Fresh Horses · Part three · Personal AI
The Charge Is Yours
If a hard decision is counted against a person and not against a job, then the most valuable thing any agent holds is its model of one person. That model is built mostly outside work and compounds only if it travels. That is why personal AI will win, and why every business needs a policy that says: bring your own.
Hayek’s point was that this knowledge is never gathered in one place. It lives, scattered and unwritten, with the people on the spot. A system works when it lets that knowledge be used where it lives, and fails when it tries to collect it at the center.
01The polite stranger
The new company hands you a laptop with its own AI already installed. It is capable and courteous, and it knows nothing about you. By twenty past ten it has asked you three questions the AI you left behind would never have asked, because that one knew you settle that kind of thing in your sleep. It has also held back the one thing you would have wanted at eight, because it had no way to know that eight is when you are sharpest.
Nothing is wrong with it. It is sorting by the job. You are not the job.
In Two Clocks I argued that a charge, a decision that carries real surprise, is never a property of the decision. Surprise is the distance between what happened and what one particular person expected. The same refund dispute is a charge for a manager in her second week and a walk for the owner who has handled four hundred of them. A role does not have a charge count. A person in a role does.
I have argued for personal AI over enterprise AI for a long time, mostly on grounds of trust and of method. People do not adopt what watches them. A hiring manager should not confiscate the wings of the person she hired for flying. Both arguments still hold. This one is more basic, and I think it settles the matter.
02Knowledge of time and place
In 1945 Friedrich Hayek published a short essay in the American Economic Review that became one of the most influential arguments in economics. The problem of running an economy, he wrote, is not a problem of calculation. If all the relevant facts were sitting in one place, any competent planner could do the arithmetic. The problem is that the facts that matter most are never in one place. They are local and fleeting: his examples were the shipper who makes a living filling tramp steamers that would otherwise sail half empty, and the estate agent whose whole knowledge is of temporary opportunities. That knowledge is scattered, mostly unwritten, and held by the person on the spot. A system succeeds by letting it be used where it lives.
Now look at what a good agent needs in order to sort your day. It needs both clocks. The spoil clock belongs to the situation: when the truck leaves, when the table books elsewhere, when the window closes. The horse clock belongs to you: when you are fresh, and which decisions will surprise you at all. And the horse clock is the purest case of Hayek’s knowledge there is. It is local knowledge about exactly one person.
03Three reasons the model of you cannot be the company’s
It is built mostly outside the job. Only Actuals retrain Prediction, so a model of what you expect can only be learned from the record of what you have actually done. Most of that record does not happen at work. Whether you are fresh at nine on Thursday depends on how you slept, on the child who was up at three, on the conversation you had in the car. An employer does not see those things and should not. An agent the company owns is sorting your charges with half of your horse clock missing.
It compounds only if it travels. A model of you gets better every week it watches you work, and its value is the sum of everything it has learned. An enterprise agent starts that sum at zero with every new hire and leaves it behind at every departure. The person’s experience walks out the door; the model of that experience stays in a system nobody will use again. Your charge count goes back up to a beginner’s on the first Monday, partly because what you know moved and what your agent knew about you did not.
You will only let it see your whole life if it works for you. This is where the old trust argument and the new one turn out to be the same argument. The only agent that can sort your charges well is one that is allowed to see the parts of your life that set your horse clock. The only agent anyone will allow to see those parts is one that answers to them. The surveillance problem and the sorting problem are one problem.
Enterprise AI is aiming at a quantity that does not exist: the charge count of a role. Personal AI aims at the real one.
04Each side brings its own clock
Once you see the two clocks, the shape of a sound policy is almost obvious. The business owns the spoil clock, because the business owns the situation. The person owns the horse clock, because the person is the horse. The sort happens at the seam between them.
The business brings
The spoil clock- Deadlines, windows and service promises, written where any agent can read them
- The systems of record and the permissions into them
- The work product, which stays with the business
- A capable default agent for anyone who arrives without one
The person brings
The horse clock- An agent that knows what surprises them and when they are fresh
- Their method, refined over years and jobs
- The record of their own decisions, which is theirs to keep
- The judgment to make the charges that reach them
Written as policy, it comes to five lines.
Bring it.
Employees may connect their own agent to the company’s systems through scoped, revocable permissions. The company does not ask them to set it down at the door.
Publish the spoil clocks.
Every deadline, booking window, delivery time and service promise lives somewhere an agent can read it. This is the company’s half of the sort, and most companies have never written it down.
The work stays; the model of you leaves.
What is done for the business lands in the business’s systems and belongs to the business. What an agent has learned about its person belongs to the person. Put that sentence in the offer letter.
Provide a floor.
A good default agent for anyone who has none, so that bringing your own is an advantage rather than a price of admission.
Plan the exit on day one.
When someone leaves, access is revoked, records stay, and the person’s agent goes with them, intact. Everyone knows this in advance, which is what makes it safe to bring the agent at all.
The earlier pieces in this series, especially The Dependency Boundary and The Method Floor, work through the mechanics of lines three and four. What the two clocks add is the reason the lines fall where they do.
05A first Monday in 2029
A general manager starts at a new restaurant. She arrives with her own agent, the one that has worked beside her for six years across two previous jobs. The restaurant’s systems hand it the spoil clocks: the delivery windows, the private-room calendar, the payroll cutoff. Her agent already has the horse clock.
Her first morning brings seven charges, not fourteen. The restaurant is new to her, so some of it is genuinely hard. But the refund disputes, the vendor substitutions and the scheduling fights arrive the way they always have for her, as walks, because the thing that knew they were walks came with her.
06Five predictions
Hiring will ask what your agent knows about you.
The way a designer is asked for a portfolio, experienced candidates will be asked how their agent works with them, because an experienced person who brings their own starts near their veteran count instead of a beginner’s.
The next fight over non-competes will be about agents.
Employers will try to claim the model of a person as company property, written into contracts the way restrictive covenants once were. That fight will be lost, slowly, because a model of a person will come to be treated as that person’s data, not as the employer’s trade secret.
Enterprise AI vendors will become the stove.
The durable business will be the spoil-clock side: systems of record, permissions, deadlines and service levels made legible to any agent that arrives. Selling every employee a company brain will look the way selling every employee a company phone looks now.
Legible companies will win the best people.
Businesses that publish clean spoil clocks will be the easiest places for a capable person with a capable agent to do excellent work from the first morning. Talent will move toward them, and it will be able to say exactly why.
Loyalty will have to be earned by the work.
When a person’s walks move with them, the cost of changing jobs falls. Companies will no longer hold people by the switching cost they used to impose. They will hold them by being a good place to spend a charge.
07Where I am in this
I tell my students to let their first employer be their first customer. A customer does not own a vendor’s tools. A contractor brings the truck; the client provides the job site. If your first employer is your first customer, you bring your own AI for the same reason, and you leave with it when the engagement ends.
A real share of the work at Charleston AI is helping people who hold ordinary salaried jobs teach a personal AI the work they do. I used to describe that as building leverage. I now think it is more specific than that. They are building a horse clock that belongs to them.
08The ledger
Claim A · Sorting a person’s charges well requires a model of that person that is built mostly outside work.
- Already true
- In the Reality Equation surprise is defined only for an Actualizer, and Expectation is learned only from the record. Anyone who has worked through a sleepless week knows that the hardest variable in their workday was not at work.
- What has to happen
- Personal agents have to be permitted to meet company systems through scoped access, so the model of the person and the spoil clocks of the business can be in the same sort without either side surrendering the other.
- Where I am probably wrong
- If what surprises a person at work turns out to be learnable from work alone, in a few weeks, then an enterprise agent can sort almost as well, and this argument shrinks to the narrower one about portability.
Claim B · Personal AI will dominate because the model of a person compounds only if it travels.
- Already true
- People already bring their own AI to work in large numbers whether or not they are allowed to, which tells you where they think the compounding lives.
- What has to happen
- The ownership question has to be settled in the person’s favor, by contract first and by law later, so that bringing an agent does not mean donating it.
- Where I am probably wrong
- If employers win the ownership fight, or if regulated industries require every agent that touches their data to be company-controlled, personal AI will dominate only among the self-employed and small firms, and large employers will run on polite strangers for a long time.
09Back to the polite stranger
It is twenty past ten on your first Monday, and the company’s AI has just asked you its third unnecessary question. It is not a bad system. It is a planner at the center, doing careful arithmetic on facts it will never have.
The facts it is missing are yours. They always were. Bring them.
Background
- F. A. Hayek, “The Use of Knowledge in Society,” American Economic Review (1945). Econlib
- John Rector, Fresh Horses and Two Clocks, parts one and two.
- John Rector, Bring Your Own AI, the book.
- John Rector, Bring Your Own AI, the field note on trust.
- John Rector, Do Not Make Them Use Your AI.
- John Rector, The Zero-Attention Economy, the book.
Related
Now a bookThis argument is set out whole, with a complete model policy you can adopt, in Bring Your Own AI: The Policy Every Business Needs (October 30, 2026).
4 thoughts on “The Charge Is Yours”