations, flags clinical trial eligibility, and alerts clinicians to missed diagnoses, how different is that from patient-facing AI that does similar things for the patient directly?
“It’s about being hallucination-proof,” Zhang emphasized. “Let me say best-in-class model. Even in the ambient side, we can say there’s a spectrum of tools that have different performance. Contextual linked evidence from the chart history—those are metrics we’ve co-designed with clinician leaders and clinician stakeholders.”
The distinction between clinician-facing and patient-facing AI may ultimately depend less on the technology’s capability than on who bears responsibility when it’s wrong.
Techy Surgeon is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
The Epic Question
No conversation about ambient AI in 2026 can avoid the elephant in the room: Epic’s entry into the market.
At the August 2025 User Group Meeting, Epic unveiled three AI assistants—Art for clinicians, Emmie for patients, Penny for revenue cycle—with licensing beginning in 2026. The announcement represented a fundamental shift. EHR vendors had historically been platforms for third-party innovation; now Epic was building competing products directly.
Survey data suggests 67% of outpatient providers may switch ambient vendors within three years. The question is whether startups can differentiate fast enough to survive commoditization.
Zhang’s answer is speed.
“That’s also the thing—velocity is going to be their Achilles heel,” she argued. “You can’t wait for another decade of a clinician’s life, or a year of a clinician’s life, to have tools that can save hours today.”
The thesis is that Epic, constrained by its incumbent customer base and enterprise release cycles, cannot innovate as rapidly as focused startups. Whether that proves true over a multi-year horizon remains uncertain. But it explains why Knowtex and peers are racing to expand from documentation into ordering, clinical trial matching, and full-workflow integration—building defensible positions before the giant catches up.
The Advice
When I asked Zhang for the best advice she’d received as a founder, she returned to relationships.
“Build long-term relationships in the industry,” she said. “Things move at speed of trust. Today I was meeting with somebody I’d met two JPMs ago that told me they were not ready for ambient in 2023. It would have to be an RFP in 2025. And lo and behold, it’s 2025. They’re ready.”
The worst advice? “That there is a set playbook that you have to follow for adoption of technology.”
Her counter-lesson: find the customers who understand what you’re building. Some hospitals aren’t ready for ambient AI—and no amount of persuasion will change their procurement timeline. Others are actively searching for innovation. The founder’s job is to identify which is which without wasting months on the former.
“You’ll talk to those individuals—they’ll say this is not possible,” Zhang reflected. “But then you just have to go find somebody who understands the magic you’re creating and the value you’re generating.”
What Comes Next
Zhang’s 2026 roadmap centers on order entry—the logical extension of ambient documentation. If the AI can capture the clinical conversation accurately enough to generate a note, it can capture it accurately enough to propose the orders implied by that conversation.
“I think this is a year where having order entry is going to be table stakes for ambient solutions,” she predicted. “You go into a room, you walk out. Now it’s not just a note, but you are going to expect that your ambient tool can do all those things—notes, coding, orders from the conversation.”
Beyond that, she envisions embedded research access, clinical trial matching, and the gradual absorption of workflows that currently require context-switching between systems.
“The clinician can be able to have evidence-backed research inside the ambient platform or inside their patient visit workflow,” she said. “Not having to switch screens—not having to look up things in GPT or Open Evidence or Doximity—but to actually do this right in the ambient workflow.”
Whether this vision succeeds depends on factors beyond any single company’s control: regulatory evolution, payer acceptance, Epic’s competitive response, and the uncertain boundary between clinician augmentation and patient-facing AI. But the direction is clear. The clinical encounter—the conversation itself—is becoming the center of gravity for healthcare AI.
The 75-year-old oncologist who thought he would die before AI could help him may have been the leading indicator. The question now is whether the technology can scale fast enough to reach the millions of clinicians still waiting—before the landscape shifts again.
Knowtex is backed by Y Combinator, Amazon Web Services (AWS), the UCSF Rosenman Institute, and MedTech Innovators among others. To learn more, visit www.knowtex.ai
Christian Pean, MD, MS is CEO and Co-Founder of RevelAi Health, Executive Director of AI & IT Innovation at Duke Health, and Assistant Professor of Orthopaedic Surgery at Duke University. He writes the Techy Surgeon newsletter on clinical AI and health policy for surgeons and health system leaders.
Techy Surgeon is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit techysurgeon.substack.com/subscribe