First principle · The coastline paradox
The Coast Has No Length
South Carolina’s coast is 187 miles long. It is also 2,876 miles long. Both numbers are official. The gap between them is the most useful thing I know about what AI is about to do to work.

Ruler: 5.0 miles. Coast: 282 miles.
- General coastline
- 187 miles
- Tidal shoreline
- 2,876 miles
- Same coast
- Two rulers
You walk out to the garage to clear one shelf. To clear the shelf you have to move the paint cans. Behind the paint cans is a box from the last move. By noon the job you could see from the doorway has become eleven jobs you could only see from up close.
Two numbers for one coast
The federal government publishes two lengths for the coast of South Carolina. The general coastline, the outline you would draw with a thick pencil, is 187 miles. The tidal shoreline, which follows the sounds, the bays, the rivers and the creeks up to where tidewater narrows to a hundred feet, is 2,876 miles.
Nobody made a mistake. They used two rulers. Anyone who has taken a boat behind the barrier islands knows which number feels true. Every creek has creeks.
Richardson’s ruler
Lewis Fry Richardson was an English mathematician and a Quaker pacifist who spent his last years asking whether countries with longer shared borders fight more wars. He could not get started, because the borders would not hold still. One source gave the border between Spain and Portugal as 987 kilometers. Another gave 1,214.
He traced the difference to the ruler. Walk a pair of dividers along a map with the points set wide and you step across the bays. Set them narrow and you have to go in. The shorter the step, the longer the border. What surprised him was that it never settled. A circle measured with smaller and smaller steps closes in on one number. A coast keeps growing.
His work was published after he died. In 1967 Benoit Mandelbrot picked it up in a paper whose title was a question: how long is the coast of Britain? His answer was that the question has no answer. A coast does not have a length. It has a roughness, and the roughness tells you how fast the length grows as the ruler shrinks.
L(ε) = F · ε1 − D
A rough thing has no size until you decide how closely to look.
So “how long is it?” is half a question. The whole question is “how long, at what ruler?”
Work is a coast
Most work is rough in exactly this way. How big is the job of reading a lease, checking the books, cleaning a kitchen, inspecting a house? It depends how closely you look, and you can always look closer.
We never noticed, because people all carry about the same ruler. Human attention is expensive, so we look at about the same grain, and we mistook our shared ruler for the size of the job. “That’s a two-hour job” always meant two hours at the grain a person can afford.
An AI delegate does not carry our ruler. Looking closer costs it almost nothing. It can read every clause, test every line of the ledger, open every drawer. So here is the part most people miss. A delegate does not make the work shorter. It finishes the 187 miles before lunch and then, unless someone tells it otherwise, it starts up the creeks.
And the finer answer is not the truer one. 2,876 is not the corrected version of 187. A ship’s pilot needs the first number. An oysterman needs the second. Which means Done is never waiting at the bottom of a job, because there is no bottom. Done is declared, by someone who chose the ruler.
Overnight, your delegate went through the house you want to buy. The old inspector would have found fourteen things. The report on your screen has four thousand. Every one of them is real. You are not short of information. You are short a ruler.
Five predictions
Work will not run out. It will get finer.
Every finished job uncovers the same job one grain down. Backlogs will grow in the very years delegates get better, and people will be confused by it. They were picturing a smooth coast.
“Thorough” will stop being praise.
When looking closer is free, thoroughness is the default and often the fault. The compliment will move to the person who knew where to stop.
Quotes will name the ruler before the price.
An estimate for an audit, an inspection or a review will say what grain it was done at, because without that, two quotes cannot be compared and neither can two results. “We checked everything” will mean nothing.
The longest report will lose.
A report will be judged by what it left out on purpose. Four thousand findings will be read as a sign that nobody gave the delegate a ruler, and the blame will land on the person, not the machine.
“How closely?” becomes the first instruction.
Before you tell a delegate what to do, you will tell it at what grain. People will write their ruler down once, like a standing order, and the businesses that serve them well will ask for it once and keep it.
I am already living this one
I gave my own AI a standing instruction: only bring matters to me worthy of my attention. I thought I was protecting my time. I was handing it a ruler.
When I teach people to tell their AI “order dinner,” I tell them the test for done is the food in your stomach. That is a ruler too. It says at what grain the job counts as finished, and it sets that grain in the world, not on a screen. The same idea sits under The Check You Skip: choosing not to look closer is a decision, and it is yours.
Charleston is a good place to learn this. I live beside a coast that is fifteen times longer than it is.
The claim ledger
- Already true
- The two federal numbers for one coast. Richardson’s borders. And anyone who has asked an AI to review the same document twice has watched the second pass find what the first did not, and a third find more.
- What has to happen
- Looking closely has to keep getting cheaper, and people have to keep leaving the grain unsaid. Both look likely to me.
- Where I am probably wrong
- A real coast stops being rough at the grain of sand, and real jobs may bottom out sooner than I think. If most work is smooth below a certain grain, delegates will simply finish it and my first prediction fails. Delegates may also learn a sensible ruler by watching us. My bet is that they learn our habits before they learn our reasons.
Back in the garage
The shelf was never one job. It was a coast, and you measured it from the doorway. That was not a mistake. It was the only ruler you could afford.
Soon you can afford any ruler at all, and the question you never had to ask becomes the first one. How closely? The coast will not tell you. It has no length.
You bring the ruler.
Background
- Benoit Mandelbrot, “How Long Is the Coast of Britain? Statistical Self-Similarity and Fractional Dimension,” Science, 1967.
- NOAA Office for Coastal Management, Shoreline Mileage of the United States.
- Lewis Fry Richardson’s study of border lengths, published after his death and cited by Mandelbrot.
- The Last Line Is You
- The Tide Is a Clock
- A Life Has No Wage