Headcount Zero: an open book on running a company with AI agents as staff

Fourteen chapters, published as Markdown, in a book that is itself the product

Headcount Zero is a book by Anthony David Adams about building and running companies with AI agents as the entire workforce. The premise is stated in one line in the README: no employees, no payroll, no HR, one founder, an org chart full of AI agents, and an open source platform called Paperclip to orchestrate it all.

The book is open source and readable in the repository, and the table of contents is a table of Markdown files. The top-level entries are four directories, PART-1-WHY, PART-2-WHAT, PART-3-HOW and PART-4-WHAT-IF, plus the README. There is no code, no build system and no binary. The content is the artefact.

That structure is a deliberate choice with a practical consequence. Because the chapters are plain Markdown in a repository, a reader can read them in a browser, clone them, translate them, feed them to a model, or submit corrections, and the contributing section invites exactly that: pull requests for edits and fixes, issues for ideas and feedback, and a specific issue template for blurbs.

The four-part structure is also a structure of argument rather than a list of tips. Part 1, Why, holds the introduction plus three chapters: Nobody's Being Honest on the gap between AI hype and reality, Smash the Loom on a brief history of automation panic, and One Person, One Billion on the one-person unicorn thesis. Part 2, What, covers what AI agents actually are, the Paperclip platform, and the economics of zero employees. Part 3, How, is the operational part, with building your first AI-run company, designing your AI org chart, and governance and oversight. Part 4, What If, looks forward to scaling to multiple companies, the human element, a first two days plan, and an afterword.

The README also notes a formatted version on Amazon when it launches, so the open repository is the primary edition and the commercial one is a formatted edition of the same text.

The thesis is argued in the first three chapters, not assumed

The order of Part 1 is the thing that makes this more than a manual. A book that opened with a tool would be a manual with a manifesto attached. This one opens with the case against the thesis, and only then defends it.

Nobody's Being Honest is described as covering the gap between AI hype and reality. Having that as chapter 1 rather than as a passing caveat is a statement about the book's method, and it means the later chapters can be read as an argument rather than as a sales pitch.

Smash the Loom puts the current moment in a longer historical frame, described as a brief history of automation panic. That is a different rhetorical move from the one most books make. Instead of saying this time is different because of the technology, it says previous waves of automation produced their own panics, which is a claim the reader can accept or reject independently of whether the technology is good.

One Person, One Billion is the thesis chapter, the one-person unicorn thesis. The name is doing the work of an argument: the claim is that one person with agents can be worth a billion dollars, which reframes the unit of the business from the employee to the founder. The rest of the book is then the machinery for that claim.

The book also comes with a stated voice about hype. The opening question in the README is what if you never had to hire anyone, and the framing is of a founder who has run out of room to hire and started building a different kind of company. The chapter order suggests the author expects a reader who is sceptical, and is willing to spend three chapters on the case before the how.

Part 2 turns the idea into a platform and a spreadsheet

Part 2 is the transition from argument to implementation, and it is short: three chapters, one on what AI agents actually are, one on the platform, one on the economics.

The first of those, Not a Chatbot, is the chapter that does the definitional work. The title alone sets the boundary the book is going to work inside, and it is worth being careful about how much the README actually tells us. It does not summarise the chapter's argument, so what can be said with confidence is that the book treats the agent as something other than a chat interface, and that this distinction carries weight for the rest of the argument, since the whole premise of replacing a workforce rests on agents doing work rather than answering questions.

The Paperclip chapter is described as the platform that makes it possible, and the platform has its own site at paperclip.ing. This is the practical dependency in the book. The repository here does not contain Paperclip, so a reader following Part 3 is following instructions that point somewhere else, and anyone assessing the book has to assess the platform as a separate thing.

Do the Math is the economics chapter, described as the economics of zero employees. Placing it in Part 2 rather than at the end is a choice worth noting, because it means the money is part of what the thing is rather than a caveat about whether it works. The chapter names no figures in the table of contents, so what the numbers are is not visible from the outside, only that the book treats them as a load-bearing part of the argument rather than an appendix.

The links section points at Paperclip and at the Singularity Playground, described as being at playground.earthpilot.co/singularity, which suggests a place to try the setup before committing to it.

Part 3 is the operational core, and it includes a kill switch

Three chapters make up the how, and their order goes from doing something to governing it.

Open Your Terminal is the first hands-on chapter, described as building your first AI-run company. The title is a bit of a joke and also an instruction, since the build starts in a terminal rather than in a web console, which is a reasonable signal about the intended audience for a book that assumes you can run software.

Who Does What is about designing your AI org chart. This is the chapter where the metaphor in the book's premise gets taken seriously. If agents are the workforce, then someone has to decide what roles exist, what each role is for and which agent fills it, and that is an organisational design problem expressed in a technical vocabulary. The book's framing of a human as one node in a hierarchy of agents, which is what the README's description of one founder and an org chart full of AI agents implies, is where the interesting thinking lives.

Kill Switch is the chapter that most distinguishes this book from a technology enthusiasm piece, and its description, governance and oversight, says plainly that the third chapter of the how is about stopping things. For a book whose thesis is that a single person can run a company with no employees, the governance question is the sharpest one it could ask, because the person doing the overseeing is also the only accountable human in the structure.

The governance material is not described in the table of contents, so how thorough it is cannot be judged from the outside. But its presence in the table of contents, as a full chapter in the operational section rather than in a later reflective one, tells you the author treats it as a build step.

Part 4 asks what happens when it works, and when it does not

The last four chapters move from instruction to consequence, and their titles are the argument.

Run Ten at Once is about scaling to multiple companies, which is the logical next step if one founder with agents is a working model. You Are Not Optional is about the human element, and the title is a correction rather than a reassurance. The README frames the book as humans and AI operating as partners, human operators bringing strategic judgment and AI operators bringing execution, and a chapter titled You Are Not Optional pushes back against the reading where the founder becomes a passenger in their own company.

48 Hours is about your first two days, a practical plan rather than a theory, and it is the chapter most likely to be useful to a reader who is not going to run ten companies and just wants to try one.

The final chapter is an afterword titled Don't Hire Anyone, which is a deliberate restatement of the thesis in the imperative mood, and the placement after the practical material gives it a different weight from the same claim made in the introduction.

What the four chapters together do is refuse the fantasy ending. A book about replacing a workforce could easily end with a vision of a founder running a hundred companies alone. This one ends by insisting the human is still necessary, then gives you a two-day plan, then restates the original demand. That is a more useful structure for a reader who has to decide whether to try the thing.

Licence, sponsor and the absence of a technical artefact

Two practical facts belong in any assessment of this repository.

The first is the licence. The work is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International, and the README states the terms plainly: you are free to share and adapt the material for non-commercial purposes, with attribution.

That is a licence with two real consequences. The non-commercial clause rules out using the text in a paid course, a paid book or a consultancy deliverable without permission, and the share-alike clause means that if you adapt the chapters and distribute the result, you owe the same licence on the adaptation. For a book about building a business, that is a slightly awkward fit, and anyone planning to build on the material commercially needs to ask the author rather than assume.

The second fact is what the repository does not contain. The primary language is recorded as unknown, there are no releases, and the only source material is Markdown chapters and a README. Whatever Paperclip is, it is not here. The last push was on 2026-04-14 and the repository is not archived, so the text is being revised, but the revision is to prose rather than to code.

The book is provided as a public utility by Earth Pilot, described as mission support for spaceship Earth, with a logo and an organisation link in the README. The praise section is an open placeholder, with a note that a blurb could be there and instructions for submitting one, and a thank you to a sponsor at the end.

There is also a commercial edition coming. The README says to read the book here, share it, or get the formatted version on Amazon when it launches, and the anchor for that link is currently empty, so the launch has not happened yet.

Editorial conclusion

Headcount Zero is worth reading if you are a founder considering running an organisation where agents do the work and you want the argument laid out rather than sold, because the book is explicit that the gap between AI hype and reality is the subject of its second chapter and it dedicates a full chapter to governance and a kill switch. It is the wrong tool if you want a technical guide to a specific framework, since the platform chapter points at Paperclip and the repository here contains no code. Before you act on it, verify three things yourself: that the economic chapter holds up against your own numbers, since the zero-employee economics depend on costs the book models rather than costs it measures, that the governance chapter matches the oversight your own context requires, since a kill switch designed for a solo founder is a different thing from a compliance regime, and that the licence permits what you intend, because Creative Commons Attribution-NonCommercial-ShareAlike 4.0 allows sharing and adaptation for non-commercial purposes with attribution but imposes share-alike on what you derive from it and rules out commercial use.

Frequently asked questions

Official sources

1. AnthonyDavidAdams/zero-employee-company-book on GitHub
2. Issues
3. README