It is 2:47 a.m. somewhere in Ohio. A woman who has been told for fifteen years that her blood pressure is "borderline" picks up her phone, taps a smart cuff on her arm, and watches a number land in a familiar little chat thread. The reply comes back in seconds. Not from a doctor. Not from a nurse. From Lark. The message is plain. It asks whether she ate later than usual. It nudges, gently, that tomorrow's walk would help. She closes the app and sleeps. That conversation - millions of versions of it - is the thing Lark Health has spent more than a decade building.
Lark Health is the Mountain View company convinced that the real chronic-care emergency is not a shortage of medicine. It is a shortage of someone to talk to about it at 2:47 a.m. The product is an app shaped like a text thread. The brain behind it is an artificial-intelligence coach trained on cognitive behavioral therapy. The patients - roughly 1.5 to 2 million of them - are not paying out of pocket. Their health insurers are. Anthem and other major payers offer Lark as a covered benefit, which is how a digital-health startup ends up looking after more than thirty million eligible American lives.
The problem they saw
American healthcare is not famously bad at heroics. It can do an open-heart surgery before lunch. What it has always struggled with is the slow part - the daily, dull, decades-long management of conditions like type 2 diabetes, hypertension and obesity. Roughly six in ten US adults live with at least one chronic condition. Most of them get fifteen minutes with a clinician a few times a year and a referral to nutritionists, classes and pamphlets they will not attend. The system treats prevention the way airlines treat legroom: nice in theory, expensive in practice.
Julia Hu noticed this earlier than most, partly because she lived inside it. She grew up with an undiagnosed chronic condition. She watched the gap between what her family was told to do and what the healthcare system actually helped them do. The gap was not medical. It was behavioral. It was logistical. It was - and Lark would later monetize this exact insight - a gap that conversational software was uniquely suited to fill.
The wager
If a chronic patient could be coached, every day, by something patient enough not to roll its eyes and cheap enough to scale, outcomes would move. Insurers would notice. Insurers did notice.
The founders' bet
Hu co-founded Lark in 2011 with Jeff Zira. The company's first incarnation was - improbably - a vibrating wristband that nudged you to sleep more. It was charming. It was not the business. The pivot came when the founders realized the wristband was less interesting than the messaging layer wrapped around it. People did not need another device. They needed a relationship with one.
That insight turned Lark into something stranger than a wearable company and more useful than a chatbot. It became an early bet on conversational AI in clinical settings - years before "generative AI" entered the marketing deck of every health-tech startup on the planet. Lark demoed at an Apple WWDC keynote as one of the first HealthKit partners. It picked up early funding from people who, looking back, look prescient. By the time the rest of healthcare caught up to the idea of an AI coach, Lark already had years of real conversations to learn from.
The product, demystified
Open Lark and you will not find a dashboard. You will find a chat. The chat asks you a question. You answer. It logs your meal, your steps, your weight, your blood pressure, your blood sugar - whichever of those Lark's connected devices are watching. Then it does the thing chronic care has always promised and rarely delivered: it follows up. The next morning. The next afternoon. Forever.
Under the surface there is a behavioral-science framework built on cognitive behavioral therapy. CBT, in plain English, is the discipline of changing what you do by gently rearranging the way you think about what you do. Lark uses it because the alternative - the lecture - does not work. Patients have been lectured about salt and sugar for fifty years and Americans are no healthier for it.
What Lark actually offers
The product menu
Diabetes Prevention Program - CDC-recognized, smart-scale-equipped, designed for the 96 million American adults with prediabetes.
Diabetes Care - real-time coaching for adults already living with type 2.
Hypertension Care - a connected cuff plus the gentlest possible reminder system about medication adherence.
Weight Management & GLP-1 - clinical screening and behavior coaching wrapped around Ozempic, Wegovy and the rest of the new pharmacopoeia.
Behavioral Health & Wellness - stress, sleep, tobacco cessation, the rest of being human.
A condensed history of an AI nurse
The numbers that interest investors
The proof
The skeptic's question about every behavioral-health app is fair: does it actually work? Lark's answer is the part that brought the payers in. The CDC granted full recognition to its diabetes prevention program, a designation reserved for programs that hit specific weight-loss and engagement thresholds in real patients. Anthem - not a company famous for handing out covered benefits casually - offers Lark to members at no out-of-pocket cost, smart scale included. Returning institutional investors keep returning. None of that is conclusive. All of it is unusual.
The 2021 Series D - $100 million led by Deerfield Management - landed at a strange moment for digital health. Most of the noisy COVID-era unicorns were starting to look frail. Lark, which had spent a decade building infrastructure that scales without adding headcount, looked like exactly the kind of unsexy compounding business a healthcare specialist fund wanted to own. Crossover investor PFM Health Sciences joined. Franklin Templeton joined. The round brought the company's total funding to roughly a quarter-billion dollars.
The mission, said out loud
Lark talks about its mission the way only a founder with a personal stake can. The pitch is not that AI will replace clinicians. The pitch is that AI will finally let clinicians focus on the patients who genuinely need a human, by handling the daily coaching, logging, nudging and listening that no clinic can afford to do in person. Care, in this framing, becomes less episodic and more ambient. It is no longer a fifteen-minute appointment. It is a presence.
That presence is also a quietly progressive idea about equity. The patients who benefit the most from Lark's model are often the ones the system has historically served the worst - rural, lower-income, time-poor, transportation-poor. A phone-based coach does not care about your ZIP code. Insurers, increasingly held to outcomes measures, do.
What it adds up to
A platform where the marginal cost of one more patient is almost zero, where the clinical evidence is real, and where the customer (the payer) saves money when the patient gets healthier. Healthcare incentives almost never line up that neatly. When they do, somebody builds a serious business.
Why it matters tomorrow
The next decade of US healthcare will be defined by two forces colliding. GLP-1 drugs like Ozempic and Wegovy are rewriting what is medically possible for obesity and metabolic disease. At the same time, the system cannot afford to put every eligible patient on them without behavior-change support that protects the long-term outcome. That problem - drugs without coaches - is a Lark-shaped problem. The company expanded its weight management program in 2024 to include guided GLP-1 care delivery for exactly this reason. The future of chronic care is pharmacological. It is also, inescapably, conversational.
Which brings us back to Ohio at 2:47 a.m. The woman with the borderline blood pressure is asleep now. Her smart cuff is on the nightstand. Her phone is dark. Somewhere in a Lark data center a model has logged her reading, compared it against a trend, decided not to wake her, and queued a gentle question for the morning. The clinic is still closed. The chronic disease is still there. But for the first time in fifteen years, so is someone willing to ask her, every day, how she is doing - and to mean it in a way the system can actually afford.