TL;DR: Medvi hit $401 million in first-year revenue with two employees, a 16.2% net margin, and $20,000 in startup capital. Its competitor Hims & Hers did $2.4 billion with 2,442 employees at a 5.5% margin. This isn't a fluke. Citadel Securities, SBE Council, and a June 2026 Harvard/INSEAD study all confirm AI is collapsing minimum efficient scale across the economy. The doctrine: systems beat slogans, and headcount is a cost you add only when a system genuinely needs human judgment.
The reactor doesn't care how many people are in the room
Medvi did $401 million in revenue in its first full year. Two employees. A 16.2% net margin. Startup capital: $20,000.1
Matthew Gallagher and his brother Elliot run the whole operation. AI tools handle branding, ad creative, customer service, and much of the compliance paperwork. Gallagher outsourced the regulated pieces, the doctor network and pharmacy fulfillment, to existing infrastructure providers, then used AI to run everything else as a system rather than a department. Compare that to Hims & Hers, a direct competitor in the same telehealth category: $2.4 billion in revenue, 2,442 employees, and a 5.5% net margin.2 Medvi runs nearly three times the margin with a headcount 1,200 times smaller.
I ran a nuclear reactor with twelve men. Not twelve hundred. Twelve. On the USS Jefferson City, the watch team in the engine room kept an entire warship moving on a system that could kill everyone aboard if a single procedure got skipped. Every man on that watch was a system operator, not a job title. Nobody's role was "supervise the guy who supervises the valve." The valve had a procedure. The procedure had an owner. That owner was accountable for the outcome, full stop.
That is what minimum efficient scale used to mean before AI broke it. Minimum efficient scale is the smallest size a business needs to reach before its unit economics work. For a century, that number kept climbing. More customers meant more support reps, more billing clerks, more middle managers to coordinate the support reps and billing clerks. Headcount was the price of revenue. AI just canceled that invoice for an entire category of businesses, and the data says this isn't a fluke. It's a pattern.
The doctrine: systems beat slogans
At DEMG we run one doctrine sentence through every decision: systems beat slogans. A slogan is "we're customer-obsessed." A system is a documented, repeatable process that produces the same outcome whether you're watching it or not. Medvi's real innovation wasn't a clever ad. It was treating every business function, marketing, support, even parts of compliance, as a prompt inside a system, not a headcount line inside a budget.
The Owner's Exit Engine framework we build with clients starts from the same premise. A business that depends on its founder's daily attention is not an asset. It's a job wearing an LLC. The engine has three gears: documented systems that run without you, AI operators that execute those systems at machine speed, and a dashboard that lets you audit the whole thing in fifteen minutes a week. Medvi built that engine by accident. Most owners have to build it on purpose.
The numbers say this is structural, not anecdotal
Skeptics will call Medvi a one-off. The macro data disagrees.
Citadel Securities' July 2026 research note, using Census Bureau data, found new U.S. business applications hit 5.6 million in 2025, up 24% since ChatGPT's public launch.3 More striking: the median seed-stage startup's employee count dropped from five to four in the first quarter of 2023, one quarter after ChatGPT shipped. Citadel's framing is precise. A company that once needed five employees to justify its early overhead now needs four. That headcount isn't eliminated. It's deferred until the business hits the scale where human judgment, not task execution, becomes the binding constraint.
The SBE Council's 2026 Small Business Technology Use Survey backs this up from the ground level: 82% of small business employers have adopted at least one paid AI tool, and the typical small business now runs five different AI tools across its operations.4 This isn't a curiosity confined to Silicon Valley telehealth startups. It's Main Street.
A Harvard Business School and INSEAD working paper published in June 2026 quantified the organizational shift. Studying more than 2,900 Y Combinator startups, researchers Hyunjin Kim and Rembrand Koning found AI-native firms employ roughly 25% fewer workers than non-AI startups in the same industry and funding cohort, yet carry comparable valuations.5 Their hierarchies run half a seniority level flatter. Their engineer share runs 13 percentage points higher. The takeaway isn't that AI-native firms cut corners. It's that they generate more value per employee because AI is built into the product, not bolted onto the workflow as a productivity toy.
And the growth gap is already showing up in revenue. PYMNTS Intelligence's May 2026 Small Business Week report found SMBs generating more than $1 million annually grew 13.7% year over year, while businesses earning under $150,000 grew just 0.6%.6 The largest SMBs grew at nearly 23 times the pace of the smallest. Scale used to buy you efficiency through headcount. Now efficiency buys you scale without the headcount, and the businesses that figured that out first are pulling away from the ones that didn't.
This isn't just telehealth arbitrage
The pattern holds outside consumer health. Cursor's maker Anysphere hit $100 million in annual recurring revenue roughly twelve months after launch, then kept compounding: $500 million by mid-2025, $1 billion by November, $2 billion by February 2026. Team size stayed in the 30 to 50 person range through most of that climb. ElevenLabs hit $100 million ARR in two years with about 50 people. Mercor scaled from $1 million to $100 million ARR in 11 months with roughly 30 employees on payroll.
None of these are two-person shops. But all three prove the same underlying mechanic Medvi exploited at a smaller scale: the marginal cost of serving the next thousand customers is inference, not headcount. A traditional SaaS company hired ahead of revenue to build the sales machine that would eventually justify the hire. These companies let the product do the selling, the onboarding, and the support, and they hired only where a human judgment call was the actual bottleneck.
What this means for your business, this quarter
If you're running a service business, a consultancy, or an agency, the Medvi story isn't a headline to admire from a distance. It's a balance sheet warning. Every employee you add to handle a task an AI system could run instead is a recurring cost against a one-time investment. That math used to favor headcount because the tools weren't good enough. In 2026, for a widening set of functions, that math has flipped.
This is where the ATLAS Model comes in. ATLAS stands for Audit, Triage, Layer, Automate, Sustain. You audit every recurring task your team performs. You triage which tasks require human judgment versus which require task execution. You layer AI systems onto the execution tasks first. You automate the handoffs between systems so nothing needs a human to shuttle data between tools. Then you sustain it with a weekly quarter-hour review, not a full-time manager.
Most owners skip straight to "automate" and wonder why it breaks. The audit and triage steps are what separate a business that scales its margin from one that just scales its chaos. Medvi's founders didn't skip those steps. They just did them instinctively, treating every function as a system before they ever called it a system.
The counterargument, and why it's incomplete
Some analysts note Medvi's "two employees" figure is doing some rhetorical work. The company reportedly uses contract account managers and outside advertising agencies alongside its two full-time hires.7 Fair point. But that critique actually strengthens the doctrine rather than undermining it. Contractors on a project basis are a variable cost tied to output. W-2 employees on a fixed salary are a fixed cost tied to headcount. The core insight survives: Medvi's founders refused to convert every function into a permanent fixed cost. They kept the org chart lean and let systems, whether AI or contracted specialists, absorb the volume.
Ken Griffin, founder of Citadel, made a related point at Goldman Sachs' Apex Symposium in June 2026: AI hasn't cost anyone at his firm a job, but it has compressed six to eight weeks of research work into two to three hours, freeing talented people to go after more opportunities rather than fewer.8 The lesson isn't "fire your team." It's "stop hiring to compensate for a system you haven't built yet."
Build the engine room before you build the org chart
On a submarine, you don't add crew because the boat feels short-staffed. You add crew when a specific watch station has no qualified operator and the casualty drill proves you need one. Every other headcount decision gets rejected at the door. That discipline is uncomfortable for founders raised on the assumption that growth equals hiring. It shouldn't be. Growth equals systems that compound. Headcount is what you add only when a system genuinely requires human judgment that AI cannot yet supply.
Medvi is not a template everyone can copy line for line. Telehealth, GLP-1 demand, and outsourced medical infrastructure created a specific opportunity. But the underlying principle travels to every industry: minimum efficient scale is falling, and the businesses treating that as a design constraint rather than an anomaly will out-compound the ones still hiring the way their industry always hired.
If you're building toward an exit, this matters even more. Buyers price businesses on systems, not on the size of the team running them. A lean, AI-operated business with clean margins is a more attractive acquisition target than a bloated one with the same revenue. Read our breakdown of the founder dependency tax for the full math on why buyers discount owner-run businesses, and what fixes it.
Frequently Asked Questions
Q: Does "minimum efficient scale is falling" mean I should fire my team? No. It means you should stop adding headcount as your default response to growth. Audit which tasks in your business require human judgment versus which require repeatable execution. Build AI systems for the execution tasks first. Reassign or hire only where a genuine judgment gap remains. Firing people to chase a headline is not a system. It's a reaction, and reactions don't compound.
Q: Is Medvi's $401 million figure representative of what a typical small business can achieve? No, and it shouldn't be treated as a template. Medvi benefited from a specific market: high-demand GLP-1 telehealth with outsourced regulated infrastructure. The revenue-per-employee ratio is an outlier even among AI-native companies. The transferable lesson isn't the dollar figure. It's the operating principle: treat every business function as a system with an AI-literate operator instead of assuming each function needs its own headcount line.
Q: How is this different from just using more software? Software you configure once and forget is a tool. A system in the doctrine sense has an owner, a documented procedure, and a feedback loop that improves it over time. The SBE Council found the typical small business already uses five AI tools.4 Tool adoption alone didn't produce Medvi's margins. Treating the tools as an integrated operating system did.
Q: What's the biggest risk in trying to run lean with AI instead of hiring? Under-triaging. Owners get excited about automation and skip the audit step, so they automate a task that actually required a human judgment call. The result looks efficient for a month and then produces a customer service failure, a compliance gap, or a bad hiring decision made by a bot with no accountability. Run the audit and triage steps of the ATLAS Model before you automate anything customer-facing.
Q: Does this trend apply outside the US market Citadel and PYMNTS studied? The specific figures are US-focused, but the mechanism, AI collapsing the headcount required to reach a given revenue level, is not geography-specific. Labor cost structures and regulatory environments will change the exact numbers, but the direction of the curve holds anywhere AI tools are available and businesses are willing to redesign their operations around them rather than layering AI on top of an unchanged org chart.
Jeff Barnes is the founder of Digital Evolution Marketing Group (DEMG). This article reflects operational experience, not investment advice. Results vary by business, market, and execution. Do your own due diligence.
Sources: 1) Forbes/Yahoo Finance, "How Medvi Found Success With Just $20,000 And AI," April 2026. 2) Business Insider, citing New York Times Medvi profile, April 2026. 3) PYMNTS, "AI Is Quietly Fueling America's Small-Business Boom," July 2026. 4) SBE Council, "The Digital State of Small Business," March 2026. 5) Harvard Business Review, "How Agentic AI Supercharges Startups and Threatens Incumbents," July 2026, citing Kim and Koning, "AI-Native Firms," HBS Working Paper 26-090, June 2026. 6) PYMNTS Intelligence, "How Digitally Fluent Small Businesses Are Reshaping Main Street," May 2026. 7) Canopy Press, "How a $20,000 AI Startup Hit $401M With 2 Employees," April 2026. 8) Hedge Fund Alpha, "Citadel's Ken Griffin: AI Will Spark A Golden Age Of Startups," June 2026.