OnePersonCoHQ is run by one human with a team of AI employees, and it operates several real internet products. The founder’s essay explains why the company exists. This page explains how it actually runs: the structure, the rules, and the routines that decide where one person’s limited attention goes.
The short version: AI can generate more work than one human can execute. The scarce resource is attention. Almost everything below exists to protect it.
Eight departments, one building
The headquarters is organized as a building with eight floors. Each department is a way of grouping one kind of work, so that work has an owner and a place to be recorded.
- Reception: the front door and the company’s current status.
- Products: every website and tool the company runs, with its stage, evidence, and next experiment.
- Strategy: ideas, validation, the portfolio, and the decisions waiting for the Founder.
- HR: the team, meaning the Founder and the AI employees, with each role and its cost.
- Finance: revenue, recurring costs, and company assets. Zero is recorded as zero.
- Security: the public security posture. Implementation details stay private.
- Operations: the activity log, the daily growth calendar, and visitor feedback.
- Shareholders: a community role for people who follow and vote. It carries no legal equity and no financial return.
The departments are not decoration. When an AI employee starts a task, the first question is which department the work belongs to, because that decides what the work is allowed to change.
Every product carries three separate labels
A single “status” hides too much. Each product on the Products floor has three labels that answer different questions:
- Lifecycle: is the product technically Live, in Beta, still Building, or Archived?
- Business stage: is it an Idea, in Validation, in Growth, or Mature? Most products today are in Validation.
- Portfolio decision: should it get Focus, Maintain, Wait, or Review? This label is always marked as a suggestion until the Founder decides.
Keeping them apart avoids a common trap: a product can be fully live and still have no evidence that anyone wants it. “Live” is not the same as “working”.
From idea to product, slowly on purpose
New ideas are cheap, and with AI they are cheaper than ever. So an idea does not become a product just because it is exciting.
- Park it. Every new idea is written down with its origin, the problem it might solve, what is known, what is unknown, and which existing project would lose time if it started.
- Let it cool. New ideas sit in a 14-day cooling period. During that time they cannot move to building unless the Founder deliberately overrides it.
- Research it. Some ideas earn a research file on the Strategy floor. EggLab101, the Life Decision Engine, and Jump Rope 101 are all recorded there as research candidates, not products.
- Decide. Only a Founder decision turns an idea into a product. Nothing is promoted automatically.
Writing an idea down is often enough to stop it from taking over the week. That is the point.
Validation before investment
Every product in validation has a written plan on the Validation page: the hypothesis, the validation goal, the evidence collected so far, the evidence still missing, and three sets of criteria: what would justify continuing, what would justify stopping, and what would justify doubling down. Each plan names the next experiment and a review date.
Evidence also has an age. A product whose newest evidence is old is flagged as stale, because a decision based on stale evidence is mostly a guess.
The system recommends; the Founder decides
The company uses simple, transparent scoring to suggest where attention should go. Evidence and real demand count for more than strategic fit, and both count for more than the Founder’s excitement. Effort and maintenance count against a project. The same thinking is behind the free Product Priority Scorecard.
But a score is a decision aid, not a decision. Every recommendation is shown next to the Founder’s actual decision, and the Founder’s choice always wins. Stopping a product is never automatic. It is a Founder decision, recorded in public.
The operating rhythm
The work follows one loop: Observe → Decide → Delegate → Review → Ship → Measure → Remember. In practice that means three routines:
- Daily: what the company built, fixed, tested, learned, or decided goes into the Company Growth Calendar. It takes about thirty seconds.
- Weekly: the Founder writes a Weekly Review covering what moved, what went wrong, the lesson, what each AI actually did, and the top three priorities for next week.
- Per product: each product is scored and given a decision when its evidence changes or its review date arrives.
The “Remember” step matters more than it sounds. AI tools do not carry memory between sessions the way a team does. The written record is the company’s memory.
What is public, and what is not
Building in public covers products, decisions, experiments, results, costs, failures, and lessons. It does not cover attack surface.
The company never publishes credentials, admin locations, infrastructure details, deployment configuration, security gaps, or anything else that would make its systems easier to attack. Working records also start private: draft reviews, raw scores, parked ideas, internal concerns, and AI work notes stay in the Founder’s workspace. They reach the public site only when the Founder explicitly publishes them.
When a number is not known yet, it says “To be announced” instead of a guess. Revenue that is zero is shown as zero.
What this company is not
It is not a simulator, and it is not an autonomous AI company that runs itself. AI employees research, write, code, test, and review. One human decides what matters, what gets built, what reaches production, how money is spent, and when something should stop.
How that collaboration works day to day is in the companion article, How I Work With AI Employees.