How True Is "My Only Employees Are AI"? 26 Solo-and-Side AI Businesses, Median ¥1.5M/Month

“By 2026 there will be a company of one human and 49 AI agents.” “The first one-person unicorn will come from an AI-staffed company.” Claims like these scroll past daily now. Bracing stuff — with almost no verifiable figures attached.

This site holds 26 cases of businesses run solo or as side projects with AI at their core (published set as of August 24, 2026, machine-classified). Every one has sources. Let us audit the “AI-only workforce” story against these 26, instead of against the discourse.

The 26, and their revenue distribution

A median of ¥1.5M a month, about 60% above the all solo-and-side median of roughly ¥900K. AI-centered businesses do out-earn the average solo profile.

But the order of magnitude hasn’t changed. The measured version of “line up AI employees and one person runs a nine-figure business” is a top quartile of ¥6.56M a month, and even the very top (Stella, an MVP in 2 days, $350K a month) sits in the ¥50M range. And what pushed Stella to the top was not AI; it was the 4 million followers its founder already had.

The jobs AI has taken over, and the ones it hasn’t

Cut the 26 by function and what AI actually replaces is clear.

Development is being replaced. Formula Bot reached $23K a month with zero lines of code. Stella shipped an MVP in two days. Chatbase, which a student ran solo to $1M ARR in 117 days, was six weeks of assembling off-the-shelf parts. The wall around building has genuinely fallen.

Support is being replaced. Chatbase’s product is the AI support employee, and PhotoAI, run with zero staff at $132K a month, openly automates its inquiries.

What is not replaced: distribution and judgment. The top cases all held distribution assets before AI, Stella’s 4M followers, Levels’ enormous X account, HeadshotPro founder Danny Postma’s track record and audience. AI mass-produces product. It produces exactly zero of the power to get that product seen. If anything, as the product moat collapsed, value migrated toward the humans who own distribution, the single clearest lesson of the 26.

AI employees are not unpaid

The payroll of an “AI workforce” is the API bill. Among our cases is Leadverse, where LLM API fees are 90% of cost of revenue. Not hiring humans means carrying a metered cost that scales with revenue, a fixed cost traded for a variable one, not for zero. And the rate card is set unilaterally by the model provider. An “AI-staffed company” is more accurately a company whose wage negotiation counterpart changed from its employees to OpenAI.

Nor is “one person” strictly one person. Post Bridge uses contractors when needed. Spot outsourcing for accounting or design appears throughout the 26. Zero employment is real. Zero human labor is not. On the exit side, Base44 (development delegated to AI, 8 employees, sold for $80M six months in) shows that “few humans plus AI” is becoming a configuration the acquisition market itself prices.

What the three shutdowns show about decay speed

The 26 include three shutdowns: Avatar AI, closed by its founder’s own hand after earning $150K in a week. Content Goblin, an AI article generator that fell from $15K MRR to under $1K in eighteen months. And Cydoc, a physician’s seven-year health AI that stopped at four paying customers. As we wrote in Where AI businesses stand, generative-AI demand decays as fast as it ignites. AI employees demand no severance, and vanish in an instant, together with the business.

Stay away from unverifiable numbers

Finally, the caution that motivated this count. Self-reports like “AI agents run everything; $500K a month within three months of launch” are circulating right now. Most cannot be checked by any third party, and the teller is often also the seller of a “let AI run your company” platform. We limited this piece to 26 cases because our listing rule counts only sourced actual amounts, and under that rule the present state of the AI-staffed company is a life-sized number: a median of ¥1.5M a month. The 40x gap between the discourse and the measurement is a gap in verification, not in AI capability.

Method and limits

The population is our published cases only. Classification is mechanical (AI-related terms in title/summary/tags, plus solo or side-project operation). Revenue figures are as of each source’s date. Overseas amounts are converted at ¥150/USD. Operators who publish numbers skew successful, so the median is likely inflated. Definitions follow the Data page.

Browse the individual cases via the AI business case list and the solopreneur case list.