The first one-person billion-dollar company is here. The real story is uglier than the headline.

In late 2025, someone asked Dario Amodei when we’d see the first billion-dollar company with a single employee. He said 2026, with 70-80% confidence. Sam Altman reportedly has a CEO group chat where they’re betting on the same question.

They can stop betting. It happened.

Matthew Gallagher launched Medvi out of his LA apartment in September 2024 with $20,000 and a stack of AI tools. No employees. No investors. No office. Eighteen months later, The New York Times verified his numbers: $401 million in first-year revenue, 250,000 customers, 16.2% net profit margin. The company is tracking toward $1.8 billion in 2026.

Two people run the whole thing. Gallagher and his brother Elliot, who he brought on in April 2025 to help filter communications.

If you stopped reading here, this would be the most inspiring founder story of the decade. A guy with twenty grand and AI tools who outperformed companies with hundreds of employees. The tech Twitter version of this story is already written: “See? AI changes everything. The playing field is level. Anyone can do it.”

But I kept reading. And the story gets a lot messier.

What Gallagher actually built

Medvi sells compounded GLP-1 weight loss drugs through a telehealth platform. Semaglutide and tirzepatide, the active ingredients in Ozempic and Mounjaro, but compounded versions that are cheaper and don’t require going through the branded manufacturers.

The AI stack is real and worth studying. Gallagher used ChatGPT, Claude, and Grok to write the platform code and website copy. Midjourney and Runway generated the ad creative. ElevenLabs handled voice-based customer interactions. Custom AI agents connected the business systems and provided real-time analytics.

That’s the kind of setup that would have required a 40-person team three years ago: developers, designers, copywriters, a customer service department, a data team, a marketing department. Gallagher replaced all of it with monthly AI subscriptions totaling maybe $10,000-$12,000 a year.

The business model itself isn’t complicated. GLP-1 drugs are the hottest product category in healthcare right now. Demand massively outstrips supply through traditional channels. Telehealth removed the friction of doctor visits. Compounding pharmacies provided the supply. AI-generated ads on Meta and Google drove the customer acquisition. Gallagher found a demand spike, removed every friction point between the customer and the product, and used AI to automate the middle.

This part of the story is worth studying. Gallagher didn’t have some secret AI technique nobody else has access to. He identified a market where people were desperate to buy, figured out how compounding pharmacies could supply it, and used AI to wire the whole thing together in weeks instead of months. The insight was the market. The AI was just the build tool.

Then the FDA showed up

On February 20, 2026, six weeks before The New York Times ran their profile, the FDA sent Medvi a warning letter.

The charges: misbranding compounded drugs. The FDA said Medvi’s website falsely implied that the company was the compounder of the semaglutide and tirzepatide it sold. Claims like “Same active ingredient as Wegovy and Ozempic” suggested FDA approval that didn’t exist for the compounded versions.

Medvi wasn’t alone in this. The FDA sent similar warnings to more than 70 telehealth companies in the past six months. At least 30% of them were affiliated with just four nationwide medical groups: Beluga Health, OpenLoop, MD Integrations, and Telegra. The whole sector has regulatory problems, not just Medvi.

But here’s the thing about running a company with two people and a lot of AI: there’s nobody whose job it is to worry about FDA compliance. No legal team reviewing the marketing copy. No regulatory affairs department checking claims before they go live. The AI wrote the copy. The AI ran the ads. And the AI apparently didn’t know (or care) that “same active ingredient as Ozempic” carries legal weight.

I want to be clear: this isn’t an anti-AI argument. But when you automate everything, you also automate the blind spots. And nobody programmed the LLM to flag “hey, this marketing claim might violate FDA misbranding rules.”

The data breach nobody talks about

OpenLoop Health, the medical infrastructure partner that Medvi relies on for its telehealth operations, got breached.

A threat actor claimed to have exfiltrated records from approximately 1.6 million patients. OpenLoop notified the Texas Attorney General in March 2026, confirming at least 68,160 affected individuals in that state alone. Patient names, emails, phone numbers, medical information, medication orders.

And this wasn’t the only security issue. A security researcher found that after submitting an intake form on Medvi’s site, you could change one digit in the URL and access another patient’s complete record. No login required. No authentication. Just increment the number and you’re looking at someone else’s name, email, phone, weight, goal weight, and medication order.

I keep coming back to this detail. A healthcare company processing hundreds of millions in revenue, handling sensitive patient data, and the access control was a sequential number in a URL. This is what happens when AI writes the code and nobody reviews it for security. The code works. It compiles. It processes payments. It just also exposes every patient’s records to anyone who can count.

If you’re building with AI, this should scare you more than any job displacement headline. The risk isn’t that AI takes your job. The risk is that AI builds something that looks like it works, passes every surface-level test, and has a gaping hole that a first-year security student would catch.

The customer service problem

Medvi has over 11,400 Trustpilot reviews with a 4.4-4.5 average rating. That’s good. Not great for healthcare, but good.

Look at the negative reviews and a pattern appears: double charges, surprise $299 monthly bills after cancellation, missing shipments, and refund processes that customers describe as nightmarish. These aren’t isolated complaints. They’re systemic.

When your customer service is AI-powered and your headcount is two, who fixes a billing error at 2 AM? Who handles the edge case where the system charged someone twice and they can’t get through to a human? Who makes the judgment call on whether a specific refund request is legitimate?

AI customer service is good at the 90% case. The straightforward questions, the status checks, the routine stuff. It’s terrible at the 10% that requires empathy, judgment, and authority to override a system. And in healthcare, that 10% can involve someone’s medication access, their financial stress, their medical data.

The AI chatbots also had “hallucination” problems. They fabricated drug prices and claimed to sell products that didn’t exist. When you’re selling pharmaceuticals, a hallucination isn’t a funny anecdote. It’s a potential health risk.

So what’s the actual lesson here?

The wrong lesson: “AI lets anyone build a billion-dollar company.” That’s the version optimized for Twitter engagement and it’s roughly as useful as “just work harder.”

The right lesson, I think, has three parts.

First, market selection matters more than your AI stack. Gallagher didn’t succeed because he used AI well (though he did). He succeeded because he found a market where demand was explosive, supply was constrained through traditional channels, and the product delivered measurable results. If he’d applied the same AI stack to selling, say, consulting services, nobody would be writing about him. The $401 million came from the GLP-1 market, not from ChatGPT. AI was the accelerant. The fuel was a massive wave of consumer demand for weight loss drugs.

Second, AI removes the cost of building. It does not remove the cost of getting things right. Compliance, security, customer trust, regulatory relationships. These are the things that Medvi is now scrambling to fix after the fact. And they’re exactly the things that AI can’t handle without human oversight. The FDA doesn’t care that your AI wrote the misleading copy. Patients whose data leaked don’t care that the insecure code was generated by an LLM.

Third, the one-person billion-dollar company might be possible, but it might not be desirable. Gallagher’s net profit margin is 16.2%. That’s about $65 million on $401 million in revenue. Impressive for two people. But a company with even a small team handling compliance, security, and customer service might have avoided the FDA warning, the data breach, and the billing complaints that now threaten the whole operation. Sometimes the reason companies have employees isn’t inefficiency. It’s because certain functions require human judgment that AI cannot yet replicate.

The playbook for AI builders who are paying attention

If you’re a founder watching this story, here’s what I’d take from it.

Ride waves, not technology. The AI stack is available to everyone. What isn’t available to everyone is Gallagher’s market timing on GLP-1 drugs. Find the wave first. Then use AI to move faster than anyone else on it. The tools are a multiplier. Zero times a big multiplier is still zero.

Automate the middle, not the edges. AI is perfect for the work between the customer and the product: marketing, operations, analytics, content. It’s dangerous at the edges where mistakes carry legal, medical, or financial consequences. Keep humans at the boundaries. Let AI handle the throughput.

Build security in, not on. Medvi’s URL vulnerability is a textbook example of what happens when AI-generated code ships without security review. If you’re using AI to write code, you still need someone (or at least an automated system) checking for basic access control, authentication, and data exposure. The LLM doesn’t think about attack vectors. You have to.

Your AI customer service has a ceiling. It handles volume. It doesn’t handle exceptions. The moment a customer has a real problem that falls outside the normal flow, they need a human. Budget for that human, even if it’s you. The alternative is the kind of billing complaint spiral that tanks your Trustpilot score and eventually attracts regulators.

Speed of scaling is not speed of learning. Medvi went from $0 to $401 million in a year. But the organizational learning, the compliance infrastructure, the security practices, those don’t scale at the same rate as revenue. The gap between how fast you can grow and how fast you can mature your operations is where the risk lives.

The $3,000 solopreneur stack vs. the $401 million question

The economics of the one-person company are real. A full AI stack in 2026 runs $3,000 to $12,000 per year. That’s 95-98% cheaper than hiring humans for the same output. Operating margins of 60-80%. I’ve priced this out myself, and the numbers hold up. One person with the right tools can genuinely cover what used to take a mid-sized team.

But Medvi’s story reveals the gap between capability and responsibility. You can build fast. You can ship fast. You can scale fast. None of that means you should skip the parts that require slow, careful, human thinking.

I think the one-person billion-dollar company is real. I also think the first version of it is going to look more like a cautionary tale than a victory lap. Medvi might clear $1.8 billion in 2026. It might also face an FDA enforcement action, a class-action lawsuit over the data breach, or a regulatory crackdown on compounded GLP-1 telehealth that kills the market entirely.

The prediction came true. Whether it’s a success story depends on what happens next.

Where this goes from here

The trend isn’t going away. The cost advantages are too large. When you can cut operational costs by 95%, you change the math on which businesses can exist and who can build them. We’re going to see a lot more Medvis. Some will be in healthcare. Others in fintech, e-commerce, education, legal services. Anywhere the traditional cost structure creates a gap between what customers want and what they can afford.

But there’s a reason most businesses in history have had employees, and it isn’t that nobody thought of being more efficient. Certain functions exist because they protect the business from itself. A compliance officer catches the marketing claim that triggers an FDA letter. A security engineer finds the URL vulnerability before a researcher does. A customer service manager makes the judgment call that keeps a billing complaint from becoming a class action.

I suspect the companies that actually last won’t be the ones that minimize headcount to zero. They’ll be the ones that figure out the right ratio: where AI handles 80% of the work and humans handle the 20% that carries real consequences. Not because humans are better at everything, but because for compliance, security, and customer escalations, the cost of an AI mistake can exceed the cost of a human salary by orders of magnitude. One data breach, one FDA enforcement action, one class action, and the savings from not hiring evaporate.

The question was never really “can one person build a billion-dollar company with AI?” The answer to that is clearly yes.

The better question: should they?

Medvi’s next twelve months will tell us a lot about the answer.

Previously: 40 minutes, 500,000 machines: the LiteLLM attack every AI builder should study | The $300 billion quarter: what Q1 2026 venture funding really means for AI founders | Why OpenAI killed Sora: the $15M/day lesson every AI builder needs to hear

Frequently asked questions

What is the first one-person billion-dollar company?

Medvi, a GLP-1 telehealth startup founded by Matthew Gallagher in September 2024, is widely cited as the first company approaching billion-dollar revenue with effectively one or two employees. Gallagher started with $20,000 and used AI tools for code, marketing, customer service, and operations. Medvi posted $401 million in first-year sales and is tracking toward $1.8 billion in 2026 revenue.

How did Medvi use AI to build a billion-dollar company?

Gallagher used ChatGPT, Claude, and Grok to write platform code and website copy. Midjourney and Runway generated ad creative. ElevenLabs handled voice-based customer communication. Custom AI agents connected business systems for real-time analytics and automated customer service. The full AI stack costs roughly $10,000-$12,000 per year, replacing what would traditionally require 40+ employees.

Is Medvi legitimate?

It’s complicated. Medvi holds LegitScript certification and its revenue was verified by The New York Times. But the company received an FDA warning letter in February 2026 for misbranding compounded drugs. Its infrastructure partner OpenLoop Health suffered a data breach affecting potentially 1.6 million patients. Customers have filed complaints about double charges, unauthorized billing, and difficult cancellation processes.

Can one person really build a billion-dollar company with AI?

Technically, yes. Medvi shows it’s possible to generate massive revenue with near-zero headcount using AI tools. Whether it’s sustainable is another question. The regulatory, security, and customer service issues at Medvi suggest that removing humans entirely creates risks that can threaten the business itself, especially in regulated industries.

What AI tools do I need to build a solo company?

A competitive solo founder stack in 2026 includes LLMs (Claude, GPT, Grok) for code and content, image generators (Midjourney) for creative, voice AI (ElevenLabs) for customer communication, and workflow automation to connect systems. Total cost: $3,000 to $12,000 per year, depending on scale. But as Medvi’s experience shows, the tools alone aren’t enough. You still need human judgment for compliance, security, and customer escalations.