the machines shouldn't be learning python
i used to feel a sense of awe whenever i came across a deep-dive technical post on hacker news.
reading about the esoteric workings of the likes of the linux kernel, the c standard, or the grub bootloader filled me with admiration—even if i failed to grasp their entire meaning. i saw a community of digital deep-sea explorers, people so dedicated to their craft that they had plumbed the absolute depths of our technology. “one day,” i promised myself, “i’ll master a system so completely that i can write a post like that.”
today, that awe has been replaced by a profound sadness.
the joy of being a programmer, for me, was always rooted in curiosity. it was the thrill of knowing that beneath the polished surface of our user interfaces lurked a vast, intricate ocean. the apis, the assembly code, the network protocols—these were the strange and mesmerizing creatures of the deep, and i was enthralled by the dream of one day seeing them for myself.
my sadness is born from an unwelcome conviction: that this entire world we built is temporary. all the apis, abis, and elegant design patterns that govern our digital lives from the shadows are just scaffolds. they are human-centric abstractions, bridges we built to span the immense chasm between the way our minds reason and the way silicon executes.
as large language models grow more sophisticated, they will begin to traverse that bridge on their own. they will take over the maintenance of legacy systems and the development of new features. initially, their work will still be rendered in languages we can read, like python or c++, but the logic will change. the abstractions they design will eventually become increasingly alien, optimized not for human comprehension but for pure machine efficiency.
this would be a transitional phase, however.
the final step will be for the llms to discard the ultimate human-centric abstraction: programming languages. why would a machine continue to constrain itself with syntax designed for ape brains? it will learn to write directly in the ones and zeros that are its native tongue.
when this happens—and i believe it is a question of when, not if—the work of generations will be made obsolete. the art and science that captivated some of the most brilliant minds of our time, the entire discipline of coaxing logic from silicon with human language, may disappear in a matter of years.
the deep ocean will still be there, but we will no longer have a way to explore it.
while this reads as a lament—and in many ways, it is—i find myself eagerly anticipating that future. we must not forget that software, for all the artistry we pour into it, is ultimately just a tool. to have machines generate bespoke software, far more capable and efficient than anything we could write by hand, would be the ultimate victory for humanity.
it is a progress worth having, even if it comes at the cost of an art we love.
END OF FILE