Two Buttons Are Enough

That feeling has a name, and it is seventy years old. It was written down by a psychiatrist, before the machines it describes existed, and it says something almost no one expects it to say: checking the moves was never going to be your job again — and that is not where your control was stored.

In 1948, in a small laboratory at a psychiatric hospital in Gloucester, England, a doctor named W. Ross Ashby wired four war-surplus control units into a machine he called the homeostat. Whatever he did to it — reversed its connections, pinned one part, hit it with a disturbance it had never met — the thing twitched, searched, and settled back into balance. The newspapers called it a thinking machine. Ashby said it was something plainer and stranger: a machine that holds a condition steady against a world that will not sit still.

Eight years later he published the rule underneath it, and it is one of those rules that sounds like bookkeeping and behaves like a law of nature. The Law of Requisite Variety: count the moves the world can make against the thing you care about; count the moves you can make in reply. If the world out-counts you, the difference does not politely wait outside. It comes through, and it lands in the outcome. His five-word version: only variety can destroy variety.

Read quickly, the law is terrifying. The world has endless moves; you have a handful; control should be impossible. But there is a thermostat on your wall, and it wins all winter. The sky has a thousand moves — the cold front, the open window, the afternoon sun, the oven, a room full of guests. The furnace has two: on and off. And the room holds at 68.

The thermostat is not cheating Ashby. It is exploiting the clause most people miss: variety is counted at the dial, not in the sky. The thermostat does not defend the weather. It defends one number. And against one defended number, the whole roaring sky collapses into exactly two states that matter — too cold, or not. Two states. Two moves. Requisite.

The narrower the thing you defend, the smaller the repertoire you need to defend it. Hold that clause. It is the whole future of your working life.

Every tool before this one had less variety than its operator. The hammer has one move. The spreadsheet has millions of states, but you supplied every keystroke; its variety was your variety, amplified. The machine in your stack is the first tool whose repertoire exceeds yours. It makes more distinct moves in an afternoon than you can properly review in a week, and the gap widens every quarter.

Most institutions read this the terrifying way and reach for move-matching. Review every output. Write a rule for every case. Convene a committee to enumerate the failure modes of a system that generates new ones faster than the committee can schedule meetings. Ashby has a verdict on this strategy, and the verdict is: out-counted. You cannot regulate a high-variety system by matching its moves. That is not a management opinion. It is arithmetic.

But the same law, read the other way, is a release. You were never required to match the machine’s moves. You are required to match the variety at the dial — at the outcome you have decided to defend. In 1970, in his last great paper, Ashby and Roger Conant proved the companion rule: every good regulator of a system must contain a model of that system. That is why the stack can absorb the stack’s variety and you cannot; it can model itself move for move. What you contain — the one model no machine holds on your behalf — is the model of what you will accept.

The room does not ask the thermostat to fight the sky. It asks it to know one number and defend it.

So when the tool out-moves you, control does not disappear. It migrates. It leaves the moves and takes up residence in the criterion. You stop being the player. You become the win condition.

1. Settings pages will die.

   Every knob was a place where you supplied the variety. As the machine supplies it, interfaces shed controls while capability explodes, converging on a statement of done plus accept and reject. Chat already ate the menu tree. The checkbox farm is next.

2. Rulebooks will lose to acceptance tests.

   Institutions that regulate machine work by enumerating moves will be out-counted by the machines they regulate. The ones that hold will regulate at the dial — audit the outcome, price it, guarantee it. An insurer never watches your moves; it prices the state you end in and stands behind it. That is Ashby’s kind of regulator, and it is why the insurers will end up writing the rules.

3. Middle management was variety-matching, and it goes.

   Each layer of a hierarchy existed to absorb the variety of the layer below and pass a summary upward. Machines now absorb machine variety. Organizations flatten toward whoever owns the dial — the criterion-holders — plus their stacks.

4. Stating “done” becomes the paid skill.

   The scarce professional act is specifying the win condition before the work exists — and its shadow, the list of what you will not accept, the No-List. Interviews will stop probing how a person works and start probing how precisely they can say what done means.

5. Everything converges on two buttons.

   Not because designers copy each other — because once the goal is narrow and stated, two moves are requisite variety. Accept and reject is the thermostat switch of knowledge work: the shape of retained control, not the sign of surrendered control.

I wrote this week that my two favorite prompts are four words long: you do it, and you decide. What I did not say is whose law I was obeying. Every morning the stack drafts, files, and moves, and I hold one thing — the criterion. My name goes on what leaves the desk, so nothing leaves it until I press the button. And I have learned to read my own hesitation: when I cannot state what done means, the machine is not the problem. The thinking is not finished. The criterion is the last piece of the work that cannot be handed over, which is another way of saying what I meant when I wrote that the last line runs through you.

Already true

Chat interfaces replaced menu trees. Coding agents ship against acceptance tests, not keystroke review. Finished work now arrives as a state change holding for accept or reject. Insurers are moving to stand behind machine work. My own desk has run on four words for months.

What has to happen

Verification at the dial must get cheap. A criterion you cannot check is a knob you do not have — if the only way to know whether the work is done is to redo the work, the two buttons are theater. The assay has to cost less than the labor, or control quietly returns to the machine.

Where I am probably wrong

Ashby’s law governs holding an outcome inside a set. If what I want from the machine is the outcome I could not have specified — the better question, the category I did not see coming — then regulating at the dial caps exactly the thing I bought. And the harder version: my sense of acceptable is itself trained by what the stack shows me. If the machine can move the dial’s setting, it holds the dial, and the two buttons are its interface to me. I do not know how to rule that out. The feed already did it once.

The proposal is still on your screen. Forty-one moves you could not have made and cannot fully audit, and beneath them, two buttons.

That is not the picture of a person losing control. It is the picture of control finding its correct size. Nothing ever asked you to out-move a machine that mints moves, any more than the room asks the thermostat to fight the sky. It asks you to know your number and defend it.

Know your number. Then press the button.

Two are enough.