Every week now, someone on my feed announces they've bought an accounting practice, a letting agency, a law firm... some services business... and automated it until it "basically runs itself." One YC company, Rational, has gone the whole way and calls itself the first zero-human accounting firm. And this spring the newspapers thought they'd found the holy grail of the genre... Medvi, a telehealth business reported by the New York Times to have done $401 million in first-year sales with two people. Sceptics, Forrester among them, say that story looks less clean up close.
The idea has been circling for a while... Sam Altman said back in 2024 that his tech CEO group chat ran a betting pool on the year the first one-person billion-dollar company appears. But lately the claims have moved from group chats into my corner of the market, with real money attached, and I kept wondering the obvious thing. Is any of it actually true?
So I spent some time sorting it out. The brags reduce to three distinct claims, and they deserve three different verdicts.
Claim one... "I built a company alone"
True... in software. Maor Shlomo started building Base44 alone in December 2024. It launched in February, did nearly $1.5 million in revenue in its first month, and turned $189,000 of profit in May after paying its LLM bills. In June, Wix bought it for $80 million in cash... roughly half a year from first line of code to exit. He'd had an eight-figure offer on the table before hiring a single person, and a team of eight by the closing.
Around him, the club is filling up. Gamma runs $50 million of ARR and 50 million users with about 30 people, and says it has been profitable for well over a year. Lovable hit a $1.8 billion valuation with 45 employees, eight months after launch. Bolt went from zero to $20 million ARR in two months with 15. And solo-founded startups have gone from 23.7% of new companies in 2019 to 36.3% by mid-2025, by one analysis.
Even here, though, where the story is truest, the famous numbers wobble. Cursor's viral "$40 million of revenue per employee" was corrected by one leaderboard that re-checked it, once Anysphere's real headcount surfaced... past 300 staff, which puts the figure near $7 million. Midjourney's legendary ratio got withdrawn by the same leaderboard, because public headcount estimates for the company run anywhere from 40 to 163 people, and any ratio built on that range is invented precision, in their words. The tiny-team revolution is real, even if its most-quoted numbers are partly folklore.
Every company behind this claim sells software. A product built once and sold many times. Nobody's licence is on the line, and no client needs a hand held.
Claim two... "I bought a firm and it runs itself"
Half-true. The efficiency gains are real... the "runs itself" part is not.
Nobody serious in services is reporting zero people... they're reporting ratios that would have been fantasy three years ago. Inside traditional firms, the vendors sell agents whose outputs arrive labelled "ready for review," and that label marks the exact edge of the automation. Not filed, not signed... reviewed.
Dwelly buys UK letting agencies and, by its own count, runs more than 300 properties per member of staff against an industry norm of around 100. Minerva, a YC-backed AI-native accounting firm, bought a practice and says it took the operating margin from 5% to 70% by rebuilding it around agents. Crescendo reports customer support margins roughly four times the call centre norm.
Klarna says its assistant now does the work of 853 customer service employees... and Klarna is also the company that cut too deep, admitted publicly that the AI-first push went too far, and started hiring humans back last year with a promise that customers can always reach a person.
The real half is already here. The five-person firm doing what took fifty is the near-term reality for accounting, legal, property and every other practice.
Claim three... "firms won't need humans at all"
Not a lie... a bet. On society, not software.
Rational's founders are honest about what stands in the way. Agents can do the work and hold identities, they write, but they don't carry trust... which is why every AI accounting startup today does the work and leaves judgment to a human. Their own launch note concedes that humans currently hold the first 25% of the business, the client relationship, and the final 25%, the sign-off. Their bet is that as agents spread through daily life, society starts trusting them, and trust-heavy industries can then run end to end on AI.
Maybe. But four walls stand in the way, and none of them is technical.
The licence... in most of the US, an audit firm must be owned by licensed CPAs, which is why every private equity "acquisition" of one is actually engineered as a split structure. The human with the licence is the regulatory product, not overhead the AI hasn't reached yet. The liability... someone has to carry professional indemnity and be sueable when the work is wrong, and an agent cannot be negligent in any way an insurer recognises. The client... nearly every AI agent in accounting runs in the back office, and the roll-ups themselves are structured so the selling owner keeps the relationships. And the judgment... Gartner's agent analyst says plainly that "we are nowhere near the point where we can just throw a lot of data at an AI agent and trust its decision," and CPA-market reporting this year is full of firms whose AI status is best described as mid-rollback, after deployments that did not deliver.
Minerva's co-founder, announcing that 5-to-70 rebuild, added in the next breath that humans are not going anywhere... accountability, empathy and context stay human, staff move up to advisory.
The honest number
And while everyone argues about zero, the more interesting number is one. The one-person firm has existed for a century... it's called a sole practitioner. What AI breaks is the ceiling. A solo accountant could serve maybe 50 clients. A solo agent, 100 properties. Those limits are what's going, and the most leveraged professionals of the next decade will be sole practitioners and small partnerships running claim-two ratios... every remaining human a licence, a signature, a relationship or a judgment call, and everything else in the building software.
Back to the title, then. The one-person company can be real when the product is software. In a firm, the people don't disappear... the ratio explodes, and that's where the money is. And the zero-human firm needs a regulator to let an agent sign. If that ever happens, everything I've written here dies. I don't expect it soon.
Two things to watch on the way. Revenue per licence-holder replacing headcount as the number buyers care about. And the first "self-running" firm to blow up publicly... because that's the day regulators sit down and rewrite what a human actually has to sign.
And since Eilla is exactly this kind of business, I'll use us as the last example. We run a number of live M&A mandates with a fraction of the team a traditional advisory would carry... we are a claim-two business through and through. And I can report the walls are real. The production automates wonderfully. The mandates are still won over long dinners, and a founder still hands their life's work to a person, not a system. If you're buying or selling a firm, let’s talk.
Do you buy the verdicts... or do you think Rational's bet lands sooner than I do? Hit reply, I read everything, and the sharpest replies shape the next edition.
See you next time.