The One-Person Institution

What can one person with AI agents really run, and what does it take to keep it running?

The slogans count employees and money: the one-person company, the one-person unicorn. This book counts something else. It follows a single operation, the author's own, through the autumn of 2026, and tells it from the records the operation keeps: dated version histories, logs that agents write as they work, a tracker that holds every piece of work, a memory of plain-text notes, and the rule files every agent reads before it starts.

Two capabilities carry the book: software development and book production. In the middle of October 2026, AI agents drafted the manuscripts of four nonfiction books within two calendar days, under one person's rules and decisions. The book shows what that took, station by station: a source register for every chapter, reviews by models from two companies, written answers to the reviewers' findings, and a list of what broke. It follows agents building software, too, with a reviewer from a second company that may read the code and may not change it.

The book does not stop at the capability. Its second part counts the burden that never shows up on an invoice: the small repairs, the queue of things only the owner can unblock, failures that made no sound for days, and the dependence on other people's companies, from usage limits that stopped agents in the middle of the night to the terms of distributors and stores. The third part asks what holds such an operation together. The rules, the tracker and the memory turn out to be the institution, and the records name one person at every gate, with no deputy on record.

Every chapter opens with a scene and gives the strongest objection a fair hearing, conceding what is true: that a company with staff would have caught some failures sooner, that "one person" hides a great deal of other people's labor, that four books in a month add to a flood nobody asked for. Composite scenes from small firms across Europe, from a publishing house in Leipzig to a documentation office in Ulm, are labeled as such.

The last chapter turns the evidence into a decision guide: six questions about volume, error, reversibility, responsibility, upkeep and dependencies, and four moves: build, buy, hire, or leave it alone. Two appendices give a one-page checklist and a worksheet for the running burden.

Written for European readers who run or are planning a small operation, the book needs no technical background. Every figure about the operation carries the date of its record; where the records are silent, as they are about the owner's hours, the book says so and invents nothing. EU laws such as the Data Act and the Cyber Resilience Act are cited from their official texts. Facts about law, products and companies are stated as of a single date, given in the book. Nothing in the book is legal, tax or financial advice.

The book is open about how it was made: AI agents drafted every chapter, models from two companies reviewed them in many rounds, and the author reviewed, edited and approved the text.