How To Write Unmaintainable Code (1999)

[-]

jon01001010@reddit

I made the worst maintainable code/html ide I was thinking do it in basic then said whats worse https://github.com/jon01001010/neos-os-

[-]

This was written as a joke in 1999, but modern Java frameworks have adopted it as a manual. Just look at the implementation standards of mainstream frameworks: a single method call requires jumping through a dozen classes, and more likely than not, those are just interfaces. You then have to hunt for the implementation, which might even be dynamically generated at runtime, making the actual logic invisible to you. It is a masterpiece of defensive programming—so complex that even the original creator cannot fully grasp the intent. Well... good luck trying to maintain that.

[-]

Icy-Huckleberry-4450@reddit

how to design unreadable webpage

[-]

yotemato@reddit

The incentive to write good, maintainable code is completely gone. Fuck it. Let’s slop it up and see what happens.

[-]

Valmar33@reddit

The incentive to write good, maintainable code is completely gone. Fuck it. Let’s slop it up and see what happens.

The sloppers were never interested in writing code in the first place. They had every incentive to avoid doing the work of learning how to program ~ how to use logic, how to problem solve ~ if they can. They want something else to do it for them. It's like... those idiots who had perfectly capable legs, but they chose to drive everywhere on mobility scooters instead.

The worst part is that these LLMs are built on top of plagiarized and stolen code ~ actual code written by actual people. So the sloppers have absolutely no idea how the LLMs actually work ~ they seem to think it's literally magic.

[-]

Lily722022@reddit

>The worst part is that these LLMs are built on top of plagiarized and stolen code \~ actual code written by actual people.

I think this in particular is really flawed logic. AI "steals code" in the same way humans "steal code"... It isn't anymore plagiarism than reading a comment on StackOverflow telling you how to solve a problem somebody else had and repurposing the solution for your own.

[-]

GasterIHardlyKnowHer@reddit

Okay, let's do another one: an AI agent or chatbot searches the internet on how to solve a problem, finds GPL licensed code and implements it.

Now what?

[-]

flatfinger@reddit

That's a more ambiguous situation. If someone were to decompose a program into constituent non-copyrightable algorithms and give a description of those algorithms to someone else who coded them without having seen the original, the clean-room approach would prove that the resulting program was not a derived work of the original program. If the new program was written by someone who had seen the original program, but is indistinguishable from something that could plausibly have been produced via clean-room methods, then it should probably likewise not be viewed as a derivative work (because of the scenes-a-faire doctrine), though proving that it shouldn't be considered one would be harder.

With generative AI tools, it's hard to tell what kind of decomposition and regeneration took place.

[-]

lelanthran@reddit

The sloppers were never interested in writing code in the first place.

That includes many of the people who claim that AI now allows them to create stuff they never had time for before.

We've all seen these claims: "I'm 50, a senior/staff/chief/principal engineer, so I am definitely a smart programmer, and now I can create a whole new product in a weekend!".

They're the class of programmer who focused on delivery over maintainability, and wished for years to be able to get their salary without writing any code.

The thing is, they could have had their wish decades ago; there's a ton of positions at every company for analysts who decode business requirements into a specification that engineers then design and implement.

They didn't choose those positions, because it pays roughly half what a SWE role pays. Now they are willingly jumping to those positions not realising that it's only a matter of time before the lag disappears and economic reality catches up.

Namely The person who comes up with the requirements and a vague high-level design (must use Azure service $FOO, must use microservices, must not be self-hosted, use protobuffers, etc) earns half what a SWE earns!

[-]

sarcasticbaldguy@reddit

We've all seen these claims: "I'm 50, a senior/staff/chief/principal engineer, so I am definitely a smart programmer, and now I can create a whole new product in a weekend!".

I'll use it like that to rapidly iterate on a prototype that my product folks can interact with, but then we throw it away and actually build the thing.

[-]

rainbowlolipop@reddit

Mercifully I get to work on scientific research stuff so maintainability and ease of understanding the complex "business logic" are more important than shinies.

[-]

ekipan85@reddit

Most software is akin to literal magic and has been for decades. Do you know the millions of lines of code connecting the keys you type to the pixels on your screen or the bits through your ethernet cable and wifi radio? Application libraries built on framework libraries built on language libraries built on operating system libraries built on kernel code and hardware drivers.

Slop turns this horrible problem into a hopeless one. At least a Linux system has source code, written with intent by many persons, that you could in principle hope to read and understand.

I think we need to go full Chuck Moore and throw all of it into the garbage. Take responsibility for every instruction the CPU ingests. At least, that's what I fantasize. I dunno that I'd ever be that willing. The hardware has also gotten so damn complex.

[-]

Kalium@reddit

I recently was in a meeting in which someone less than seriously suggested pushing four unrelated software packages, all of which do different things, into an LLM and asking it to combine the best of them. This was and is obvious nonsense - they do different tasks, work in entirely different ways, and are implemented in wholly different languages.

There was one person in the meeting that I'm convinced took it entirely seriously. This manager has never been a software developer and appears to genuinely believe that LLMs are magic. I'm just glad I don't report to them.

[-]

rzet@reddit

I work with 1 total LLMbrainrot folk and 2 half baked as well. cry...

[-]

OffbeatDrizzle@reddit

found the sloperator

[-]

SaxAppeal@reddit

They can’t stop you from ordering a steak and a glass of water!

[-]

MehYam@reddit

I used to be a real piece of shit

[-]

reborngoat@reddit

I still am, but I used to be too.

[-]

SaxAppeal@reddit

I said I WAS!!

[-]

PuppyCocktheFirst@reddit

I bet your hair slicks back reeeeeeeal nice!

[-]

Leihd@reddit

Unless it's a clueless manager who thinks you're underperforming.

I once worked with a guy who churned out a lot of code, apparently he thought it was a good metric of skill. This was pre-ai, and his code was still slop. I'd forgive it maybe if the code even worked.

[-]

youngggggg@reddit

I mean if AI is the one maintaining it then what does it matter ultimately? Code quality and “maintainability” feel much more important for human readers of code and I worry all this stuff is quickly losing its value

[-]

HighRelevancy@reddit

I think pretty much every investigation into the subject has discovered that readability, maintainability, and other such qualities of code are basically the same for humans as for AI. AI produced code isn't any more maintainable by AI than by human. AI can read more of it faster than you can and can catch up faster, but it's still got to do the same process of reading all the shit to deduce what it means.

[-]

youngggggg@reddit

I hope this remains true into the future. I worry about what this all looks like in a couple years

[-]

HighRelevancy@reddit

I think even as they get faster and smarter, it's still basically a given that it will be a factor. They'll get faster at resolving the mysteries of bad code, and maybe fast enough that it won't matter for many use cases, but it will still be slower/take more work than if the code was good/clean/readable/etc.

The only way I see this changing is if we let them start naming/commenting things in non-human-language terms. I've seen experiments where LLMs iterated on "thinking" in raw embedded tokens instead of unembedding them into written text and reingesting them, and it was effective in improving quality of output. Current thinking models are putting all their intermediate thoughts/working into human-readable terms and it's a narrowing of what they can represent internally. It's like writing notes for your future self but you're only allowed to use a beginner's French dictionary - it works but it's sure a lot harder to express complex ideas than working in your native tongue.

[-]

youngggggg@reddit

👍 I appreciate the thoughtful response here, I’m very much speaking from a place of fear

[-]

HighRelevancy@reddit

I think we're very much in an era where AI is a major tool forever. My job now is markedly different to six months ago. But I don't think, no matter how good it gets, it will actually replace humans. Someone needs to have ideas, make design decisions, validate what the AI is producing, and actually work with the other humans that software engineer provides services to.

The actual act of programming is still fun and I'm not saying you have to change what you do, but I do think you should at least give the free GitHub copilot or something a go. It's worth understanding these tools and what they're actually good for.

[-]

youngggggg@reddit

I’m a professional and use a variety of AI tools everyday. But I think my perspective is ultimately biased by working on a SaaS app that simply isn’t that large or complicated. For what my company is doing, every dev feels insanely replaceable by AI to me.

[-]

lelanthran@reddit

I mean if AI is the one maintaining it then what does it matter ultimately?

Maybe not to the code, but to the developer, certainly! Roles where you build up a specification existed for decades, but they pay very little.

The ability to program was a large reason why you, the developer, was paid double what the people with business-domain knowledge were being paid to produce the business requirements.

If that ability does not matter any more, what extra value do you, the now-ex-developer, can bring that will justify a salary larger than the people who were already doing what you now started doing?

[-]

0xbenedikt@reddit

How To Write Unmaintainable Code (2026)

chatgpt.com

[-]

ea_nasir_official_@reddit

Or claude/codex/openclaw or whatever the ai bros use to write their code for them while pretending to be smart

[-]

VelvetWhiteRabbit@reddit

If you are writing web servers in 2026 without LLMs to write for you, I feel sorry for you.

[-]

FriendlyKillerCroc@reddit

Aha there it is. This subreddit has shifted again since there is no point arguing that LLMs aren't obviously stupidly powerful anymore and calling them glorified auto complete would even make you look stupid here.

We are now at the stage of feeling superiority over people who use LLMs.

[-]

GasterIHardlyKnowHer@reddit

Warning, the user I'm replying to is a known AI troll

[-]

ea_nasir_official_@reddit

I use locally hosted llms. Vibecoding is just really stupid. It's hard to maintain and AI is not a good coder. It's a average coder. It's very bad at what it does. It makes output that works, but no human nuance or decision.

[-]

MadCervantes@reddit

Locally hosted LLMs are vastly less powerful than frontier hosted models.

[-]

ea_nasir_official_@reddit

I've used big boi llms like Claude and Codex but really none are any good and it's better results to just write your own code with a small local llm with RAG and web search for assistance

[-]

neithere@reddit

What are you using locally? What are the resource requirements, what are your use cases / processes? Any articles on that?

I had a personal mini-hackathon with Codex just a while ago and got very tired from it. It's great for initial prototyping and not bad for keeping docs and code in sync but then even for high-level stuff and with detailed ADRs it starts drifting and it's just easier to read and write some code than doing the same through this thing :/ still trying to find the sweet spot.

[-]

ea_nasir_official_@reddit

I use llama.cpp (Vulkan). There aren't set requirements other than what can fit on your hardware. My main productivity machine is a Thinkpad P14S with 32GB ram and an 8840HS. My bigger but slower model is Qwen3.5 35B A3B (UD-IQ3_XXS), but i'm experimenting with Gemma 4 26B a4b. My small model when I don't need the big ones eating my ram is Qwen 3.5 4B. Check out the unsloth quants and see what fits in your hardware. Q4 is generally considered the smallest that you can actually get quality output from.

Noteably the biggest bottleneck with LLMs these days is not compute, but memory bandwidth. If you have a dedicated GPU you should get some massive speed boosts.

[-]

jug6ernaut@reddit

I think this is where i've landed also. Using LLM's for super small focused changes, basically using them to automate writing code I already have planed.

Then web search LLM's are very useful for learning new topics.

[-]

RecursiveServitor@reddit

It can improve its own code. All non-trivial programming is iterative. A super 100x programmer isn't going to one-shot perfect code either. So, the thing that matters in terms of AI is how fast it can turn shitty code into good code. So far my experiments suggest the answer is "faster than me".

[-]

dillanthumous@reddit

The problem is they are simultaneously stupidly powerful and powerfully stupid. 10x a mistake is still a mistake.

[-]

lolimouto_enjoyer@reddit

There are many ways to use LLM some are valid and good others are not.

[-]

nickcash@reddit

inb4 they find this thread and copypaste one of their three preprogrammed responses about how you're just not using the right model. you have to use the new model from ploob. no one uses claude anymore. it's all about heebee these days

[-]

Valmar33@reddit

inb4 they find this thread and copypaste one of their three preprogrammed responses about how you're just not using the right model. you have to use the new model from ploob. no one uses claude anymore. it's all about heebee these days

The great thing about this sort of absurd logic is how absurd it sounds when you use it in a different context:

"You're just not using programming language", "you just have to rewrite in Rust", "nobody uses C anymore", and so on. Oh, wait...

[-]

Full-Spectral@reddit

But the thing about it is that Rust is that it's not about spitting out more code faster than the C++ it's primarily replacing, and is actually often more front loaded. It's about BETTER code, and it provable is capable of providing that. And it's not thinking for you, it's making you work harder to understand your data relationships. The payoff though can be very significant.

[-]

bonerfleximus@reddit

You'll figure it out eventually. World is changing whether we like it or not unfortunately

[-]

fiah84@reddit

except SQL, SQL is forever

[-]

Truenoiz@reddit

Right, but SQL is so easy that it's trivial. Please hand over the production database so I can fix it.

[-]

fiah84@reddit

Please hand over the production database

I grant thee select, and that's it ya fricking clanker

[-]

RelatableRedditer@reddit

and then it constructs the most convoluted selector with 28472 redundant joins

[-]

bonerfleximus@reddit

Haven't had issues getting good results for SQL, even for more advanced edge cases like dynamic SQL referencing tempdb objects in different scopes and varied schema.

I guess having a strong enough understanding of what the finished SQL should look like was a prerequisite but it's a lot better than I've seen people give credit for.

[-]

dillanthumous@reddit

YoU NeEd tO LeaRN To PrOoMpt!

[-]

ofcistilloveyou@reddit

I'm proompting all over the place.

[-]

lolimouto_enjoyer@reddit

You forgot you have to use skills, subagents, planner and... what else was there?

[-]

alex-weej@reddit

MCP, Tools, Spells, Incantations...

[-]

dillanthumous@reddit

Prayers to the Omnisiah.

[-]

reborngoat@reddit

This is the answer. I stopped prompting and started praying to the Omnissiah, and now I have a harem of lovely toasters!

[-]

Kalium@reddit

ZERO ZERO ONE ZERO ZERO ONE ONE ONE...

[-]

QuickQuirk@reddit

You forgot to install openclaw as your developer.

[-]

Disastrous_Crew_9260@reddit

Ok syntax monkey 🐒

[-]

BaNyaaNyaa@reddit

I don't get you people. I'm using Magnum 6.7 and I'm running 420 agents to do my job! You must be using it wrong

[-]

AceLamina@reddit

Literally had an argument with someone who's a SENIOR engineer because he thought AI will eventually replace all engineers

After a bit of digging, he was hired out of college back in 2021 and works at Amazon and vibe codes at work everyday.

[-]

gimpwiz@reddit

They hand out titles like candy sometimes, eh

[-]

AceLamina@reddit

Back then, 100% The SWE boom meant literally anyone could become a engineer, he's just one of them

[-]

Full-Spectral@reddit

The ultimate goal of that track is to then start a Youtube channel handing out wisdom from the mount as a Former Amazon Lead.

[-]

gc3@reddit

I've found the main sin of Ai code so far is copying and pasting functions. It's usually pretty good about the other issues listed here. Claude code even has to generate documentation when it maintains code it wrote itself.

I've had to tell it to combine functions in the interest of dry

[-]

MisterMeow35@reddit

Let's not forget that LLM models were trained to write code on real human projects.

[-]

thomasmitschke@reddit

The bible we all followed

[-]

LazyAAA@reddit

Make sure that every method does a little bit more (or less) than its name suggests. As a simple example, a method named isValid(x) should as a side effect convert x to binary and store the result in a database.

Pure gold - simple but yet so damaging :)

[-]

The_Northern_Light@reddit

I’m cleaning up some researcher code (Matlab / Python) and it’s amazing, literally they do literally all of this (except the Java / C specific stuff)

[-]

zippy72@reddit

I feel this would pair well with the InterCAL manual

[-]

fagnerbrack@reddit

Classic!

[-]

worldofzero@reddit

This feels like it could be retitled "Best Practices of a Vibe Coder" and it'd be equally accurate... We lost so much of the profession so fast recently.

[-]

shizzy0@reddit

Yeah, but at least it writes comments.

[-]

lolimouto_enjoyer@reddit

Horrible stating-the-obvious comments most of the time.

[-]

Kered13@reddit

What bugs me is when you will tell it to do (or not do) something in the prompt. Something obvious that you shouldn't have to tell it, but you do because it's AI, and then it decides it needs to add a comment saying that thing. No, you don't need a comment in the code reminding you to use best practices.

[-]

SnugglyCoderGuy@reddit

// Load the config from the specified file
config, err := loadConfigFromFile(filename)

[-]

EliSka93@reddit

Stating the obvious in 5 times as many words as necessary.

[-]

jmasterfunk@reddit

Only obvious to those who can actually code.

[-]

Globbi@reddit

What are you talking about? I'm reading the list and it has ZERO to do with vibe coding. Not because coding AIs didn't exist back then, but because vast majority of complaints are not relevant to AI-written code.

[-]

beenny_Booo@reddit

It's wild how many of these 'tips' from 1999 are still accidentally implemented by junior devs today. Or even senior devs on a bad day, lol.

[-]

andree182@reddit

> Make "improvements" to your code often, and force users to upgrade often - after all, no one wants to be running an outdated version. Just because they think they're happy with the program as it is, just think how much happier they will be after you've "fixed" it! Don't tell anyone what the differences between versions are unless you are forced to - after all, why tell someone about bugs in the old version they might never have noticed otherwise?

Huh, this sounds like exact definition of AI code tools, which keep changing/optimizing/rearranging stuff you never asked it to do...

[-]

dillanthumous@reddit

I get a laugh out of reading the 'reasoning' chain sometimes. The LLM spooling out reams of reminders to itself not to do things incorrectly while simultaneously justifying making extensive breaking changes is the clearest evidence that rationality is not an intrinsic property of language.

[-]

tabacaru@reddit

Once I was angry it gave me a wrong solution and I showed it the right solution with an example of the correct output.

It proceeded to still tell me I'm wrong, then start showing me an example of inputs where I was wrong, only to work itself out that the example it generated actually did match the correct output and proceeded to then say the example is actually correct.

So in a single paragraph it managed to vehemently suggest I'm wrong, give me an example where I'm wrong, but the example turned out to confirm I'm right.

It's insanity that people can take the outputs of LLMs and just assume they're magic.

[-]

TheDevilsAdvokaat@reddit

chatgpt once told me that zero is an even number greater than one.

I wish I had kept the screenshot...this was back when chatgpt was very new and it's better now...but it absolutely taught me not to rely just on ai, but to double check everything it says.

[-]

turunambartanen@reddit

Not restricted to AI software. It's a property of early release versions in general.

When writing a UI in rust egui is a solid choice. It's very easy to use and has most of the features you need. But fuck me, every new version has a small breaking change. Nothing major. I can fix my code to use the new methods no issue and it doesn't take long. But OH MY GOD, bring out 1.0 already!

[-]

TheDevilsAdvokaat@reddit

This is quit hilarious. I've actually SEEN comments like this.

i=i+1; //add one to i

In which case why the hell bother to write the comment?

[-]

LessonStudio@reddit

Obviously using AI and not paying attention is a very good way.

But, I would suggest that certain languages and frameworks can really encourage it. Yes, being very very careful will help, but:

Enterprise Java - this works so hard to organize things that it just induces a higher level of difficulty for the smallest of things. Maybe, it then caps out an acceptable level
PHP - the worst code I've written is in PHP. I can write clean code, but the temptation for really nasty shortcuts is so in your face. The frameworks are the worst on the planet. They all say, "Non-opinionated" and then scream in your face "OBEY!!!!" as these frameworks solve a narrow set of problems well, but outside of that and you are just hacking, working around, and writing garbage.
C - This is more of a cultural thing. If you look at raylib, that API is the most beautiful C I've seen and it encourages more beautiful C.
C++ with templates - Templates buried inside a library can make that library so very easy to use. But, once programmers start using them unnecessarily in their code, it often becomes showing off, not helpful in any way at all. Makes code a nightmare to test to exhaustion.
React - What the F is wrong with those people.
Flutter - For small projects it is great. But you can see the primary flaw when you look at how there is a new library about ever 8 hours for passing data throughout the system.
Rust - I love rust. I've written a zillion lines of code. But, it does not compile in my head. I can easily miss a ? .ok .unwrap .copy .clone and not even notice it. The compiler makes this so easy to fix. But, I don't make those mistakes in C++, python, julia, C.
Javascript - For small things it is fantastic. But, the fact that typescript was needed is all that you need to know. Typescript bought a bit larger project sizes before it all goes to hell.

[-]

jerosiris@reddit

We can now write unmaintainable code at a rate that would make 1999 people’s heads explode.

[-]

SnugglyCoderGuy@reddit

[-]

AutomateAway@reddit

you merely adopted the unmaintainable code. I was born in it. Moulded by it. I didn’t see refactoring and good design patterns until I was already a man, by then it was nothing to me but an anti pattern. The unmaintainability betrays you, because it belongs to ME!

[-]

SnugglyCoderGuy@reddit

If that is true, then you managed to live long enough that you are you're own villain because you wrote the mess you were molded by.

[-]

imihnevich@reddit

Good one. Should've gone with "nothing to me but over-engineering"

[-]

card-board-board@reddit

snafucated

I'm going to use this.

[-]