He Built a $1.8 Billion Telehealth Company With Two Employees and a $20,000 Budget. Here Is the Architecture Behind It.
By Creatives Takeover · June 22, 2026
How two employees built an $18B healthcare company.
In September 2024, Matthew Gallagher started a telehealth company from his apartment in Los Angeles with $20,000 and no employees.
He was not new to building things. He had previously founded Watch Gang, a watch subscription company that grew to 60 employees and ultimately failed. That failure taught him something most founders only learn the expensive way. A bigger team is not automatically a better business. Headcount creates overhead, communication friction, and cost structures that do not move with revenue. He carried that lesson directly into what he built next.
The company was called Medvi. The product was access to GLP-1 weight-loss medications, the same category of drugs behind Ozempic and Wegovy, delivered through a direct-to-consumer telehealth model. The market opportunity was obvious. The execution was not.
In its first month, Medvi attracted 300 customers. The next month brought 1,000 more. By the end of 2025, its first full year in business, the company had generated $401 million in sales and served more than 250,000 customers. It is now tracking toward $1.8 billion in revenue for 2026.
The team that built this was Gallagher, alone, for the first several months. In April 2025 he made his only hire: his younger brother Elliot, brought on specifically to filter communications so Matthew could focus on strategy. That remains the entire headcount.
The Architecture Nobody Else Was Using
The most important decision Gallagher made was not a product decision. It was a structural one.
He did not try to build a vertically integrated healthcare company. He did not hire doctors. He did not open a pharmacy. He did not pursue a medical license. Instead, he identified the parts of the business that were regulated, capital-intensive, and would take years and millions of dollars to build internally, and he rented them.
Medvi partners with CareValidate and OpenLoop Health, two companies that handle the entire regulated layer of the business. Licensed physicians who review and approve prescriptions. Pharmacy fulfillment. Shipping logistics. Regulatory and HIPAA compliance. Everything that requires a license, a credential, or years of infrastructure investment was outsourced to specialists who already had it built.
What Gallagher kept for himself was the layer that actually drives revenue: the brand, the customer acquisition engine, and the customer experience. In an industry where the regulated infrastructure is commoditized and available to rent, the only thing that differentiates one telehealth company from another is how well it acquires and retains customers. Gallagher understood that distinction with total clarity, and he built his entire company around owning that one layer exceptionally well.
This is the core architectural insight. Identify the highest-value layer in your industry, usually brand, distribution, or customer experience, own that layer completely, and rent everything underneath it.
The AI Stack That Replaced a Company
Gallagher told The New York Times plainly: "It's not an A.I. company, but I did it with A.I."
The breakdown of what AI actually replaced is specific. ChatGPT, Claude, and Grok handled the majority of the platform's codebase, the marketing copy, and the website itself. Midjourney and Runway generated ad creative and video content. ElevenLabs handled voice-based customer communication. Custom AI agents connected these disparate systems together and monitored business performance in real time, effectively functioning as the analytics and operations team a traditional company would need to hire for.
Every function that traditionally requires a department, engineering, marketing, design, customer service, was handled by a different AI tool, orchestrated by one person. Traditional companies allocate roughly 70 percent of their budget to salaries. Gallagher's primary cost was LLM usage and advertising spend, not payroll. A solo founder's full AI tool stack in 2026 costs somewhere between $3,000 and $12,000 per year. Hiring a single employee costs upwards of $130,000 in salary alone before benefits or overhead. The math is not incrementally different. It is categorically different.
The Margin That Proves the Model
The clearest evidence that this architecture works is not the revenue figure. It is the profit margin.
Medvi operates at a 16.2 percent net profit margin. Hims and Hers, its closest large competitor in the same category, generated $2.4 billion in revenue with 2,442 employees and a 5.5 percent net profit margin. Gallagher is running nearly three times the margin of his largest competitor with a headcount that is 99.9 percent smaller.
That comparison is the entire thesis in one number. Hims and Hers built a large organization to do everything internally. Medvi built a small organization that owns one layer and rents the rest, with AI doing the operational work that would otherwise require dozens of people. The capital efficiency gap between those two models is not a rounding error. It is a structural advantage that compounds every single quarter.
The Part of the Story That Deserves Honesty
This architecture is genuinely impressive. It is also not the whole story, and a full account of Medvi has to include the part that generated real and warranted criticism.
Medvi's marketing website used AI-generated before-and-after weight-loss images. The people in those images do not exist. The transformations they depicted were never real outcomes from real customers. This is a meaningfully different thing from using AI to write code or generate ad copy. It is using AI to manufacture the appearance of clinical results that never happened, in a regulated medical category, to influence a purchasing decision about a person's health.
The unit economics of Medvi's business model are real. The customer acquisition numbers are real. The architecture of renting regulated infrastructure while owning distribution is a legitimate and replicable structural insight. None of that is undone by the deepfake controversy. But none of it excuses it either. The two things are both true at once, and a founder studying this story should walk away with the structural lesson while explicitly rejecting the shortcut that came with it.
The architecture does not require manufactured results to work. That is worth saying plainly, because the temptation to conflate "this growth tactic worked" with "this growth tactic was acceptable" is exactly the trap that produces more of this kind of controversy across the industry.
What This Means If You Are Not Building a Telehealth Company
Most founders reading this are not going to build a GLP-1 telehealth platform. The specific industry is almost beside the point. The structural lesson is what travels.
Every industry has a regulated, capital-intensive layer that takes years to build and a customer-facing layer that drives the actual revenue. The founders who win the next decade will be the ones who get precise about which layer they are actually in the business of owning, and rent everything else without hesitation or ego about needing to build it all themselves.
The second lesson is about sequencing. Gallagher did not spend months perfecting a product before anyone saw it. He spent his initial budget on customer acquisition first, then iterated on the product based on real demand. Most early-stage founders do the opposite, polishing a product in private while distribution remains an afterthought. The founder with a clear customer acquisition engine and an adequate product will consistently outperform the founder with a polished product and no distribution.
Five Things Worth Taking From the Architecture
Identify what you must own versus what you can rent. The regulated, capital-intensive parts of your industry are rarely where the differentiation lives. Distribution, brand, and customer experience usually are. Get precise about which one you are actually in.
AI replaces functions, not judgment. Gallagher used AI for code, copy, design, and customer service. He did not use it to decide what to build, who to partner with, or how to price. The architecture decisions stayed human. The execution did not.
Margin is the real scoreboard, not revenue. A 16.2 percent margin with two people is a fundamentally different business than a 5.5 percent margin with thousands of people, even when the revenue numbers look comparable on the surface.
Distribution first, polish second. Spending early capital on customer acquisition before the product is fully built is uncomfortable. It is also how you find out whether the demand is real before you have sunk months into something nobody wants.
Capability does not grant permission. The fact that AI can generate convincing before-and-after images does not mean it should be used to fabricate clinical outcomes. The architecture that built Medvi's economics is worth studying. The shortcut that damaged its credibility is worth rejecting outright.
Gallagher did not invent GLP-1 drugs. He did not invent AI tools. What he did was see, earlier and more clearly than most, which layer of a regulated industry was actually worth owning, and he built a company with the discipline to rent the rest.
That architecture is the real lesson.
What you build with AI is your decision.
What you choose not to fabricate is also your decision.