I’ve been in the industry for like 7-8 years and I’ve seen what “experienced conscious developers” write, past myself included. LLMs are not beating humans in writing slop, not even close. Median Sonnet output is more clean and sustainable than median output of a middle software engineer. And even seniors with 5+ yoe aren’t always doing better than that.
No sir, LLMs aren’t writing slop, it’s users that prompt them to write it this way 
You have no clue what “Decent” code is if you cranked that in 2 weeks with an LLM. Just being frank.
Yeah there are a lot of terrible programmers out there. Most of us are and don’t know it.
ok mr gatekeeper. @anon35988380
That guy is like saying “googling” stuff doesn’t help you program… just read the API docs only. AI is a tool, like any other. To not even use it even a little, means you just like wasting your time
Agree, but a reliance on just AI and not any code skills is dangerous. If someone moves too fast, like 3.5m lines in a few weeks, there is zero chance they know what that code is actually doing. And that is the risk, I use AI quite a bit - but I know exactly what its doing and how its doing it every step of the way. I don’t have to write every line, but I have looked at every line and what functions it generates and directed it into the specific patterns I want for something - knowing where the project is eventually headed.
With AI you don’t have to write modular and maintainable code, you don’t have to worry about DRY or maintaining a single concern. You can just shove everything together however it does - and AI being so fast and powerful can keep changing it as you need. Total rewrites can happen daily. So why worry about good coding practices when AI can brute force them? Well to me, the #1 reason is any time code changes you risk introducing a bug, a vulnerability, or just accidentally deleting functionality. Well designed and implemented code lets you make changes in a targeted fashion with minimal impact outside the planned scope.
I 100% agree. I have been spending the last hour going through AI generated code and cleaning it up. It was rewriting the same logic in dozens of methods instead of calling one centralized logic method… drives me crazy. It was doing a bunch of unnecessary conditionals making it hard to navigate what was going on if a human ever had to read it. Bad variable naming. AI can be super helpful, but if left to its own it can be so sloppy, and yea people are going to get themselves in trouble maintaining this stuff. But if AI is the one maintaining it… idk maybe that would work if you had enough money, but if humans still need to get in there and look around, AI code is often not the best. Now its better than probably a lot of mediocre or lazy programmers, so it’s going to be a godsend for them. But it is definitely a tool for me. I can’t afford to not know what is going on in my code. Maybe if I worked for some large company and the responsibility was diffused and I could kick the can down to the next guy when I move on, maybe, but if you own a project, AI slop is a real concern.
You’re experiencing the exact same thing as me. Developers who have only ever written and seen mediocre at best code, will see what AI does and think is exceptional because its much better than what they’ve written or worked with.
When you cut your teeth in a world where your IDEs couldn’t do autocompletion or syntax highlighting, didn’t have linters and formatters, and for the web - typescript was still 15 years away. You had no choice but to write code you could easily maintain and understand - and reuse. You didn’t have time to rewrite the same logic every time, and you sure didn’t have memory or bandwidth to duplicate anything.
Maybe I’m just an old has been for thinking any of that stuff still matters. Just money and electricity at the problem, and it doesn’t exist anymore.
I mean at some point it may not matter if a human could ever read the code. Maybe it won’t even be human readable.
I am over 65 years old started coding on punch cards and have seen it all.
whats clear is there has been very little oversight on the quality of code these models are being trained on
If this becomes a reciprocal downward spiral due to AI training on AI generated code,
there becomes an opportunity for creating a better code curation dataset for the entrepreurs out there, lets hope this happens soon.
Agree! I’m not over 65, but I wrote my first line of code in 1989. I can definitely tell AI has not been trained on the kind of code thats been expected of me. It knows the patterns and concepts but doesn’t know how to write them unless you load your context with examples of them - and it definitely wont use them on its own spontaneously.
Considering if privacy mode is off, Cursor can train on your code - that means Cursor is getting trained on a LOT of AI generated code. Not sure how that turns out.
Look, I think it’s going to taper out. The issue is not with AI generating garbage. The issue is with a lot of folks jumping on the bandwagon to quickly pump out a product that can make them money. But as I said above no one can argue that AI can generate pretty decent boilerplate code and get one understanding complex systems in a much shorter time.
It will be a tool that literally every programmer uses on a daily basis and the people just chasing the “new thing” will get bored.
No company can survive on bad code, yet literally every company uses tools like Cursor now.
I don’t know anyone that will still write each and every line of code themselves anymore, why? Because AI makes the job so much easier. Explaining things etc. Claude 4 really nails page designs if your prompt it right. Like genuinely beautiful, designed logins and dashboards that looks amazing in seconds.
Why won’t I use a tool in my arsenal, just because people say AI produces slop…
The reason of price skyrocketing is quite simple: models are burning infinite amount of tokens to think until they hit the limit of your plan.
AI is dumb, in reality a self-regulating token spending AI would solve it. But indeed until then… They all sell smokescreens hoping this to happen.
I may have some start of answers, but I will not give it to any of this big tech companies, I myself will make my own
try and create new chats often when there is an opportunity to break into new context areas of concern. it seems the longer the chat goes, the higher the token rate usage in some cases.
“No company can survive on bad code” is something someone who’s never worked in the industry would say. You have NO IDEA how “bad” most legacy code actually is..
totally agreed, ■■■■ solution often brings much more value than the perfect one if it’s “good enough”. Hell, the whole MVP concept is about delivering a barely living and often not maintainable feature as fast and as cheap as possible. A huge chunk of the industry LIVES on bad code.
And you know what now that I’m thinking about it, it makes sense that the business models of these LLMs is failing. It’s probably difficult to design a half-■■■, barely performant piece of software that’s just getting by as an MVP, versus recursively “thinking” and “considering” the most efficient, best-practice way to do it. LOL. Peak human ingenuity
Wow this ‘bad code’ problem is far worse than i imagined.
Back in the mid 80’s when i was writting code for Hospital’s we used to meet and have code walk through’s examining code structure as a team, to try and optimize every cpu cycle there was, which was nanoscopic by todays standards, code effiecency was important - it still is.
The machines required a lot of respect from coders and analysts. I have written a lot of code, and none of it was bad code. Glad I am not part of the “Industry”
are you one of them idiots who overengineer and overoptimize meaningless stuff?
your “back in my days” flex doesn’t work here, not everything needs to be optimized to its limits. perfect systems are VERY expensive and VERY slow to develop.
engineer’s task is to solve problems, and you are pleasing yourself with perfecting code instead of solving them, apparently…