A family physician closes her last chart at 10:47 p.m. Her toddler is asleep. Her dinner is cold. Somewhere in San Francisco, a piece of software has been listening to her patients all day - and the only thing it asked of her was nothing.
Freed is the reason that note got finished before midnight. The product is technically called an "ambient AI medical scribe," but the clinicians who use it - and there are now more than 20,000 of them - tend to describe it in less industry-flavored terms. They say it gave them back their evenings. They say it lets them look at patients again. They say it is the first software in healthcare they have ever wanted to pay for with their own money.
Which, in a sector famous for committee purchasing and seven-figure rollouts, is the most interesting thing about it.
01The Problem They Saw
Medical documentation is the dark matter of American healthcare. It is invisible from the outside, it weighs nothing on paper, and it bends the entire system. For every hour a U.S. physician spends with a patient, she spends roughly two hours on the EHR. The acronym for what this produces is "burnout." The plainer English word is "leaving."
By 2022, this was not a niche complaint. Surveys put physician burnout above 50% across most specialties. Practices were losing senior clinicians not because the medicine had gotten harder but because the typing had. A generation of trained physicians was discovering, very late, that they had unwittingly signed up to be data entry clerks who occasionally see sick people.
Erez Druk, an engineer at Meta, was watching this from the kitchen. His wife, Gabi, is a family physician. Her notes came home with her. They sat on the table during dinner. They kept her awake.
02The Founders' Bet
Druk made a bet that turned out to be unfashionable in 2022 and very fashionable shortly thereafter: large language models had crossed the line from "fun toy" to "competent listener." A model could sit through a patient visit, understand what was said, and produce a usable clinical note. Not perfectly. Usably.
He left Meta. He recruited Andrey Bannikov, another Meta engineer, as co-founder and CTO. They did not start with a press release or a pitch deck full of TAM. They started by shipping something Gabi could use on Monday morning. Then they shipped something her colleagues could use. Then her colleagues' colleagues.
This sounds quaint. It is also, in hindsight, the most strategic decision they made. Freed's earliest user was a real doctor whose feedback the founders heard every night at home. That feedback loop is not a slide in a deck. It is a marriage.
03The Product
Freed runs in a browser tab or a phone app. The clinician hits "start visit." She talks to her patient the way she always has. When she is done, Freed has written the note - structured, formatted to her chosen template, in her preferred style, ready to paste into the EHR. The whole interaction with the software, end to end, is one button.
There are clever pieces under the hood. Specialty-aware templates. Learned formatting per clinician. A visit-prep mode that pulls in prior context. A Chrome extension that handles EHR copy-paste. Mobile apps on both stores. HIPAA compliance, SOC 2, the whole compliance trellis you have to grow before any clinician will let you near a patient.
But the design choice that matters is the absence of design choices. Freed asks the clinician to learn nothing. There is no template-builder workshop. No two-week implementation. No IT ticket. You sign up, you record a visit, you have a note. The product hides everything that would normally require a consultant.
04The Proof
Freed crossed 20,000 paying clinicians by early 2025. The product runs across more than 1,000 small healthcare organizations - independent practices, single-shingle providers, the long tail of American medicine that big enterprise sales teams routinely ignore. Users report saving two to three hours a day. The App Store rating is 4.8, which is the kind of number you usually see on dating apps and meditation timers, not on enterprise software.
How Freed grew, roughly
The growth happened almost entirely through word of mouth. Doctors talk to other doctors. Nurse practitioners talk to other nurse practitioners. When a physician walks out of the exam room with two free hours, she tells someone. Freed's marketing department, such as it is, is mostly that.
// Freed, abridged
05The Mission
"Focus on people, not paperwork" is the tagline on Freed's homepage, and it is one of those rare marketing lines that survives contact with the product. The company's mission is small in scope and enormous in implication: take a single, profoundly hated task out of a clinician's day, and let the time that opens up flow toward patients.
It is worth noting what Freed is not trying to do. It is not trying to diagnose. It is not trying to replace the clinician. It is not trying to sell an "AI doctor." This restraint is unusual in 2026 healthcare AI, where the temptation to oversell the model is nearly irresistible. Freed has chosen the boring promise. The boring promise is also the believable one.
06Why It Matters Tomorrow
The AI scribe market is now crowded. Abridge raised a much bigger round. Microsoft has DAX Copilot. Suki, Heidi, Nabla, DeepScribe, Augmedix - the list grows weekly. The case for Freed is not that it is the only one. The case is that it is the one independent clinicians keep choosing for themselves, on their own credit cards, without an enterprise sales cycle. That is a quieter market than the hospital-system deals the press tends to cover. It is also, by clinician count, a larger one.
If Freed is right - if the ambient AI scribe becomes a standard part of practicing medicine - then the next decade of healthcare gets a small, structural gift: a few hours back, per clinician, per day. Multiply that by every general practitioner, psychiatrist, and physical therapist in the country, and the math gets serious. Two hours times five days times millions of clinicians is not a productivity upgrade. It is a reclamation project.
Back to the kitchen. The family physician closes the laptop at 10:47 p.m. - but in the new version, she closed it at 6:30. She read a book. She tucked the toddler in. The notes were already done, accurately, in her own voice, by the time she walked out of clinic. Freed did not perform medicine. It performed something humbler and more useful: it picked up the pen.