If you have put artificial intelligence (AI) to work in the last two years, somebody promised you an hour back. The draft that used to take a morning takes a minute. The research that used to take an afternoon is waiting in a tab before you have finished the coffee, and the notes from the meeting are written before you are back at your desk. By every measure a vendor would put on a slide, the hour was saved, and I would like to know where it went, because I have been looking for mine. The evening still arrives with the screen open. The list is no shorter than it was, and there is a new habit you did not have before, which is checking what the machine did while you were not looking.
I spent thirty years on a payroll, the last stretch of them responsible for AI governance at a global enterprise of more than 120,000 people, and I retired in July of this year with a plan. The plan was to live a more analog life than I had managed since I was a boy on my grandparents' land, where a day was measured in what you had fished and fixed and carried, and nothing beeped. I had the tools for it, or I thought I did. I had written a book with a small roster of agents while holding down that job, and I had every reason to believe the same agents would carry the administrative weight of the business I was starting so that I could be outdoors while they worked. What I found in the first weeks was that I was no good at it. The quiet arrived and I filled it with the screen, and what I was doing on the screen, if I am honest, was checking.
So I will state the problem the way I have come to see it, from the side of the person who has it. A person who puts AI to work is promised time and receives output. The tools multiply what one person can do, and the person, being a person, either finds more to do with the multiple or stands beside the machine and reads everything it produced, because nothing in the arrangement has earned the right to run unread. What the person does about it today is buy a better tool, or read something about productivity, or decide that the screen is simply where life happens now. None of it works, because the hour the machine saved did not disappear. It was spent on watching the machine.
The word for what the machine gives you is the one every vendor leads with, and I will use it exactly once, in the sentence this essay and the business I run both rest on: trust is what turns leverage into time. Take the first word away and the middle one produces supervision. A more capable machine that you cannot trust is a machine you have to watch more closely, and the more it can do while you are gone, the less you can afford to be gone. That is the arithmetic that was running in my first weeks, and it runs in every company I have looked at closely. The tools got better, and the watching got longer.
There is a test for whether the time is real, and I did not want to hear it when it was first put to me. The analog time is genuine only if it is unmonitored, and checking the agent's output from the trailhead means the machine lengthened the leash rather than releasing you. I knew the shape of that sentence before anyone said it about software. I have been sober for eighteen years, and one of the things sobriety teaches you is to distrust your own report of how much control you have. "I can stop whenever I want" is the sentence that tells you the person cannot. "I only check it now and then" is the same sentence with a phone in it. The returning loop, the small reward of seeing that the machine did the thing, the pull to look once more, is a familiar architecture to anyone who has lived inside one, and the story that the machine has set you free can be true and still be the story the person on the leash tells.
I tried willpower first, because that is what a person my age tries. I left the phone in the truck and set hours, and I lasted about as long as you would expect, and when I failed it was for a reason other than discipline: I had no grounds to trust what the agents did when I was not looking, because I had never given them any. They were capable, but capable is not the same as trusted, and the distance between those two words is the whole subject of the company I now run.
The page you came from lists seven things a leader can come to trust about the AI a company uses, and I will not list them again. I will say what they add up to for one person on one morning. The framework I spent a book drawing lets you trust the AI you use because its governance travels inside it, and it shows you the evidence instead of asking you to keep watch. Every one of the seven is, at bottom, a reason not to look.
The way trust accumulates is the part I had wrong for most of my career, and the book describes it as a flywheel that turns in one direction: trust is earned, trust unlocks autonomy, autonomy is surveilled, demonstrated reliability earns more trust, and more trust unlocks more autonomy. The third clause is the one that matters to this essay, and it looks at first like the leash by another name. If autonomy is surveilled, somebody is watching, and if somebody is watching, it is me, from the trailhead, with the phone out. But it does not have to be me, and in the agents I trust it is not. An agent built from the inside out reports itself. Every action it takes carries its reason, nothing it does is silent, and the record is there when I come back to it. The surveillance in the flywheel is done by the structure the agent lives in and not by the person who would rather be fishing, and the difference between watching and reading the record is the difference between standing at the trailhead and sitting down on Sunday morning with the week's log and a cup of coffee. The first is a leash, and the second is a review, which has a time and a place and an end.
I call what comes back the analog dividend.
The analog dividend is the payoff of trusted AI measured in life rather than in output: the time that comes back off the screen once the work can be trusted to run without the person watching. It is not a productivity figure, and I have stopped keeping those. It is the morning that did not happen on a screen, the afternoon that was not interrupted, the week in a trailer with no signal in which nothing in the business falls over, and the person you are in those hours, who is not the person you have been for most of your working life, with somebody always needing something by Friday.
The business runs on three words in an order, and the order is the point. Trust is the gate, time is the measurable deliverable, and creativity is the payoff. I put time in the middle on purpose, because time is the one of the three you can count. You can count the hours you were not on a screen, and you can count the hours you were on one checking, and whether the second number is falling is a better measure of whether your AI is trustworthy than any score a vendor will sell you. The measure I hold the company to is inverted from the one every product you own is built on. Success means attention withheld, not attention captured. The product is built to let you go, and it keeps the record so that you can.
That standard cuts the other way, and it cuts at me first. The founder proves the payoff by living an analog life, and a founder chained to a screen refutes the pitch. I wrote that down as a rule for the business before I had tested it on myself, and the test has been less flattering than I expected. If I tell you that trusted AI gives you your life back and you find me at eleven at night reading agent logs, you have learned everything you need to know about whether I believe it, and you should go and buy something else.
So I will tell you where I actually am, and you can decide. The line I drew for myself is the one I would draw for any company I advise. Anything that touches money, anything that goes out in public under my name, and anything that becomes the record the business runs on waits for me before it lands, and I sit with it at a set hour, once, and not whenever the pull arrives. Everything else runs while I am out, and the agents that run it carry the rules with them and write down what they did and why. I arrived at that line by noticing, rather than by any wisdom about it, which checks I was doing because something needed me and which I was doing because the looking felt good, and the second list was longer than I would have guessed.
What I have handed over is most of the weight I never loved. The drafting, the filing, the first pass at anything that has to be written, the reading of the long documents other people send, and the bookkeeping of a business run by one person and a roster of agents all happen without me in the room. What I have not handed over is the judgment, and I do not intend to, because a machine that writes my sentences is a convenience and a machine that decides what I believe is something I would not accept from a person either.
And I have not passed my own test, not every day. I have gotten good at doing nothing in short stretches. I can manage a morning, and sometimes a whole day with the rod and no signal, and then I come back to the truck and the first thing my hand does is reach for the phone, and the first thing I look at is the log. Some of those days the log is a review and I read it once. Some of those days it is the leash, and I know the difference from the inside, and I am not going to pretend to you that the line between them is one I have stopped crossing.
The question I carry out of this is the one I would most like to hear your answer to. The hour the machine saved you went somewhere. Was it a review, or was it the leash? And if the honest answer is the leash, what would the machine have to show you before you could leave it alone?
The thing that lets me leave the agents alone is a structure rather than a feeling about them: one picture on which every agent, vendor and rule has a place, and a record that I read on Sunday instead of watching on Tuesday. The picture is on the next page: One Map.