Competence in the Age of AI

What chess figured out before the rest of us.

There's a particular kind of comedy in watching a Magnus Carlsen game with the engine running. The evaluation bar sits along the edge of the screen, white and black, perfectly certain. Carlsen plays a move. The bar twitches. A bright "??" pops up next to it, and somewhere a viewer rated 1400 leans back on his couch and mutters that he wouldn't have done that.

The best chess player alive, gently corrected by a man eating chips.

I do it too. We all do. Stockfish has made armchair grandmasters of everyone, and the strange part is how normal it feels. You point your phone at a position and it tells you, to a hundredth of a pawn, exactly how wrong everyone is.

It wasn't always like this. Rewind thirty years and the world champion was something closer to an oracle. When Kasparov played a move you didn't understand, the assumption ran the other way. You assumed there was something there you couldn't see. His judgment was the outer edge of what humans knew about the game, and you had no way to fact-check him because there was nothing to check him against. That gap, between his understanding and yours, got filled with reverence.

Chess is also where we first met the future. In 1997 a machine beat Kasparov, and a small universe of sixty-four squares became the first place humans were decisively, permanently outplayed by something we built. People panicked, briefly, about what it meant. Then they got over it. Today no serious player trains without an engine, and no one calls it cheating to analyze with Stockfish afterward. No one. The thing that once felt apocalyptic turned into a utility, somewhere between a calculator and a spell-checker. Superhuman intelligence in your pocket, and it's the least remarkable thing about the pocket.

So here's the part that should be surprising and somehow isn't.

The players who grew up with that god-tier tool sitting right there, the ones who could have outsourced every hard thought to a 3500-rated engine, mostly got good by refusing to. Gukesh, the current world champion, trained for years under a coach who deliberately kept engines away from him. No instant evaluations. No optimal lines handed over on a plate. He had to sit with ugly positions and work out for himself why a move was good or bad, using nothing but his own head. Carlsen is a version of the same story: largely self-taught off books and replayed games, and known for reaching for his own judgment before he reaches for the machine. Kasparov once put it as Carlsen using the engine like a tool rather than an oracle, building the mental muscles instead of having the answer handed to him. That habit is more or less the whole thing.

One of most viral quotes on AI :

That's the article, more or less. You can hand the answer to a machine. The understanding is the part that has to stay yours.

The engine will give you the move. It will not give you the why in a way that becomes yours, and if the why was never yours to begin with, the answer is close to useless. You can't extend it to the next position. You can't tell when the tool is confidently wrong. You can't do the thing one square over that nobody pre-chewed for you. A player who only knows the engine's verdicts knows a list. A player who built their own evaluation knows the game. The two look identical right up until the moment the crutch gets taken away.

There's a second thing chess quietly taught us, and lately it reads less like a chess observation than a forecast. Being the best calculator in the room stopped mattering the day the calculators showed up. At the very top, everyone prepares with the same engines and drinks from the same superhuman well, so raw analytical horsepower has become the floor rather than the edge. Nobody wins on it, because nobody is short of it. What's left to compete on is everything the engine can't hand you.

And that everything turns out to be the stuff we used to file under "soft." Nerve. The discipline to study when studying is boring, to keep grinding the same endgames long after the novelty has worn off. The composure to lose the first game of a world championship match at eighteen and not fall apart, which is roughly what Gukesh did before winning the thing. His coach used to make him watch footage of Alex Honnold free-soloing El Capitan, three thousand feet of granite with no rope, as a lesson in staying calm where a single slip ends everything. That isn't a chess lesson. It's a lesson in managing your own head under pressure, and it's exactly the sort of thing no engine will ever do for you. This is the quiet inversion of the whole AI moment: as the analytical part of any job gets commoditized, the traits we treated as secondary start carrying the weight. Discipline. Emotional steadiness. The read on a room, a negotiation, a person across the table who needs to trust you. Knowing which problem is worth solving in the first place, which is a question no model will answer for you because it isn't a question of computation at all. Intelligence, it turns out, was only ever one axis among several.

Ilya Sutskever , founder of OpenAI said it with less hedging than I just did:

We mistook intelligence for the decisive quality because for most of history it was the scarcest, and the moment it stopped being scarce, its price fell like anything else in oversupply.

I think about all this whenever someone asks whether AI is coming for their job, because chess already ran the experiment. The machines got better than us, completely and apparently forever, and the game didn't die. More people play now than at any point in its history. The professionals adapted. They stopped competing on the one thing the machine had won and got better at the things it couldn't touch. The work changed shape. It didn't vanish.

That's probably where most of us are heading. Not unemployment so much as different employment, doing the parts the tools can't, in roles that look a little strange from where we're standing now. And as these systems start to genuinely speed up the sciences, they'll crack open whole fields that don't have names yet, the way nobody in 1990 was hiring a prompt engineer. The honest answer about what we'll all be doing is that we can't quite picture it. We almost never can.

Which leaves a practical question, and it's the one I'd actually want answered if this were my career on the line: how do you prove you're good at anything once the machine can produce the same output you can, faster and for free? For most of history, competence signaled itself through the work. You wrote the report, drew the diagram, wrote the code, and the artifact was the proof. That link is breaking. The artifact no longer says much, because anyone with a subscription can generate a polished one in seconds. A clean deliverable used to be evidence. Now it's the price of entry, and it proves about as much as a quoted engine line proves you can play chess.

Chess, naturally, already solved this. Anyone can memorize Stockfish's top choice and recite it at the bar. Nobody is fooled, because the way you prove you can play is to sit down over the board, on the clock, with no phone, and perform when the tool is switched off. The credential is what you do under conditions where the crutch isn't available. The same shift is coming for the rest of us. Competence will signal itself less through the artifact and more through the things the artifact can't fake: the judgment to choose the right problem, the taste to know when something is actually good versus merely finished, the ability to defend a decision live when someone pushes back, the track record of being right over time, and the trust of people who'll vouch that you were the one steering rather than the one copy-pasting. It's a move from showing the answer to showing your understanding of it. Which is, conveniently, the same thing that made you worth listening to in the first place.

If you're sixteen or seventeen and reading this between mock tests, the question lands differently and a lot harder. You're grinding through NEET or JEE prep, twelve-hour days, a coaching syllabus that treats you like a machine for solving standardized problems quickly, and meanwhile there's a chatbot on your phone that solves those same problems instantly and never gets tired. It is reasonable to look at that and feel the floor move. Why memorize the reaction, why drill the integral, why do any of it, when the thing in your pocket already does it better?

Here's the part the panic hides. The exam hall is your over-the-board moment, the same test chess uses, the tool switched off and just you and the page. NEET and JEE are built that way on purpose, to separate the students who understand from the ones who can only quote answers. The AI that solves your problem set is the engine on the couch. It can hand you the answer, but it cannot sit the exam for you, and it cannot install the understanding that lets you solve a variation you've never seen, which is the only kind of problem these exams actually reward. So the trap isn't AI. The trap is using it the way Gukesh's coach refused to let him use the engine: to skip the struggle. Every time you ask the bot for a worked solution before you've fought the problem yourself, you're outsourcing the one thing you came to build.

So use it the other way. Solve it yourself first, badly if you have to, then ask the AI where your reasoning broke. Make it explain the step you don't get, in three different ways, until one of them clicks. Have it generate variations on a problem you just got right, to test whether you actually understood it or just got lucky. Treat it as a sparring partner that never tires and never judges you, not an oracle that does the rep on your behalf. And take the soft part seriously, because it isn't soft: the students who crack these exams are rarely the most brilliant in the room, they're the ones with the discipline to keep showing up and the composure not to spiral after a bad test. If the pressure is genuinely crushing you, that is a signal to talk to someone you trust, not a personal failing and not something to grind through alone. The exam is hard by design. You are not supposed to face it in isolation.

The eval bar can hand you the move. It can't hand you the understanding underneath it. And sooner or later in the exam hall, in the meeting, in the quiet where it's just you and the problem, the bar goes dark, and what you understood is the only thing you get to keep.