You must log in or register to comment.
2 monthsI don’t, and probably never will. A whole bunch of reasons:
- The current state of affairs isn’t going to last forever; at some point the fact that nobody’s making money with this is going to catch up, a lot of companies providing these services are going to disappear and what remains will become prohibitively expensive, so it’s foolish to risk becoming dependent on them.
- If I had to explain things in natural language all the time, I would become useless for the day before lunch. I’m a programmer, not a consultant.
- I think even the IntelliSense in recent versions of Visual Studio is sometimes too smart for its own good, making careless mistakes more likely. AI would turn that up to 11.
- I have little confidence that people, including myself, would actually review the generated code as thoroughly as they should.
- Maintaining other people’s code takes a lot more effort than code you wrote yourself. It’s inevitable that you end up having to maintain something someone else wrote, but why would you want all the code you maintain to be that?
- The use-cases that people generally agree upon AI is good at, like boilerplate and setting up projects, are all things that can be done quickly without relying on an inherently unreliable system.
- Programming is entirely too fun to leave to computers. To begin with, most of your time isn’t even spent on writing code, I don’t really get the psychology of denying yourself the catharsis of writing the code yourself after coming up with a solution.
- 2 months
You wrote this all a lot better than I could have, but to expand on 2) I have no desire whatsoever to have a “conversation” (nay, argument) with a machine to try and convince/coerce/deceive/brow-beat (delete as appropriate) it into maybe doing what I wanted.
I don’t want to deal with this grotesque “tee hee, oopsie” personality that every company seems to have bestowed on these awful things when things go awry, I don’t want its “suggestions”. I code, computer does. End of transaction.
People can call me a luddite at this point and I’ll wear that badge with pride. I’ll still be here, understanding my data and processes and writing code to work with them, long after (as you say) you’ve been priced out of these tools.
- 2 months
Here’s my opinion on those points, if people care.
- Indeed, that’s why I’d love for local models to be enough for my needs. Not yet the case but hopefully someday
- It’s already half my week. Explaining in natural language, to my coworkers, what they need to do. Either directly or through tickets
- I don’t think it’s always the case. IDE already have powerful tools but they don’t come close to the “understanding” an LLM can have on your code.
- The generated code is reviewed exactly like any other of my coworkers code. I check it does what it says on the tin and passes tests.
- All code ends up like other people’s code. You don’t always remember what something does a few months from now. Overall we read more code than we write, so the importance of the code being readable and easy to understand is the same as before. You’re still the one getting woke up a 4am, not the LLM.
- Not really true. I can implement a feature once and have it implemented and extended on all other services. The LLM mostly understand there are differences between them and work around that. Is it perfect? Ofc no, but I’ve never seen an automated tool you can just point at a file and say “do the same for x and y”.
- Programming still needs to be done by humans most of the time. LLM are just parrots so they cannot create some new API that a human previously had to create. I like making architectural decisions, organizing the code and structure it how I want. I don’t like having to edit 25 files to make some small adjustments. In the same way a “Refactor” button did not make us stop programming, LLM won’t either.
- 2 months
But in those 4 months AIs evolved a lot
Has it really? I don’t feel like it’s much different for programming compared to 4 months ago.
- 2 months
Yes, it evolved almost exponentially in these 4 months. It’s just bizarre what recent models can do and how consistent they do it.
If you never tried it, of course, you won’t know the difference. But for those who tried surely saw a huge improvement.
- 2 months
It’s not that I’ve never tried it, I’ve dabbled in it consistently over the last few years. If you had said there was a major difference compared to 2 years or maybe even a year ago, sure. In the last 4 months, I guess we’ve gotten stuff like Claude 4.6, which saw an increase in coding performance by 2.5% according to SWE benchmarks. An improvement, sure, but certainly not an exponential one and not one which will fix the fundamental weaknesses of AI coding. Maybe I’m out of the loop though, so I’m curious, what are those exponential improvements you’ve seen over the last 4 months? Any concrete models or tools?
- 2 months
I only ever started using and evaluating LLM agents this past December, but in my experience it kinda works now. I can’t ascertain exactly why and how, but I was taken away how a simple prompt could get workable results.
- 2 months
I decided to try Qwen 3.5 Plus via Qwen Code CLI (Gemini CLI fork) and it’s bizarre what it can do.
It can figure out when it’s struggling to something, look on the internet for questions and docs to understand things better. It takes a lot of actions by itself, not like that bad models from 4 months ago that gets stuck on endless thinking and tweaking and never fix anything.
Recent models are thinking each time more like human programmers.
- jkercher@programming.devEnglish2 months
I think you’re mistaking improvements in tooling as improvements in the LLMs. LLMs are plateauing. The idea of exponential growth is an illusion. We took 20+ year old technology, geared it toward text (the LLM), and trained it on the entire Internet. Then, it’s popularity grew exponentially.
This is the hype narrative that Altman, Dario, Jensen, etc. push. They are trying to convince everyone that what we have is the Model T Ford of AI. Just imagine where we’ll be in 6 months!
- 2 months
Of course.
My reasons for not using AI are the same as they were four months ago and will be the same in four months, regardless of what the models can or can’t do.
Ask again in four years.
- 2 months
What are your reasons?
The place you work don’t force you to use it?
I’ve been noticing all companies are forcing devs to use AIs to be more productive, even for simple things like write git commits.
- 2 months
I noticed how quickly my own skills started deteriorating when trying to work with it. I’m trying to build my skills, not outsource them.
I also don’t love the environmental impact, nor the immorality of how they got/get their training sets for the base models.
If my work tried to force me to use it, I would be looking to change employer. Or lie and say I use it. But our AI use is heavily regulated and generally disencouraged, so luckily no issues there.
- 2 months
I don’t think your code being used for training is a concern anymore. They’ll eventually keep finding new codes until it reaches its peak. Refusing to share your code for training will just postpone the inevitable, AI code will improve to its peak sooner or later.
- 2 months
You replied to only one of my points, and that’s not even what I said…
They train new models on base models, and I’m talking about how they scraped the internet without permission or how websites sold their users data without compensation and how no one was ever given any opportunity to opt out of sharing your work and your words to train these base models on.
Without that grand scale theft we would have no base models anywhere near what we have now.
I’m not opposed to willingly sharing, I’m opposed to profiting from stealing.
- 2 months
Your mistake is to think that I want to prove something, I don’t want to mention all your points, this is just a comment, not a scientific discussion.
Please, continue to “use AI daily”. Rot your brain, see if I care.
If my competitors want to shoot themselves in the foot that’s fine by me, I won’t stop them.
- 2 months
Wait, you guys still haven’t tried cocaine?
Using it you can work with much more energy and focus! You don’t get tired either!
- 2 months
Zero use. No need. Been doing this for 15+ years. Plus, if I don’t know how to do something I kind of want the mental reward from figuring it out myself.
I don’t, it’s not better than simply thinking about things myself. There isn’t institutional pressure to use it and if there was I would simply lie and not use it.
- 2 months
I have stopped using it, because the skill atrophy kicked in and I don’t want to turn into someone chatting with a bot every day.
I work as a software developer and over the last months, I slipped into a habit of letting ChatGPT write more and more code for me. It’s just so easy to do! Write a function here, do some documentation there, do all of the boilerplate for me, set up some pre-commit hooks, …
Two weeks ago I deleted my OpenAI account and forced myself to write all code without LLMs, just as I did before. Because there is one very real problem of excessive AI useage in software development: Skill atrophy.
I was actively losing knowledge. Sometimes I had to look up the easiest things (like builtin Javascript functions) I was definitely able to work with off the top of my head just a year ago. I turned away from being an actual developer to someone chatting with a machine. I slowly lost the fun in coding, because I outsourced the problem solving aspects that gave me a dopamine boost to the AI. I basically became a glorified copypaster.
- 2 months
I don’t. Personally, I don’t believe that AI assisted coding has a future, and I don’t like the quality of what it produces. Either we achieve AGI or whatever the finance bros are hyping up this week, or we don’t. If we do, then AI can write code without having a human in the loop, so “vibe coding” is dead. If we don’t, then AI code stays where its at right now, which is garbage quality. After a few years vibe coding disasters, demand for human coding will increase, and my skills will be much more valuable than even before all this craziness. In that case, letting those skills atrophy today would be the wrong move.
And if I’m wrong? Well, if AI code generation makes it so anyone who doesn’t know how to code can do it… then surely I’ll be able to figure it out? My existing skills wouldn’t hurt.
- 2 months
Online AI might crash, burn and go away
But open weight local models are here to stay and not going anywhere. We’re not going back to pure intellisense and simple tab completes
- 2 months
Not using it, not gonna use it. I prefer my skills to be improving, not growing reliant on a glorified “smart” copy-paste.
- 2 months
I use it at work because my colleagues only use it so it’s the only way I can deal with the LLM slop without total killing myself. And it’s horrendously bad still. I fucking hate it. Makes the worst fucking decisions.
I’m considering a career change honestly. I can’t stand this shit anymore.
- 2 months
Opus 4.6 Gemini 3 pro
Name em Ive tried em. Its all so sub par for anything beyond a one off script.
People have the impression they should be outputting 10x with them so they abuse them into doing more “thinking” Then they should.
Edit: it’s fucked the work culture more then it already was fucked when it comes to what defines software quality and expertise
- 2 months
Never used it. Don’t see any reason to. I just type stuff in my IDE. Works like a charm.
Most of the time I’m not writing large volumes or boilerplate code or anything, I’m making precise changes to solve specific problems. I doubt there’s any LLM that can do that more effectively than a programmer with real knowledge of the code base and application domain.
I also work on open source software and we haven’t seen a meaningful uptick in good contributions due to AI over the last few years. So if there’s some mythical productivity increase happening, I’m just not seeing it.
- Guttural@jlai.luFrançais2 months
Not just you, bud. We’ve seen the science confirm that the supposed productivity increases are a mirage.
