The number of questions on Stack Overflow fell by 78 percent in December 2025 compared to a year earlier. Developers are switching en masse to AI tools in
Also depends on your level of expertise. If you have beginner questions, an LLM should give you the correct answer most of the time. If you’re an expert, your questions have no answers. Usually, it’s something like an obscure firmware bug edge case even the manufacturer isn’t aware of. Good luck troubleshooting that without writing your own drivers and libraries.
Yeah but in that edge case SO wouldn’t help either even before the current crash. Unless you were lucky. I find LLM useful to push me in the right direction when I’m stuck and documentation isn’t helping either not necessarily to give me perfectly written code. It’s like pair programming with someone who isn’t a coder but somehow has read all the documentation and programming books. Sometimes the left field suggestions it makes are quite helpful.
I’ve found some interesting and even good new functions by moaning my code woes to an LLM. Also, it has taken me on some pointless wild goose chases too, so you better watch out. Any suggestion has the potential to be anywhere from absolutely brilliant to a completely stupid waste of time.
If you’re writing cutting edge shit, then LLM is probably at best a rubber duck for talking things through. Then there are tons of programmers where the job is to translate business requirements into bog standard code over and over and over.
Nothing about my job is novel except the contortions demanded by the customer — and whatever the current trendy JS framework is to try to beat it into a real language. But I am reasonably good at what I do, having done it for thirty years.
Yeah the internet seems to think coding is an expert thing when 99.9% of coders do exactly what you described. I do it, you do it, everybody does it. Even the people claiming to do big boy coding, when you really look at the details, they’re mostly slapping bog standard code on business needs.
Boring standard coding is exactly where you can actually let the LLM write the code. Manual intervention and review is still required, but at least you can speed up the process.
Code made up of severally parts with inconsistently styles of coding and design is going to FUCK YOU UP in the middle and long terms unless you never again have to touch that code.
It’s only faster if you’re doing small enough projects that an LLM can generate the whole thing in one go (so, almost certainly, not working as professional at a level beyond junior) and it’s something you will never have to maintain (i.e. prototyping).
Using an LLM is like giving the work to a large group of junior developers were each time you give them work it’s a random one that picks up the task and you can’t actually teach them: even when it works, what you get is riddled with bad practices and design errors that are not even consistently the same between tasks so when you piece the software together it’s from the very start the kind of spaghetti mess you see in a project with lots of years in production which has been maintained by lots of different people who didn’t even try to follow each others coding style plus since you can’t teach them stuff like coding standards or design for extendability, it will always be just as fucked up as day one.
That is a bit … overblown. If you establish an interface, to a degree you can just ignore how the AI does the implementation because it’s all private, replaceable code. You’re right that LLMs do best with limited scope, but you can constrain scope by only asking for implementation of a SOLID design. You can be picky about the details, but you can also say “look at this class and use a similar coding paradigm.”
It doesn’t have to be pure chaos, but you’re right that it does way better with one-off scripts than it does with enterprise-level code. Vibe coding is going to lead people to failure, but if you know what you’re doing, you can guide it to produce good code. It’s a tool. It increases efficiency a bit. But it also don’t replace developers or development skills.
Also depends on your level of expertise. If you have beginner questions, an LLM should give you the correct answer most of the time. If you’re an expert, your questions have no answers. Usually, it’s something like an obscure firmware bug edge case even the manufacturer isn’t aware of. Good luck troubleshooting that without writing your own drivers and libraries.
Yeah but in that edge case SO wouldn’t help either even before the current crash. Unless you were lucky. I find LLM useful to push me in the right direction when I’m stuck and documentation isn’t helping either not necessarily to give me perfectly written code. It’s like pair programming with someone who isn’t a coder but somehow has read all the documentation and programming books. Sometimes the left field suggestions it makes are quite helpful.
I’ve found some interesting and even good new functions by moaning my code woes to an LLM. Also, it has taken me on some pointless wild goose chases too, so you better watch out. Any suggestion has the potential to be anywhere from absolutely brilliant to a completely stupid waste of time.
If you’re writing cutting edge shit, then LLM is probably at best a rubber duck for talking things through. Then there are tons of programmers where the job is to translate business requirements into bog standard code over and over and over.
Nothing about my job is novel except the contortions demanded by the customer — and whatever the current trendy JS framework is to try to beat it into a real language. But I am reasonably good at what I do, having done it for thirty years.
Yeah the internet seems to think coding is an expert thing when 99.9% of coders do exactly what you described. I do it, you do it, everybody does it. Even the people claiming to do big boy coding, when you really look at the details, they’re mostly slapping bog standard code on business needs.
Boring standard coding is exactly where you can actually let the LLM write the code. Manual intervention and review is still required, but at least you can speed up the process.
Code made up of severally parts with inconsistently styles of coding and design is going to FUCK YOU UP in the middle and long terms unless you never again have to touch that code.
It’s only faster if you’re doing small enough projects that an LLM can generate the whole thing in one go (so, almost certainly, not working as professional at a level beyond junior) and it’s something you will never have to maintain (i.e. prototyping).
Using an LLM is like giving the work to a large group of junior developers were each time you give them work it’s a random one that picks up the task and you can’t actually teach them: even when it works, what you get is riddled with bad practices and design errors that are not even consistently the same between tasks so when you piece the software together it’s from the very start the kind of spaghetti mess you see in a project with lots of years in production which has been maintained by lots of different people who didn’t even try to follow each others coding style plus since you can’t teach them stuff like coding standards or design for extendability, it will always be just as fucked up as day one.
That is a bit … overblown. If you establish an interface, to a degree you can just ignore how the AI does the implementation because it’s all private, replaceable code. You’re right that LLMs do best with limited scope, but you can constrain scope by only asking for implementation of a SOLID design. You can be picky about the details, but you can also say “look at this class and use a similar coding paradigm.”
It doesn’t have to be pure chaos, but you’re right that it does way better with one-off scripts than it does with enterprise-level code. Vibe coding is going to lead people to failure, but if you know what you’re doing, you can guide it to produce good code. It’s a tool. It increases efficiency a bit. But it also don’t replace developers or development skills.