2. Blog
3. vibe-coding-paradox

November 6, 2025

5 min read

Previous Post blog init Next Post You’re at the newest post!

yo, so this is basically a bunch of thoughts i’ve had about vibe coding and generative ai that have been bouncing around in my head for a while. i asked claude to structure this properly so it’s actually readable, but these are my genuine thoughts on the whole thing.

the old days

i started programming around 2019, maybe 2020. back then, there were no ai tools whatsoever - completely different time.

i used to work with flask a lot. knew pretty much every single error by heart because i’d seen them all a hundred times. stack overflow was my home. i’d spend hours and hours troubleshooting a single bug, going through random articles, github issues, even hitting page 2 of google.

and when you finally solved it? that feeling was incredible. unbelievable. there was something deeply satisfying about wrapping your head around why something wasn’t working, then finally cracking it after hours of struggle.

enter chatgpt

got introduced to generative ai in april 2024 through chatgpt. i’d actually refused to use it before that. but then i was working with a friend who was using it, and it felt like so much power.

it could fix problems so fucking fast. faster than i ever could. we finished that project in record time.

few months later, started using github copilot through their student program. even crazier. the fact that it could write code in my codebase, understand different files, context-switch between them - felt like the next big thing. i remember being obsessed with their update logs, checking what new features they’d pushed.

then a friend introduced me to claude code. started with claude pro, now i’m on claude max. basically paying a fuck ton of money to llms at this point.

but here’s the thing

i hate it.

weird part is, i’m someone who’s in this modern space but loves the old way things were done. going through random articles, diving deep into github issues, actually understanding what’s happening under the hood.

it’s easier now, sure. get an error? ask chatgpt. fixed in seconds. but i learned so much more the hard way. that troubleshooting process, trying to wrap your head around why something isn’t working - that’s where real understanding came from.

the productivity thing

from a productivity perspective, i get it. the gains are insane. i believe in that. but from a developer perspective - someone who’s passionate about code, who cares about understanding it - vibe coding isn’t it. at least not for me.

claude code is at a point where i can tell it to get details from the database, publish an event with that id, set up a listener for the new feature, and boom - it’s done. no testing needed. (obviously you should always test, i’m just showing how powerful these tools have become.)

i’ll get a ticket done super fast now. but what’s the point? there’s no satisfaction in it. every time i open claude code, i want to use it less, but i’m trapped because it makes me unbelievably productive.

the “anyone can code” thing

there’s this idea that anyone can vibe code now. but there are levels to this.

your years of experience, your understanding of systems - that changes everything about how you use these tools. that’s why i call bullshit on business owners thinking they can build entire applications through ai without any technical knowledge. even the best business person needs someone technical who actually understands what’s happening under the hood.

the future

what happens long-term?

most university students right now (including me) are using generative ai to write their code. when the pre-chatgpt generation leaves the industry, we’ll be left with a workforce that’s heavily reliant on these tools. sure, some will use ai for productivity while maintaining deep knowledge, but how many?

we’re heading toward a really different, dynamic workspace. interesting, but we don’t know how it’ll play out.

my personal struggle

i’ve become too reliant on these tools.

my rule is: if i can’t instruct the ai step by step, if i don’t have a mental map of what needs to be done, then i don’t really know what’s happening. that’s my bare minimum - i need to understand enough to prompt properly, to go in-depth about what i want.

my end goal? stop using ai tools completely. which i know is probably impossible at this point. even this article - i’m dumping my thoughts to claude and asking it to structure them properly so humans can actually read it.

the grunt work question

if it’s just grunt work that i’m giving to these llms, if it’s that basic, that low-level - why don’t i just do it myself? how hard can it be?

but i still use them. to keep up. to do more. it’s a weird time.

where i’m at now

i’ve reduced my usage a lot compared to before, and i’ll continue to. want to get to a point where i’m not heavily reliant on ai, even for basic shit.

i love coding, man. but all of this has made me view it on a task basis - let’s get this done, let’s get that done - without the actual satisfaction that i originally signed up for.

for someone who cares more about programming itself than what can be done with it (i know, crazy statement), vibe coding is the antithesis of everything i love about this field.

but here i am, still using it. still paying for it. still hating every second of it.

Previous Post blog init Next Post You’re at the newest post!