I'll begin this with the preface that I'm just about to go on holiday. So make - Pastebin.com

Pastebin

API<br>tools<br>faq

paste

Login<br>Sign up

SHARE

TWEET

Untitled

a guest

Jul 30th, 2026

22

Never

Add comment

Not a member of Pastebin yet?<br>Sign Up ,<br>it unlocks many cool features!

text 4.80 KB

| Software

0 0

raw<br>download<br>clone<br>embed<br>print

report

I'll begin this with the preface that I'm just about to go on holiday. So make of that what you will. You might conclude the author is likely burnt out, tired and in need of a break - and I'll agree with you. All of these statements are true. Never the less, this article has been brewing in my head for about a week, so as some personal therapy I'll get it out onto paper. Or pastebin as the case may be.

I'm so fucking jaded by the current state of software engineering, and LLMs are to blame.

They have sucked the joy of programming out from under me.

Like many of you reading this, I'm an old school developer. I get a buzz out of solving problems. But it's not _just_ the problem solving. It's the journey to get there.

I've enjoyed the banging my head against the wall, and waking up the next morning with the solution. The shower thought that has you rushing into work in the morning to find out if your subconscious has had a brainwave. The air punch of excitement when you find an old thread hidden on some obscure website that points you in the right direction. The feeling that you get when you delete a swathe of code, and things still work beautifully. elegantly. The learning something new. The feeling of helping your peers work through some architectural problem, explaining the best way to do things.

OG programming. As I like to call it now.

All of it, for me - more or less gone. Talk to the LLM and find an answer. Maybe. There's still a certain amount of satisfaction in telling the token maxing machine that its a fucking idiot, and to go read something first before giving me such a ridiculous solution. But I digress.

Right now, what am I? I'm a glorified condom for an LLM. A slop reviewer. A last defence against the dark arts of poorly thought out code sent over at breakneck speed - delivered by an LLM.

And slop it is. I don't care what headline Sam Altman is currently feeding you. Ever single AI model I've tried is horseshit and produces horseshit code.

It might look good, it might work. But review it and perfect it is not. Far from it. In fact, without a human in the loop we'd be in an unmaintainable mess - and given many development shops are heavy in on AI - I suspect that debt is going to come knocking like a 100 pound gorilla for a shit ton of companies.

I gave the supposed frontier model from Anthropic a problem the other day - it needed a two line fix - it delivered a 600 line change. Now granted, it gave me enough of a hint about the problem that I was able to apply the two line fix myself, but still. COME ON. The AI singularity? Give me a fucking break.

PR after PR - filled with no brain LLMing. Unused parameters. Overly defensive code. Rewrites of existing library code. Rewrites of existing project code instead of abstractions. And this isn't some open model. This is the creme de la creme of LLM models. Fable, GPT 5.6. As a result I trust approximately zero output from LLMs.

Maybe I'm doing it wrong I thought to myself. Maybe I need to have a cursor file the length of my arm. Better guardrails. Maybe I need to distill all of that senior programming knowledge down into markdown. Train the monkeys better. Pass a boat ton of skills to my peers and ask them to run that first before sending in PRs.

Nope. tried that. The majority of the time? Still fucking terrible output - and yes, it's definitely LLMs writing it, I can tell by the amount of verbose comments it litters over the code base. Despite, this is. A rule telling it to do otherwise.

My role is one thing. I've shifted from writing code to reading it much more; which I'm ok with. I just worry about my peers the most. How are they learning anything new if the game now is who can write the tightest prompt to get it to obey your whim? Where does the 'this doesn't feel right' ever come from? The sitting down and planning out before coding.

The instinct that the rest of us have from building up years of experience, and knowing just from looking at a piece of code that there's a better way? No longer gonna exist. You are going to be relying on us old guys in the corner. The ponytailed fuckers drawing on cigarettes screaming at clouds and complaining about how back in my day we squeezed everything onto a computer with 4MB of memory and every byte counted.

Maybe it doesn't matter. Maybe this is the new world order that us old school guys need to get used to. Maybe this is the role now, if the code works. Ship it.

But Christ on a bike its depressing.

On the plus side. It does look like we are in an AI bubble. If this week of my current work is anything to go by, there's not a snowballs...