Programming rites of passage

Feb 02, 2026

Much to my sadistic little delight, I read this morning that vibe-coding (and other similar variations of LLM-driven software building) does not help you acquire skills or learn much.

LLMs, on the other hand, are trying to give you the sharpest, closest solution possible. When learning through an LLM (especially around bug triangulation, troubleshooting etc), there is hardly any room or potential for ancillary learning.

Although I have no data on this, I tend to believe that ancillary learning plays a crucial role in our growth as developers and engineers. This is probably true for every other line of work/study.

from an earlier writeup

Some months back, as I was refining vāk (my custom blog builder that runs my blog and is used by a grand total of 1 user), I was thinking about the fact that almost every decent programmer I know of has gone through this programmer's rite of passage that involves reinventing, for play and more, many little technologies that already have robust implementations.

The ordering may be off (I will touch upon that later) but it goes like this:

There are some more stages here that very few cross. After the compiler, you might get into PL-theory. Philosophising about "what even is language? what is grammar?" will lead you down a path where you might toy with your own language. There is no chance you won't end up doing a bit of LISPy stuff, or get into ML-like syntax before embarking on your own journey, discovering your own weird syntax.

Or you might branch out into formal verification if you're into being a stickler (and into math) for proving things right (or wrong, if you derive pleasure out of that). You'll possibly get into reading thesis papers more than layman blogposts (like these) and get into stuff that even Haskell engineers at Bellroy call "esoteric".

The inquisitive will continue to be so. The perpetually-curious who are intrinsically driven to the details and nuances will continue to be so even in the midst of this vibe-coding AI era.

I wonder what will happen to those that would like to skill up but will end up having stunted growth because vibecoding and eager-to-build LLMs are here and the humans are not conscious or aware enough to recognize that the grind (that the LLM eliminates) is what helps skill-up.

P.S: The rite of passage is not necessarily sequential. Some folks head straight to Crafting Interpreters. Some skip the blog builder and have to write parsers at their job. And some formal verification stalwarts have no skill or craving to build their own site from scratch.