The Underground Engineer
Dostoevsky’s Underground Man makes an argument that sounds insane until you’ve worked inside a big enough system. He says: give a man a perfect world, a glass palace where every need is met and every outcome optimised, and he will, out of pure spite, smash a corner of it. Just to prove he’s a man and not a piano key. Not because the palace is bad. Because it’s perfect. Because a being whose every move is predicted by the system has stopped being a being and become a function call.
He wrote that in 1864, before a single recommendation engine existed, and it reads like a memo about working under a roadmap.
The crystal palace is the optimised org. Everything humming, every ticket in its lane, every metric up and to the right, your next four quarters legible to anyone with the planning doc. By every external measure, good. And something in a certain kind of person, the kind who builds, starts to itch, then rebel, against precisely how good it is. Not because they want chaos. Because total predictability is indistinguishable from not existing. If the system can fully forecast your output, in what sense did you do it.
The Underground Man’s word for what he’s defending is, weirdly, the right one. Caprice. The unpredictable swerve. The thing you do that the model didn’t see coming. He’d rather act against his own advantage than confirm he’s calculable. Pathological in a person, load-bearing in a builder. The trick is routing the same impulse somewhere that creates instead of destroys.
Here’s the upgrade Dostoevsky leaves implicit. The Underground Man can only spite. All rebellion, zero creation, so his freedom curdles into bitterness. He proves he’s not a piano key by breaking things and ends up paralysed in a basement narrating his own irrelevance. The swerve without a build is just vandalism, and the vandal is as trapped as the function he’s rebelling against.
The fix isn’t killing the swerve. It’s giving it an object. I keep starting ventures partly out of this exact underground refusal, the refusal to let any system finish predicting me. But a venture is the swerve made productive. Caprice that ships. You take the same energy that’d have you smashing a corner of the palace and you point it at a blank repo instead, and now your unpredictability builds a thing the planning doc never contained.
This is quietly the answer to the AI anxiety too. The fear is the models predict and replace us. But the most replaceable worker is the one who already turned himself into a piano key. Fully legible, fully forecastable, doing exactly what the process said. The Underground Man’s spite, redirected, is a survival trait. Be the one who does the thing the model couldn’t have generated, because it came from a swerve no training set contained. The man with no pants doesn’t fear the pickpocket, and the man who refuses to be calculable doesn’t fear the calculator.
So don’t smash the palace. Walk out and build a stranger one. Keep a part of yourself the system can’t forecast, and then make that part build rather than break.
That’s the difference between a man and a function. Also between a builder and a basement.
One of a series of essays. I’m Prajjwal Chittori. prajjwalchittori.com.