Arine.

On a Tuesday morning, an algorithm flags a prescription that should never have been written.

It happens quietly. A 71-year-old enrolled in a regional Medicare Advantage plan is taking nine medications. Three of them interact. One was discontinued at her last hospital visit but never removed from her chart. Her pharmacy, her primary care doctor, and her cardiologist each see a slice. Nobody sees the whole. By Tuesday afternoon, a pharmacist working from Arine's platform has the entire picture, a ranked list of risks, and a draft intervention. By Friday, she has a phone call. By the end of the quarter, she is not in the hospital.

Multiply that by 30 million lives. That is Arine in 2026.

Americans take 4.5 billion prescriptions a year. A surprising number of them are wrong.

Adverse drug events send roughly 1.3 million people to the emergency department annually. Non-adherence and medication-related problems are estimated to cost the US healthcare system upwards of $500 billion. The strange part is not that the problem exists - it is that everyone in healthcare knows about it, and yet most of the work to fix it has, until recently, been done with Excel spreadsheets and phone calls.

This is the gap Yoona Kim noticed. Kim is a PharmD with a PhD in pharmaceutical sciences from UCSF. She had spent years in clinical research watching the same pattern: a patient on a complex medication regimen, a pharmacist who could spot the problem in five minutes, and a system that gave the pharmacist neither the data nor the time to find them.

Her co-founder, Penjit "Boom" Moorhead, was a software engineer. The two of them did the obvious thing, which is also the thing nobody had done. They built software that does the looking.

The bet was that AI would not replace pharmacists. It would finally let them do their jobs.

This is the unfashionable version of the AI-in-healthcare story. There is no chatbot. There is no robotic dispenser. Arine's platform ingests claims, EHR data, lab values, and social determinants - then ranks who is most likely to be harmed by what they are taking, drafts an intervention, and routes it to a human clinician for the last mile.

That last sentence is the entire thesis. Kim and Moorhead bet that the highest-leverage place to put machine learning was not in the diagnosis, and not in the prescription, but in the boring middle: the work of figuring out which 5% of a health plan's members are about to fall off a cliff, and what specifically to do about it.

It is not an obvious bet. It is also, by most accounts, the correct one.

Three products. One uncomfortable question: who is the riskiest person on your roster, and what are you going to do about them?

The platform is HITRUST certified, hooks into Epic and Cerner, and runs on the predictably unglamorous AWS stack that healthcare buyers expect. The interesting part is not the architecture. It is the ranked list it produces every morning: a finite, prioritized set of patients a clinician can actually call in a workday.

Receipts, as the kids say.

This is the part of the healthcare-AI story where most companies hand-wave. Arine's numbers, drawn from public statements and customer case studies, are unusually specific. They are also the reason five national health plans and seven Blues plans signed up.

The customer list - 30+ plans, five nationals, seven Blues - is the harder receipt. Health plans do not switch vendors for vibes. They switch for star ratings and for the medical loss ratio. Arine moves both.

"The safest, most effective, and most appropriate treatment." Plain words. Hard to argue with.

The official mission statement is unglamorous on purpose. There is no talk of disruption, of revolutionizing care, of rewriting the future of medicine. There is a quiet assertion that every person should get the right drug, the right dose, at the right time, and a company built to make that happen at scale.

What is interesting is who funds this version of the future. Town Hall Ventures, which led the Series C in June 2025, is built around investing in care for underserved populations. Kaiser Permanente Ventures wrote a check on behalf of an integrated delivery system that runs its own pharmacies. Neither of those is a coincidence. Both organizations look at medication mismanagement and see the same line item: a preventable cost showing up in their own data.

Arine sells them the tooling to make it go away.

Why it matters tomorrow.

Two things are about to happen at once. Medicare is aging into its largest cohort in history, and the regulatory environment around CMS Star Ratings is tightening every year. Health plans that cannot demonstrate measurable improvement in medication adherence and outcomes will lose stars, and stars are worth billions in bonus payments. That is the unglamorous demand curve under Arine's growth.

Then there is the quieter shift: generative models that can read clinical notes, draft pharmacist outreach, and triage millions of records overnight. Arine has been doing the hard part of that for years - the data plumbing, the HITRUST controls, the integrations into Epic and Cerner. The new models are not the product. They are an accelerant on a platform that already worked.

Which brings us back to Tuesday morning, and the 71-year-old whose interacting prescriptions got caught before they did damage. She does not know Arine exists. She probably never will. The company is happy with that.

The bet was that the right place to put machine learning in healthcare was not in front of the doctor or the patient, but quietly underneath both. Eight years and $84 million in, the bet looks like it is paying off.