A Guy Built a $401M Business in Year One With $20K and No Employees. The Skill Behind It Isn't Prompt Engineering.

In September 2024, Matthew Gallagher launched Medvi, a GLP-1 telehealth platform, out of his Los Angeles home with $20,000 in capital, more than a dozen AI tools, and one employee: his brother. In 2025, Medvi's first full year, the company posted $401 million in revenue and $65 million in net profit, a 16.2% margin. It's now generating more than $3 million per day and tracking $1.8 billion in 2026 sales.

The headline number is extreme enough that it invites dismissal. Don't dismiss it yet. There's a real lesson here about how solo operators are building in 2026, and there's also a cautionary footnote that most coverage skips.

What Medvi actually is

Medvi connects patients with licensed healthcare providers for GLP-1 prescriptions (Ozempic, Wegovy, tirzepatide) entirely through a digital async workflow. Patient fills out an intake form, a licensed provider reviews it and issues a prescription, a compound pharmacy ships the medication. Medvi handles the platform, the provider matching, the patient communication, and the back-office. The AI tools handle most of the grunt work.

The business model worked because of regulatory context: in 2024-2025, compound pharmacies could legally produce and ship compounded semaglutide to fill a nationwide shortage. That window created a massive unmet demand in GLP-1 prescriptions, and Medvi moved fast to capture it. Gallagher didn't invent the market. He built the fastest intake-to-prescription pipeline for it.

None of that required a hundred engineers. It required a smart use of AI tools to handle intake processing, provider-patient matching, patient communication, prescription routing, billing, and customer support, and a disciplined system for giving those tools the right context to work reliably.

Context engineering, explained without the buzzwords

"Context engineering" has been showing up in every indie hacker newsletter for the past few months. Most explanations are frustratingly vague. Here's what it actually means.

Prompt engineering is crafting the right one-shot prompt to get a good output. Context engineering is building the system that ensures the AI always has the right information at the right time to complete a task reliably, regardless of who runs it or when.

The concrete difference: a prompt engineer might write a great prompt for reviewing a patient intake form. A context engineer builds the CLAUDE.md file that tells the AI assistant what a compliant intake review looks like, what edge cases to flag, what the provider's decision criteria are, what to do when information is missing, and what the escalation path is. Then they write another file that documents the patient communication templates and which one to use in which situation. Then they document the tool interfaces. Then they write structured handoffs between the intake review agent and the communication agent.

At the end of that process, you have a system where a junior contractor following the documentation could do the work, and so can an AI. That's the point. Context engineering is documentation-as-operating-system.

For Medvi, this meant building AI-readable documentation for every repeatable task in the intake-to-prescription workflow, so the AI tools could execute reliably without Gallagher making a judgment call on every edge case. At $3M per day in revenue, you cannot make every judgment call yourself.

What solo-scale capital efficiency actually looks like

The conventional startup math: headcount drives execution speed, so you hire to grow. The AI-augmented math: context-engineered AI agents can handle most repeatable work at $200-$500/month in tool subscriptions, and headcount becomes the bottleneck you add only when the work can't be systematized.

Medvi operated with two people and a stack of AI tools during the $401M year. That's not a stunt. At $65M in net profit on $401M revenue, the cost structure was genuinely lean. The AI tools were doing intake processing, patient communication, billing operations, and customer support triage.

The capital efficiency number that stands out: Gallagher invested $20,000 and made $65M in net profit. That's a 3,250x return on invested capital in 12 months, without VC funding, without a large team, without a San Francisco office. The capital efficiency is only possible because the AI tools replaced the headcount that would otherwise be required.

The cautionary footnote that most coverage skips

Medvi is not a clean success story. It has three significant problems.

In February 2026, the FDA issued Medvi a formal warning letter for misleading product claims: specifically, the company was marketing compounded semaglutide in ways that implied FDA-equivalent safety and efficacy testing that compounded drugs don't have.

In March 2026, a class action lawsuit was filed in California alleging that Medvi was benefiting from affiliate spam networks, with at least 100,000 people potentially involved in the claim. The lawsuit alleges the company's rapid growth was partly driven by deceptive marketing practices.

Also in March 2026, a security researcher found that Medvi's patient health records were exposed through sequential, unauthenticated URLs: all 250,000 patients' personal health information, accessible without a login. The company fixed it after disclosure, but it was sitting open.

These aren't minor operational issues. An FDA warning, a class action, and a HIPAA-level security vulnerability in the same quarter suggest that the system Gallagher built was optimized for growth speed, not compliance or security. When you're running a healthcare operation at scale with two people and AI tools, and you skip the compliance review that a larger team would catch, the FDA warning letter is what you get.

The lesson isn't "context engineering doesn't work." The lesson is that context engineering for growth is easier than context engineering for compliance. Documenting how to process an intake form is simpler than documenting a compliant HIPAA information security policy and then actually implementing it.

What the Medvi model actually translates to for a solo operator

The $401M number is specific to a regulatory arbitrage window in GLP-1 prescribing that has mostly closed: the FDA cracked down on compound semaglutide availability in late 2025. You cannot replicate it directly.

What does translate: the operational architecture. If you have a business with repeatable processes (customer intake, content production, client communication, data analysis, fulfillment routing), those processes can be documented and context-engineered for AI execution. The result is a system that scales without proportional headcount.

The key question to ask about any task in your operation: can I write documentation so clear that an AI tool with no prior knowledge of my business could execute this task correctly 95% of the time? If yes, context-engineer it. If no, figure out why not: usually it's missing information, ambiguous decision criteria, or a genuine judgment call that requires human experience.

The honest take: Medvi proves the model works at extreme scale. Most solo operators don't need $401M scale. They need $50K-$500K revenue with sustainable margins and no full-time employees. At that scale, the context engineering overhead is even lower and the compliance risk is manageable. The architecture is right. The cautionary footnote is about what happens when you skip the unglamorous parts.

Author

Lukas

@lukcombinator

Sources

• From $20K to $1.8B in Revenue: How One Founder Built a Telehealth Giant With AI and No VC: Dealroom
• A $1.8 Billion Startup With Just 2 Employees Was Hailed as the Future. Now, the Negative Allegations Are Piling Up: Yahoo Finance
• The One-Person Unicorn: How Solo Founders Use AI to Build Billion-Dollar Companies: NxCode
• Matthew Gallagher's Medvi: How AI Built a $1.8B Telehealth Startup: Mirror Review