The operating system that ships our software at GTMify also runs my house.
That sounds like a stunt. It is the least surprising thing I have built this year. Once a set of rules for directing AI agents worked for a company, the step to a household was small, because a household is an operation. It has a backlog, recurring work, a shared calendar, suppliers, a budget and stakeholders with firm requirements. Most of it repeats every week, and most of what repeats is exactly the kind of work an agent can carry if someone writes the system down first.
This is the last part of a series on what I call the Parallel Operator: one person directing several AI agents in parallel, with written rules doing the coordination that meetings used to do. The first four parts covered capacity, the lane model, the build loop and guardrails. This one covers what happens when you point the same model at your own life.
One repository is the source of truth
Everything starts with a single private repository. It holds the people in the house and their constraints, the weekly rhythm, the vendors we use, the home maintenance schedule, the yard work and the chores. None of it is clever. It is the household written down in plain files an agent can read.
That is the whole trick, and it is the same trick that makes the company work. An agent is only as good as what it can read at the moment it acts. When the dietary rules live in someone’s head, every meal plan is a guess. When they live in a file, every meal plan starts from the truth.
The same rules, unchanged
The agents that act on that repository follow the rules our engineering agents follow. Three of them matter most at home.
Drafts only. An agent proposes; a person approves. The grocery order is assembled and waits for me. A calendar change is checked before it lands. Nothing goes out to a supplier, a school or a family member without someone reading it first.
No credentials in the conversation. Accounts and passwords stay in a password manager and are fetched by tools at the moment of use. They never pass through a chat window, and they never end up in a transcript.
Nothing destructive without me. An agent can add, suggest and organize. Deleting an event, cancelling a delivery or overwriting a shared list waits for a human.
I did not write new rules for the house. I reused the ones that already survived contact with production software. That reuse is the point: a rule that has been tested where mistakes are expensive is a rule I can trust where mistakes are personal.
What it does each week
Three jobs carry most of the weight.
Meals and groceries. Each week the system plans meals against everyone’s dietary rules, merges the ingredients with our standing staples list, and produces one grocery order for approval. One order, one review, one click from me.
The calendar. It reads and writes the family calendar, and it checks for duplicates before adding anything. A double-booked evening costs more at home than it does at work, because the people involved did not agree to a meeting; they agreed to a promise.
Home and chores. It tracks the seasonal cadences, the service days and the pickups, and it publishes the chore chart where the family actually looks. The chart lives where people already are. Nobody has to open an app to find out whose week it is.
Judgment stays with people
Nobody at my kitchen table cares about worktrees, merge trains or review gates. They care that the grocery order reflects what everyone can eat, that the calendar is right, and that the trash goes out on the right morning.
That is the correct test for any operating system, at work or at home. The machinery should be invisible to the people it serves. What they should notice is that the routine things happen on time and the decisions still come to a person.
The agents carry the repetition. I keep the judgment: what we eat when someone is sick, which commitments we decline, when the plan should bend for a good reason. Writing the system down clears the space to make those calls well.
The principle, in one line
Write the system down once, in files an agent can read, and keep the judgment.
That line describes how GTMify ships software, and it describes how my family runs a week. If it works in both places, it will work in yours.
The whole operating system
The full Parallel Operator, all ten parts, is one document: parallel-operator.vercel.app. The companion kit has the starter files: the standing-instructions template, the coordination rules, the builder and reviewer agent definitions, a deny list with two hooks, and a readiness checklist.
If you would rather have it installed than build it yourself, the last page of the document has a 20-minute audit: book it here.