Pay-per-patient

On April 2, The New York Times ran the most-shared founder story of the year: a Los Angeles entrepreneur named Matthew Gallagher had built a telehealth company, MEDVi, to roughly $401 million in 2025 revenue — its first full year — with a starting stake of $20,000, a dozen-odd AI tools, and a headcount of two. The company projects $1.8 billion for 2026 and claims a 16.2% net margin. Founded September 2024. Two employees. A billion-dollar run rate. The takeaway wrote itself, and a hundred thousand reply-guys typed it out: AI just built a billion-dollar company with two people.¹

I want to argue the opposite. The AI is the least interesting true thing about MEDVi. Strip the headline off and you find a machine that is older, simpler, and far more powerful than a stack of prompts — and you find, wired to the same machine, a fuse that was already lit when the profile ran. This piece is about both: the engine, because it's the real lesson and it's genuinely new; and the fuse, because the two are the same part.

The business was unfair before the AI showed up

Begin with the thing the headline skips: what MEDVi sells. It sells a monthly subscription to a compounded GLP-1 weight-loss drug — the cut-price cousins of Ozempic and Mounjaro — shipped to your door after a quick online consult. Hold the legality of that for a few minutes; it matters enormously, and it's the fuse. Look first at the shape of the business, because the shape is what prints money.

A direct-to-consumer telehealth subscription is one of the best business models software ever borrowed from medicine. There is no warehouse and no inventory: a partner pharmacy compounds and ships each order. There are no returns and no size-exchanges: it's a vial, not a hoodie. The revenue is recurring — a patient on a weight-loss protocol doesn't buy once, they re-up every month for as long as the number on the scale keeps moving. Compare it to the e-commerce business most founders cut their teeth on: a $50 average order, a thin one-time margin, 20% of it walking back through the returns door. Telehealth is the inverse on every axis — high ticket, high margin, high frequency, near-zero physical friction.

That shape has a financial consequence that is the hinge of this entire essay. Write the lifetime value of a patient as plainly as it deserves:

where p is the monthly price, c the cost to compound and ship, m the months a patient stays, and a what you paid to acquire them. In e-commerce the first term is small and m is often one. In subscription telehealth, the monthly margin (p − c) is fat and m is large — and that means the amount you can rationally spend to win a customer, a, can be enormous and still leave you with a profit. The unfairness isn't the AI. It's that a high-margin recurring product can outbid everyone on earth for the next customer and still come out ahead. Remember that sentence. It is the whole reason the engine I'm about to describe could exist.

And the timing was a gift. Through 2024 and 2025, demand for GLP-1 drugs ran years ahead of supply; the branded versions were expensive, rationed, and constantly out of stock. Compounded copies were cheap, available, and — for a window — legal. MEDVi didn't invent that demand. It positioned a clean, fast, cash-pay storefront directly in front of the largest consumer-health stampede in a decade. Telehealth made Gallagher rich. The AI just kept the org chart at two.

How unfair is the shape, concretely? Set it beside the public benchmark — Hims & Hers, the listed telehealth company that ran the same playbook a few years earlier and did about $2.4 billion in 2025 revenue at a roughly 5.5% net margin, with thousands of employees and a decade of building behind it.²

The margin and the headcount are the eye-catchers, and they're real enough to take seriously. But a number that small on the right column is not a triumph of automation. It's a tell. Two people did not do $401 million of work. Two people coordinated it. The work — all of it — was rented. Which is the actual innovation, and where the AI hype has been pointing in exactly the wrong direction.

The engine: pay-per-patient

I've spent a lot of this year writing about a different inversion in how attention gets bought — clipping, where a brand posts a price per thousand views and a distributed crowd goes and manufactures the views, and the brand pays only for the ones that land. Pay-per-view. The audience becomes the sales force. MEDVi is the same inversion, pushed one rung further down the funnel — past the view, past the click, all the way to the closed sale. Call it pay-per-patient.

Here's the mechanic. MEDVi itself is a brand, an offer, and a checkout funnel — that's the part the two people own. Customer acquisition is outsourced to third-party affiliate agencies who are paid a bounty for every patient they convert. They run the ads, write the advertorials, cut the "I lost 30 pounds" listicles, buy and whitelist the Meta placements, and point the traffic at MEDVi's funnel through tracked landing pages. One such agency, The Offer Inc., signed on in September 2024 and was, by its own court filings, billing MEDVi tens of millions of dollars a month for the patients it delivered.³ That is the engine. Not a model. A marketplace of strangers, paid per acquired patient, racing each other to feed a funnel.

Now recall the LTV identity. Because MEDVi's (p − c) × m was so large, the bounty a it could pay an affiliate for a converted patient was larger than almost any competitor could match. The unfair economics funded the unfair distribution: pay the highest commission in the category, and the best affiliates bring their traffic to you instead of the other guy. The recurring margin and the affiliate army are not two facts about MEDVi. They are the same fact, seen from two ends.

And the rest of the company? Also rented. The clinicians who write the prescriptions, the software that routes a patient to one, the pharmacy that compounds and ships the drug — none of it is MEDVi. It's a telehealth infrastructure provider, OpenLoop, and its affiliated pharmacy network (Triad Rx, Link Pharmacy) sitting underneath the brand like rails under a storefront.³ "Two employees" is true the way "a restaurant is one chef" is true — right up until you look in the kitchen and find it's a commissary, a delivery fleet, and a packaging plant that all belong to someone else.

Step back and the playbook is clean enough to name. Own the brand, the offer, the funnel, and the data. Rent everything else — acquisition, clinicians, pharmacy, even the intelligence — and pay for outcomes, not headcount. AI is one of the rented parts, not the protagonist: it writes the ad variants, drafts the funnel copy, answers the support tickets, so that two people can sit at the center of a machine that would have needed two hundred a decade ago. That is real, and it is new, and it deserves the attention it's getting — just pointed at the right noun. The miracle isn't that AI did the work. It's that the work no longer has to live inside the company at all.

$20K → $401M

Starting stake to first-full-year revenue, in ~18 months¹

~$200M

Revenue per employee — the entire firm is two people¹

500,000+

Patients acquired by April 2026, almost entirely through affiliates⁴

The same machine is the bomb

Here is the turn, and it is not a footnote. Every property that makes the engine fast makes it dangerous, because it's the same machinery. A distribution engine you don't own is a distribution engine that can detonate.

Start with the ads. When you pay strangers per acquired patient and ask no questions, you are buying performance and importing whatever produced it. What produced MEDVi's was, in part, synthetic: investigators documented thousands of active ads across Meta featuring AI-generated "patients," fabricated before-and-after transformations, and invented "doctor" personas — advertising that looks like authentic testimony and isn't.⁵ This is the same seam I keep flagging across the attention economy: the model pays on a number, and a number can be manufactured. The U.S. FTC's endorsement rules require that paid and fabricated endorsements be clearly disclosed; an AI actor swearing by a drug is the cleanest possible violation of them.⁶

The regulator was already there. On February 20, 2026 — six weeks before the Times profile — the FDA issued MEDVi a warning letter (#721455) for misbranding, including marketing that falsely implied MEDVi was itself the compounding pharmacy.⁷ Read that sequence again. The press canonized, in April, a company the FDA had formally warned in February. The founder myth and the enforcement file were open on the same desk; only one of them went viral.

And then the foundation itself. MEDVi's entire product is legal only because of a loophole that is closing. Compounding pharmacies are permitted to mass-produce copies of a branded drug essentially only while that drug is in shortage, under sections 503A and 503B of the federal rules. The FDA declared the tirzepatide shortage resolved in December 2024 and the semaglutide shortage resolved in February 2025; the enforcement grace periods for compounders lapsed through 2025.⁸ By 2026, mass-market compounded GLP-1 is, increasingly, simply not allowed to exist. MEDVi is not a company with a regulatory risk attached. It is a regulatory arbitrage with a company wrapped around it — and the arbitrage has an expiry date that has mostly already passed.

What's real, and what just goes fast

So is the MEDVi story a triumph or a fraud? The honest answer is that it's two different stories wearing one headline, and you have to cut them apart to learn anything.

The playbook is real, and it's the durable lesson. A single operator can now assemble a category-leading company out of rented parts: rent the demand generation, rent the clinical layer, rent the fulfillment, rent the intelligence, keep the brand and the margin, and pay for everything by the outcome. That pattern is going to mint companies for the next decade, in every vertical where the supply chain has been unbundled into rentable pieces — which, increasingly, is all of them. The person who repeats the X-thread version of MEDVi ("AI did it") will draw the wrong lesson and reach for a code generator. The person who reads the org chart will draw the right one and go assemble a distribution machine.

But velocity is not durability, and that's the second story. A business built on a product that is going illegal, sold through advertising it can't legally stand behind, growing on an engine that belongs to its contractors, is not a durable company. It's an arbitrage moving at the speed of a company — and arbitrages close. The interesting question MEDVi forces isn't "can two people build a billion-dollar company?" They evidently can. It's the harder one underneath: when assembly gets this cheap, the scarce thing is no longer the building — it's whether what you've built is allowed to keep existing. Distribution got cheap. Legitimacy didn't.

The right way to read MEDVi, then, is as a proof and a warning at once. Proof that the lean, rented, outcome-priced company is not a thought experiment — it's doing nine-figure revenue with a two-person letterhead. Warning that the same un-owned, un-disclosed, un-licensed machinery that makes it fast is exactly what will take it apart. The next great telehealth brand will run this playbook on a drug it's allowed to sell, with affiliates it actually polices, behind ads with real faces. It will grow slower than MEDVi. It will also still be here in 2028.