Started with $20K, Running $40B with Just 2 Employees: Dissecting Medvi's Complete Operations Blueprint

Let me get the conclusion out of the way first.

The core of Medvi isn’t the dollar amount. It’s the design of “what gets handed to machines, and what humans hold onto.”

Three domains AI handles, two domains outsourced, and one domain the two humans never let go of. Once you grasp this 3-layer structure, the dollar figures don’t matter. You walk away with a blueprint any solo founder, in any industry, can translate.

There’s a company that started with $20,000 (about ¥3 million) in initial capital and pulled in $401M (about ¥60 billion) in its first year.

It’s Medvi, launched by Matthew Gallagher from his Los Angeles home in September 2024. A telehealth company specializing in GLP-1 (the umbrella term for injectable medications used to treat obesity and diabetes). English-language outlets are reporting that the 2026 run rate (an annualized estimate based on current revenue) is on pace for $1.8B (about ¥270 billion). The headcount is two: just Gallagher and his brother Elliot. (PYMNTS.com)

“It’s medical, so it’s a special case, right?” I hear you. But before industry, look at the structure. Why does this run with just two people? That answer applies to any business.

I touched on Medvi’s numbers briefly in a previous article (→ Article explaining “10x founders”). This time I’m going deeper. I’m going to dissect the whole operational design.

Medvi’s 3-Layer Structure: Start with the Big Picture

Let me BLUF the overview of the 3-layer design first. Layer 1: AI Automation Layer (code, marketing, CS); Layer 2: Outsourced Compliance Layer (doctors, dispensing, prescriptions); Layer 3: Human Decision Layer (strategy, quality standards). Because these three layers are cleanly separated, two people can run it.

What Gallagher did wasn’t “do everything yourself” or “hand everything to AI.” He looked at the nature of each task and assigned it to the optimal owner. Things machines can handle, go to machines. Things requiring expertise, go to outside professionals. Everything else, and only that, humans hold.

There are two patterns solo founders tend to fall into. Either “I have to do everything myself,” or “AI will solve everything if I just hand it off.” Both are wrong. Medvi designed the middle ground.

One footnote: the $1.8B run rate is an annualized estimate, not a valuation from an external party. Multiple English-language outlets have reported it, but the calculation details are based on Gallagher’s own statements and aren’t fully disclosed. What I want to talk about today is “the design behind those numbers,” so I’ll use the figures only as context. (NewsNation)

img: Concentric 3-layer diagram. From the inside out: “Layer 3: Decision Layer (2 humans),” “Layer 2: Compliance Layer (outsourced),” “Layer 1: Automation Layer (AI tools).” | type: diagram | style: clean infographic, minimal flat design, soft pastel palette

The Roles of 3 AIs: How ChatGPT, Claude, and Grok Split the Work

Let me lay out the AI tools Medvi uses. ChatGPT, Claude, and Grok divide the labor across code, marketing, and CS. The core insight isn’t tossing everything at one AI — it’s matching tools to the nature of the task.

Here’s the division I can read from Gallagher’s public statements:

• Code generation and product development: ChatGPT, Claude → Backend builds, admin UI implementation. Built the platform without hiring an engineer
• Social ads and marketing copy: Grok → High affinity with X (formerly Twitter), generating copy that reflects real-time trends
• Image and video ads: Midjourney, Runway → Auto-generates ad creative. No designer
• Customer service: AI voice via ElevenLabs + Claude → The backbone of running 250,000 customer touchpoints with 2 people

“Three tools? Wouldn’t one work?” Fair question. But even within “AI,” different models have different strengths. The output required when writing code, writing social copy that lands, and talking to a customer are completely different in nature. Try to do everything with one tool, and everything comes out half-baked.

This is a takeaway you can use right now if you’re a solo founder. A lot of people are “throwing everything at ChatGPT,” but the moment you start matching specialized tools to specialized tasks, the output quality jumps. Step one is just asking the question: “Which AI is best suited for this particular task?”

The fact that Midjourney and Runway power the ad creative is also worth noting. If you can run marketing assets without a dedicated designer, the decision of “should I even hire a creative role?” itself changes.

“But can you trust the quality of AI-generated work?” When you’ve acquired 250,000 customers in year one, my answer is: it’s not my place to worry about that level.

”Neither Doctors nor Pharmacists Are Employees”: Why They Outsourced Compliance Entirely

I think this is the heart of the design.

Medvi has neither doctors nor pharmacists on staff. Through partnerships with two companies — CareValidate and OpenLoop Health — they’ve outsourced the entire “diagnose, prescribe, dispense” compliance chunk.

Why does this matter?

Trying to launch a business in healthcare, regulatory overhead is brutally heavy. Recruiting, training, and managing physicians; coordinating with dispensing pharmacies; building HIPAA-compliant (HIPAA is the U.S. medical privacy law) systems. Try to do it all in-house, and $20K of initial capital is laughably impossible.

Gallagher broke through this “expertise wall” by outsourcing to the pros in each domain. What you don’t need to do yourself, you hand to professionals. It’s not just about cost efficiency — there’s a major secondary effect: “you can separate regulatory risk from your own company.” What Medvi handles in-house is platform design and marketing. Everything else lives outside.

Translated to a universal principle, this works across industries. For tasks with an “expertise wall,” outsourcing to specialists is usually faster, more accurate, and cheaper.

For example, typical outsourcing candidates for a solo founder:

• Accounting & tax: Cloud accounting service with AI assistant + tax accountant
• Legal & contract review: AI legal services (AI features in services like Bengoshi.com)
• Recruiting & talent sourcing: Crowdsourcing platforms + scout AI
• Logistics & inventory (for e-commerce): Fulfillment services

What Medvi did wasn’t “throw money at the expertise wall.” It was a design choice: “send the entire expertise wall outside.” That shift in framing is huge.

What the 2 Humans Refused to Let Go: The True Nature of Decisions Not Given to Machines

So what do the Gallagher brothers actually do?

The answer is “strategy and product direction” and “setting quality standards.” They haven’t handed the core business judgment to any AI or outsourcing partner.

The GLP-1 market exploded in 2024-2025. The Ozempic and Wegovy boom sparked a flood of new telehealth services offering GLP-1 prescriptions. Medvi acquired 250,000 customers amid a saturated field of competitors because they kept making accurate calls on “which market, what to offer, and to whom.”

Pricing, competitive analysis, product direction. You can have AI sketch out the outline. They actually do that. But the final judgment stays with humans. Across multiple interviews, Gallagher has consistently held the line: “AI is the executor. We’re the ones who decide.”

There’s one more thing they didn’t let go of: “customer experience quality control.” AI voice handles the front of CS, but the design routes high-stakes problems to humans. They explicitly avoid a design where machines alone respond when customers are genuinely in trouble.

The lesson for solo founders, distilled: there are two situations where you can’t hand things to AI. The moment you’re deciding the direction of the business, and the moment customer trust is directly on the line. As long as humans hold those two, you can hand everything else over to machines without hesitation.

What Gallagher kept in the “human layer” all has something in common. They’re all “decisions you can’t redo.” Miss on business direction and the cost of correcting it skyrockets. Break customer trust and rebuilding takes forever. That’s exactly why humans hold them. AI is fast at execution. What humans should own isn’t execution speed — it’s the weight of judgment.

A “Role Design Document” for Solo Founders: Translating the Medvi Model into Your Own Business

I translated Medvi’s 3-layer structure directly into something solo founders can use. Copy and use it.

Layer 1: Automation Layer (handled by AI tools)

Task	Recommended Tools
Code & internal tool building	Claude, ChatGPT
Content & social copy	Claude, Grok
Image & video creative	Midjourney, Runway
First-line customer inquiry response	ElevenLabs, Intercom AI
Data analysis & monthly reports	ChatGPT (Advanced Data Analysis)
Newsletter & article drafts	Claude

Layer 2: Outsourcing Layer (handled by specialists / outside services)

Task	Example Outsourcing Partners
Accounting & tax	freee + cloud-based tax accountant
Legal & contracts	Bengoshi.com AI, LegalForce
Recruiting & freelancer sourcing	Lancers, CrowdWorks
Ad operations & SEO	Contract marketing specialists
Logistics (e-commerce)	Amazon Fulfillment, Logikura

Layer 3: Human Layer (you handle this, never hand it over)

• Business direction and strategic judgment
• Relationship-building with key clients and partners
• Setting product / service quality standards
• Escalated customer issues (final judgment)
• Cash flow and financial decisions

The belief that “I have to do everything myself” is what eats up most of a solo founder’s time. What Medvi proved is the fact that “if you split it into 3 layers, 2 people can move $40 billion.” Scale is irrelevant. This design philosophy works for a one-person freelance business too.

img: Comparison table of a 3-layer design document for solo founders. Left column: “Traditional do-it-all-yourself model.” Right column: “3-layer design model (Medvi-style).” Compares task allocation and time cost. Each row shows representative tasks and tools. | type: comparison-table | style: clean side-by-side layout, soft contrasting colors, clear typography

If You’re Starting This Week, Here’s the First Job to Let Go Of

For those saying “got it, let me try” — here’s the priority order to start with.

Tackle this week: 3 tasks to hand off to AI

1. Social post drafts → Just instruct Claude or ChatGPT: “Write a 300-character post for this reader segment on this topic.” You’ll find the rhythm in the first week

2. Templatizing customer email replies → List 5 common inquiry patterns and have AI generate replies for each. You can cut response time by an average of 60% (my own measurement)

3. Article and blog outlines → Just give the theme and target reader, and you get H2 structure and key points for each section. Eliminates the “thinking from zero” overhead

Tackle next month: 3 tasks to outsource

1. Monthly bookkeeping → Start with the AI assistant features in freee or Money Forward. Once auto-categorization is running, your tax accountant fees drop too

2. First-pass design work → Use Canva’s AI features to auto-generate banners and landing page assets. “Prototype with AI before outsourcing, and use AI to prepare the material you hand to the pro” — a flipped approach

3. Recurring research work → Have an AI agent run competitive analysis, market trends, and keyword research on a schedule. You get 2-3 hours back per week

I’ll be honest — when I first became a solo founder, I was hoarding everything myself. I genuinely believed “if I delegate, quality drops” and “doing it myself is faster.” But once I actually handed things off to outsourcers and AI, more tasks ended up with higher quality than when I was doing them. “Things only I can do” turned out to be overwhelmingly fewer than I thought.

img: Two-phase execution roadmap showing “This Week” and “Next Month” phases. Each phase has 3 actions with owner labels (AI / Outsource / Self) as tags. Simple timeline layout. | type: roadmap | style: clean horizontal timeline, soft pastel tags, minimal flat design

Wrap-up: What Medvi Proved Is “A Win in Design”

Just looking at the numbers, it looks like “something only a genius could pull off.” But the reality is different. This is a design story.

• $20K to $401M in year one, with a 3-layer structure behind it
• Give AI roles. ChatGPT, Claude, and Grok each have different strengths. Don’t throw everything at one AI
• Send compliance and specialized work to external partners. You don’t need specialists in-house
• What the 2 humans hold is only “strategic judgment” and “the final line of customer trust”
• If you’re starting now, hand “social posts, email reply templates, article outlines” — those 3 — to AI

“This is healthcare, so it doesn’t apply to me.” If that thought crossed your mind, hear me out again. What Medvi did wasn’t a healthcare story — it was an allocation design of “what gets handed to where.” Industry doesn’t matter. Whether you’re at ¥500,000 monthly revenue or ¥5 million monthly revenue, the principle of the 3-layer design doesn’t change.

If you’re saying “I don’t know where to start,” just look at this week’s task list. Mark anything where you think “couldn’t AI do this better?” That becomes your first design document.

The person who decides on the design moves ahead. Doers win. I’m out in front, and I’ll keep documenting the how-to.