Thats the point. To make you permanently depend on Anthropic and be a little paypig forever.

File: m3l6TGD5-8k.jpg (159 KB, 958x1080)

159 KB JPG

>>109874343
Next month they'll release an even smarter clanker that will fix the problems caused by the current clanker.

You mean you arent using claude to build you how to not need claude anymore?

File: 1778165395786301.jpg (47 KB, 852x854)

47 KB JPG

>>109874343
I'm using codex though

>>109874377
Same shit.

>claude, keep the code simple
Wow, that was hard

>>109874355
But claude isn't capable of maintaining the code it writes either. Is the future of software engineering just telling claude to implement the entire project from scratch every time you want to add a new feature?

>>109874386
Doesnt work, still produces a royal mess.

>>109874386
Show me one actual example of this working. If claude could actually write well structured, readable, and maintainable code, wouldn't that be the default?

>>109874391
Yep. The future of SE is two societies:
An elite luddite club of those who still know how to write their own software
and vast fields of goycattle paypig slaves to AI companies.

>>109874383
Same luddite

Personally not a fan coder harnesses, I just start a chat window on an LLM and just start discussing and it writing the initial code and then manually add code blocks for each feature/patch. I am not reinventing the wheel aka using C/C++/asm so it's doable.

>>109874343
I dont understand how people build complex thing with it. Last time I foolishly let it code the logic for traversing n-ary tree and matching stuff(strings basically) and holy fuck it polluted the entire thing with tiny little bugs and redundant stuff. On any attempt to fix it it made even more mess.
I might as well erase the whole part and rewrite it myself.

>>109874386
You forgot "make no mistakes"

File: Another Jew SaaS Post.png (12 KB, 546x152)

12 KB PNG

>>109874343

>>109874343
>10x over-engineered, unreadable by humans, and unmaintainable*
accurate

>>109874343
>claude, if you write code make the code simple and use this sample as the coding style to follow, do not use a single dependancy that I do not ask, do not use OOP, if you don't know something pause and ask me, make no mistakes, and do not create anything, follow my instructions strictly
customize it based on your preference and input this in the parameters so that every single prompt contain this

at least with human slop it was obviously terrible and you could imagine their two beain cells rubbing together to produce the O(n2) abomination, but with ai code it all looks sensible on the surface and then you read what's actually happening and it's worse. but dressed up really nice with documentation, like decorating turds with sprinkles

>we're back to counting lines of code to determine efficiency
the more things change the more they stay the same

>>109874451
Yes. Luddites are based. What is your point?

>>109874343
how do you think they get the money then?

>>109874614
jeeted, the code has been jeeted

>>109874386
OP is kind of right but you can work around it if you point the LLM to the sources and re-iterate the codebase every now and then with a fitting workflow.
Unfortunately it costs you so much tokens to re-iterate and loop over the codebase for performance improvements and less sloppy code.
You can't use lightweight models for code that must be clean and optimized.

Most of you are probably webslop or app fags so it doesn't really matter to you.
Try building a high performance game engine or whatever and find out yourself how often you have to tell the LLM to revisit, refactor and rewrite parts of the codebase.
It won't just code it with good optimizations form the start regardless of how "perfect" you think your initial prompt is.

>>109874649
they make no distinctions on the different technologies so they are subhuman retards who are unable to think for themselves and to determine objectively good or bad

>>109874343
Not my problem, won't ever notice because idk how to code or read code

>>109874694
Non-luddite cope

>>109874688
it is good at building blocks that you review and copy paste, that's how you use AI

>>109874706
truth is not cope and mistakenly thinking that the other is the one coping is a very large cope

>>109874729
Lol you cant code.

>>109874734
I do it for money since 2011

>>109874744
And Claude is still better than you? Rough.

>>109874391
It costs more money to do that, so yes.

>>109874753
what does that even mean, cringe

AI users who also know how to code, is it actually faster than just writing it yourself? It sounds frustrating to use outside of very specific issues which only require a few lines (or just an explanation) to solve.

Glad that I'm not the only one who doesn't understand code which was generated by Claude.
I hate ChatGPT, but that's code I can understand and follow through.

>>109874343
I feel like I'm missing something because I don't fully understand the LLM hype. If you know how to code then just wrap low-level/hard-to-read/verbose code in high level abstractions. Now your codebase reads like plain English especially with comments, you save money (no Claude subscription), you theoretically don't even need an internet subscription or extremely beefy GPU and you internalize how computers work on a deeper level than any vibe coder could.

>>109874343
Just let it do that, let it write the tests for it. Then open a new session and say

>remove all the bloated shitware slop, do not touch the tests at all.

>>109874343
The code Claude writes for me is like 50% comments, its annoying in that regard. But its pretty good at it, navigating large code bases, doing end to end testing on all its changes, handling git stuff in the background. This must be brutal for Jeet coders who have immediately been rendered obsolete.

>>109874841
In a lot of cases yeah, it's much faster. I wouldn't trust it with any core logic but for boilerplate or tests it's 100x faster than I can ever be, and I'm a pretty fast coder.
Not a replacement for a software engineer by any means yet.

>>109874343
>over-engineered
Too true. You need to prompt against the patchwork and make refactors frequently

>>109874841
For individual tasks not really, the main benefit for me is I can run 2/3 sessions at the same time and then clean up the slop once it's done. Basically the "do it yourself vs slop it and clean it up" tradeoff gets offset by the fact the slop can be generated in parallel while you wouldn't be able to do that manually.

>>109874343
Easy. Have Claude write a complex over-engineered tool to "tidy" all the code it makes. If a language does not already have a fully functional tidy tool then that language is incomplete and not production ready.

>>109874343
why does a code even need to be human readable. What, you're going to fix issues yourself one variable at a time?
>unmaintainable
prompt skill issue unironically