My university professors used to tell stories about feeding stacks of punched cards into mainframes, only to watch the entire job fail because one single hole was punched in the wrong column. The machine didn’t argue or “misunderstand.” It simply didn’t work. Their job was to debug, repunch the card, and try again.
Half a century later, I catch myself muttering the exact same sentence when an AI gives me the wrong output:
“This thing is not doing what I told it to do.”
That familiar frustration is the giveaway. It’s not a conversation that went badly. It’s a program with a bug.
Real programmers don’t treat the failure like a human misunderstanding. We don’t get increasingly polite or emotional. We debug. We narrow the requirements, add examples, enforce output structure, chain reasoning, and iterate. Non-programmers usually don’t. They rephrase nicely, then impatiently, then give up — treating the AI like a stubborn colleague instead of a machine that needs precise instructions.
This is the clearest proof that prompting is not natural social language. Real human conversation is fuzzy and forgiving. Prompting is not. The moment you start systematically fixing a broken prompt, you’ve crossed from talking to programming.
Every layer of abstraction in computing history looked like “not real programming” to the previous generation. Assembly looked like witchcraft to the card-punchers. Python looked like a toy to the C crowd. Now prompting looks like “vibe coding” to traditional developers. But the mental model is identical: define requirements clearly, handle edge cases, verify output, refactor when it breaks.
And here’s the important part for the AI era: this is exactly why programmers are more necessary than ever.
Every leap in abstraction — from punch cards to assembly, to C, to Python — didn’t eliminate programmers. It elevated them. The same thing is happening now. The people who truly thrive with AI aren’t the ones having casual chats with it. They’re the ones who treat prompts as code: writing them with intention, testing them, refactoring them, and building reliable systems on top of them.
In a world full of AI, the scarce skill isn’t access to the model. It’s the engineering mindset — the debugging instinct, the ability to turn vague desires into reliable outcomes. That’s still a programmer’s job.
So next time your prompt fails, don’t try to “explain it better like you would to a human.”
Pick up the virtual punch card, fix the hole, and run it again.
Because that quiet, stubborn refusal to accept “it just doesn’t work” isn’t frustration.
It’s the reason programmers will remain essential in the age of AI.

P.S. I wrote this while using AI to develop a small hobby web app https://www.poll-sim.com , which uses AI to run.