The engine prepares the grandmaster. It does not play the game

Before a big tournament, a grandmaster spends weeks with a chess engine. It sees further into any position than its owner ever will. It finds the resource nobody saw. Then the player walks to the board, shakes hands, and sits down alone. After the handshake, the engine is banned. It sits at home, rated above every human in history, and not invited. Chess, a game not famous for leniency, wrote that rule down. The moves belong to one mind, under a clock.

The same rule now governs most professional work. Nobody wrote it down.

A lawyer reads a clause an assistant drafted with AI in minutes. A marketing chief gets a campaign rationale written overnight. It is fluent, and it is precisely on brief. An editor finds a finished story on the desk before the reporter is halfway down a second coffee. The doctor's inbox holds a suggested diagnosis. Every trade has its own version of that morning. Nobody asked for the speed to come with a catch. It came with one anyway.

The catch is that the bottleneck has moved. It used to sit where the work was made. How long to draft the clause or file the story. It now sits where the work is tested. Does the clause hold, and is the story true? Most firms have rebuilt their workflows around the first bottleneck and left the second where it was. Untrained, unbudgeted and largely unloved. The row over whether AI will ruin the professions or rescue them misses the point. Judgement has not been eliminated. It has moved house.

The methods for the new address already exist. Most of them are old. Cross-examination was testing confident witnesses centuries before the witness was a chatbot. Marketing scientists in the Ehrenberg-Bass tradition test a brand claim against purchase data rather than the story a focus group enjoyed telling. Empirical researchers ask what evidence would prove them wrong. Reporters refuse to print a quote sourced once. Auditors assume, with great politeness, that everyone might be mistaken, including the auditor. They have a word for numbers that agree too eagerly, and it is not a kind one. Pick any trade, and you will find a habit like these, built for exactly this problem. A fluent, confident source with no duty to be right.

This is where chess earns its place. A grandmaster's edge is not a better memory for engine lines. It is knowing which moments call for calculation, and which call for the judgement engines still rate poorly. In 2005, an online tournament was opened to any pairing of humans and machines. No grandmaster won it. No supercomputer won it either. Two amateurs with off-the-shelf software beat a machine that outranked every human alive. Garry Kasparov, beaten by a machine himself eight years earlier, drew the lesson without flinching. A weak player with a better process beats a strong player with a worse one. It beats the machine too.

The easy answers fail this test. Asking a similar system to mark its own homework compounds the problem it was meant to solve. Reading the draft again, slowly and without a method, mostly produces a slower yes. In 2023 a New York court fined two lawyers over a brief resting on cases a chatbot had invented whole. The authorities they cited were beyond dispute, never having existed. One of the lawyers grew suspicious and asked the chatbot whether the cases were real. He asked the machine that had invented them. It said yes. Of course it said yes. The fine was small enough that no firm lost sleep over it. The lesson was not. Nothing in how the work was organised had sent him anywhere better.

What survives is not the knack of operating the tools. Within a decade, that will be as common as typing. Nor is it collecting methods by the dozen, since mastering one properly has always beaten mastering ten badly. It is a narrower, harder thing. Command of the few methods your own role requires to test what machines and powerful people hand you. And the judgement to know which one the moment in front of you calls for. Treat an AI draft that way and it becomes what Hobbes called counsel rather than command. It arrives with no authority of its own. It earns whatever authority it gets from the method that examines it.

None of this requires a verdict on AI, destroyer or saviour. Both assume the technology decides the outcome on its own. It does not. It multiplies whatever it finds. Whoever walks back into that meeting with the confident draft faces the question a grandmaster answers every time the clock starts. Not what the engine said. Which method does this position require? And who, exactly, is playing the game?