The one-person company
A book about the AI-first company that comes with one installed.
Every role becomes a repo. Every procedure becomes a playbook. AI runs it on your machine, and you make the calls that matter.
7 parts. 31 chapters. One payment of $17.
Get the book, $17Instant access. 14-day refund. No subscription, ever.
The operator's manual
HeadcountZero
How one person runs an entire company on AI
7 parts · 31 chapters
Get it now for $17
Secure checkout · Payments by Stripe · USD
14-day refund. No subscription, ever.
Nobody gets replaced. Nobody was hired.
01 / The trap
Not because people are the problem. Because of what happens to everything you teach them.
This pattern is ours, not a borrowed case study. Years of hiring. Dozens of people, trained one procedure at a time. Every one got good. Every one eventually left, and the company forgot what they had learned.
Not the files. The reasoning: why we stopped doing it the other way, which supplier fails in August, what the number has to be before you kill it.
Better hiring, better onboarding, a wiki: we tried it all. What worked was structural. Stop keeping the company in people's heads. Every role in a repo, every procedure in a playbook a stranger could run, the reasoning written down. An AI runs it. You keep the judgment.
An employee quits and takes the reasoning along. A repo does not. That is the difference between a company that resets every eighteen months and one that compounds.
A plan is not a company.
Most AI advice hands you a to-do list. This hands you the architecture, then the thing itself.
02 / A test you can run right now
If the answer is yes, you are not AI-first. You bolted AI onto work that is still employee-shaped.
Retrofitting: a person owns the procedure and asks AI for help inside it. AI-first: the procedure lives in a file, the AI runs it end to end, and the person shows up at the decision.
Retrofitting feels productive, but when the person leaves, the procedure leaves too. You bought a faster typist for a company that still cannot remember.
Chapter 7 runs the test on four common setups and tells you which parts of yours survive.
03 / The mechanism
Four moving parts. The book builds them one at a time.
One orchestrator that knows who does what and where everything lives. It routes work, checks results, and stays thin so it never becomes the bottleneck.
Operations, marketing, finance: each a folder with its craft, memory, playbooks and a log of why. If a fresh session cannot act from it, the folder is broken, not the AI.
Every procedure in four blocks: command, verify, recover, gotcha. It runs the same on a bad day as on a good one.
No autonomy on day one. A role climbs four rungs: advise, propose, apply within a cap, apply. You grant each rung, and you can take it back.
The AI does the two hundred hours. You make the twenty decisions. AI-first, not AI-only.
04 / Where this is going
What people actually said and did, with sources. Nothing rounded up.
Sam Altman, talking with Alexis Ohanian (Fortune, 4 February 2024):
"In my little group chat with my tech CEO friends there's this betting pool for the first year that there is a one-person billion-dollar company. Which would have been unimaginable without AI and now will happen."
At Code with Claude in May 2025, Anthropic's Dario Amodei put the first billion-dollar solo company at 70 to 80 percent likely by the end of 2026.
Close, they think. Neither says it has happened.
WhatsApp, 2015: ~900 million users, ~50 engineers.
Gumroad, 2023: $20.7M revenue, $8.9M net profit, zero full-time employees (Sahil Lavingia). Pieter Levels, 2025: "I don't write code anymore."
A telehealth company, profiled by the New York Times in April 2026: $401M in year one with two employees.
Then the other half: an FDA warning letter, a class action, fabricated doctor personas, and a Times editor's note.
Chapter 3 tells both halves. The second is why a human stays at the gate.
No one-person billion-dollar company exists yet. But the ceiling is rising faster than anyone planned. This is the manual for the people racing at it.
05 / Who is behind this
No biography to sell you. The operating business stays unnamed on purpose. What you see is the architecture, working.
An optional upgrade after purchase, not part of the $17. Shown so nothing later is a surprise.
Built running a real company, rebuilt clean for you. No client facts, nothing borrowed.
06 / What's inside
Each chapter proves one thing, then tells you what to do about it today.
Part I
Part II
Part III
Part IV
Part V
Part VI
Part VII
The install step
A concrete action in your own folder, doable that day. By the end of Part VI: a front door, one working role, one procedure that runs itself.
07 / The line, drawn honestly
Column three is not softened. The first two are only credible because of it.
08 / What else you could do
Including doing nothing, which is a real option and sometimes the right one.
| Option | What it costs | What you end up with |
|---|---|---|
| Do nothing | $0 | The same company, one year older. Fine if it is working. |
| Hire a person | $60,000+ / year | Real capacity, and a copy of your procedures stored in a head that can leave. |
| Agency retainer | $3,000+ / month | Output while you pay. The knowledge lives at their office, not yours. |
| An AI course or prompt pack | $50 to $2,000 | Techniques. Nothing that keeps running after you close the tab. |
| This book | $17, once | The architecture, written down, plus one install step per chapter and a company that remembers what it learned. |
09 / Why $17
The usual move: a cheap book that exists to sell you a subscription. You have bought that $7 doorway before.
There is no continuity here, and there will not be. No community, no monthly fee, no renewal.
$17 is what a real book costs: low enough that a wrong guess costs little, high enough to keep writing it. The runnable system is offered once, after you buy. Decline it and you never hear about it again.
Claude Code is Anthropic's product, with its own subscription (from about $20 a month at the time of writing). We earn nothing from it, and you can read the book first.
10 / The offer
HeadcountZero, the book
$17
One payment. Not a trial, not a first month.
The one-payment pledge, in plain words.
$17, charged once. Every add-on and upgrade is a single payment too, always at the same price, and none is needed for the whole book. No subscription, no trial that converts, no renewal.
11 / If you are not buying today
Roughly weekly: what we changed and why, what broke, and which AI-first claims fell apart when we checked the source.
No sequence, no pitch dressed as a lesson. One click unsubscribes. We never sell the list.
Free. Nothing is charged, and subscribing does not buy the book.
You are on the list.
The next note goes out with everyone else's. Nothing else happens in your inbox in the meantime.
12 / Questions
No. You open a terminal and type instructions in English. The architecture is folders with words in them, and the technical parts come as complete examples you copy and adjust (chapter 23).
The real skill is writing a procedure down before handing it over. That is not a technical one.
Anthropic's tool that lets an AI work in your files and run things on your machine. Not ours, not included: its own subscription, from about $20 a month at the time of writing. We earn nothing from it, and you can read the whole book before subscribing.
No, on purpose. Prompts go stale fastest and teach the least. The book is about structure: how roles, procedures, autonomy and decisions get written down. That has outlived several model generations. Last year's prompts have not.
No, and the book argues against it. It is for the operator who never scales payroll in the first place.
Chapter 29 covers public AI layoffs that were reversed. Chapter 30 covers the realistic outcome: one operator plus a few specialists.
Arguably more useful. Five people with everything in five heads are as fragile as fifty. Written-down roles let people work on the company instead of being it. Chapter 30 is written for teams.
The architecture already built: front door, three seeded role brains plus a template, playbooks, registry, mandate ladder. 35+ files that run in your terminal.
Not required. Offered once after you buy, always at the same price. The book is complete without it.
Checkout charges $17. Optional add-ons may appear as checkboxes, never pre-ticked. After the book is in your hands, we offer the Company OS ($97) and a wider edition of it.
Each is a single payment, none is required, none holds back anything in the book, and prices stay the same later. Never a subscription.
14 days, full refund, unconditional. Email us and it is usually done within 48 hours. We may ask what did not work. Ignore the question if you like: the refund is yours either way.
No. No community, no monthly fee, no renewal, now or later. Your purchase never turns into a recurring charge.
Naming the operating business would drag its customers into a sales page. What is sold is the architecture, rebuilt clean. Judge it on its contents: the reasoning is in the book, the files are in the OS.
It will, and the architecture assumes it: the mandate ladder, the verify step in every playbook, and a hard stop on anything that commits money or your name.
Chapter 8 names the failure modes (forgetting, stale instructions, thin docs) and the structural fix for each.
Part VI is a 90-day path for someone with a company to run. Day zero is an evening. The first working role takes a couple of weeks. Then a weekly run, capture, write-down cadence.
Do not start a second role until the first is reliable. That is the most common way this fails.
Whoever proves the prediction right will not do it with prompts. They will do it with a company written down well enough for a machine to run, and the judgment to keep the twenty calls that matter.
One payment. Instant access. 14 days to change your mind. No subscription, ever.