Devs gaining little (if anything) from AI coding assistants
Code analysis firm sees no major benefits from AI dev tool when measuring key programming metrics, though others report incremental gains from coding copilots with emphasis on code review.
For me, it is a glorified auto-complete function. Could definitely live without it.
Same for me, but that glorified auto complete helps a lot.
Hell yea. Our unit test coverage went way up because you can blow through test creation in second. I had a large complicated migration from one data set to another with specific mutations based on weird rules and GPT got me 80% of the way there and with a little nudging basically got it perfect. Code that would've taken a few hours took about 6 prompts. If I'm curious about a new library I can get a working example right away to see how everything fits together. When these articles say there's no benefit I feel people aren't using these tools or don't know how to use them effectively.
We always have to ask what language is it auto-completing for? If it is a strictly typed language, then existing tooling is already doing everything possible and I see no need for additional improvement. If it is non-strictly typed language, then I can see how it can get a little more helpful, but without knowledge of actual context I am not sure if it can get a lot more accurate.
Good devs gain little.
I gain a lot.
Its basically a template generator, which is really helpful when you're generating boilerplate. It doesn't save me much if any time to refactor/fill in that template, but it does save some mental fatigue that I can then spend on much more interesting problems.
It's a niche tool, but occasionally quite handy. Without leaps forward technically though, it's never going to become more than that.
Just beware, sometimes the AI suggestions are scary good, some times they’re batshit crazy.
Just because AI suggests it, doesn’t mean it’s something you should use or learn from.
Feel the same way!
Generative AI is great for loads of programming tasks like helping create regular expressions or syntax conversions between languages. The main issue I've seen in codebases that rely heavily on generative AI is that the "solutions" often fix today's bug while making future debugging more difficult. Generative AI makes it easy to go fast in the wrong direction. Used right it's a useful tool.
While I am not fond of AI, we do have access to it at work and I must admit that it saves some time in some cases. I'm not a developer with decades of experience in a single language, so something I am using AI to is asking "Is it possible to do a one-liner in language X where it does Y?" It works very well and the code is rarely unusable, but it is still up to my judgement whether the AI came up with a clever use of functions that I didn't know about or whether it crammed stuff into a single unreadable line.
It introduced me to the basics of C# in a way that traditional googling at my previous level of knowledge would've made difficult.
I knew what I wanted to do and I didn't know what was possible or how to ask without my question being closed as a duplicate with a link to an unhelpful post.
In that regard, it's very helpful. If I had already known the language well enough, I can see it being less helpful.
Great for Coding 101 in a language I'm rusty with or otherwise unfamiliar.
Absolutely useless when it comes time to optimize a complex series of functions or upgrade to a new version of the .NET library. All the "AI" you need is typically baked into Intellisense or some equivalent anyway. We've had code-assist/advice features for over a decade and its always been mid. All that's changed is the branding.
This is what I've used it for and it's helped me learn, especially because it makes mistakes and I have to get them to work. In my case it was with Terraform and Ansible.
Haha, yeah. It really loves to refactor my code to "fix" bracket list initialization (e.g. List<string> stringList = [];) because it keeps not remembering that the syntax has been valid for a while.
It's newest favorite hangup is to incessantly suggest null checks without asking if it's a nullable property that it's checking first. I think I'm almost at the point where it's becoming less useful to me.
I learned bash thanks to AI!
For years, all I did was copy and paste bash commands. And I didn't understand arguments, how to chain things, or how it connects.
What about just reading the documentation?
I'm a penetration tester and it increases my productivity a lot
Penetration tester, huh? Sounds like a fun and reproductive job.
But it can be very HARD sometimes
I mainly use AI for learning new things. It’s amazing at trivial tasks.
so it's a vector of attack?
Everyone keeps talking about autocomplete but I've used it successfully for comments and documentation.
You can use vs code extensions to generate and update readme and changelog files.
Then if you follow documentation as code you can update your Confluence/whatever by copy pasting.
I also use it a lot for unit tests. It helps a lot when you have to write multiple edge cases, and even find new one at times. Like putting a random int in an enum field (enumField = (myEnum)1000), I didn't knew you could do that...
Yeah. I've found new logic by asking GPT for improvements on my code or suggestions.
I cut the size of a function in half once using a suggested recursive loop and it blew my mind.
Feels like having a peer to do a code review on hand at all times.
My main use is skipping the blank page problem when writing a new suite of tests—which after about 10 mins of refactoring are often a good starting point
No shit. Senior devs have been saying this the whole time. AI, in its current form, for developers, is like handing a spatula to a gourmet chef. Yes it is useful to an extremely small degree, but that’s it…for now.
A convoluted spatula that sometimes accidentally cuts what your cooking im half instead of flipping it and consumes as much power as the entirety of Japan.
It's when you only have a pot and your fingers that a spatula is awesome. I could never bother finish learning C and its awkward syntax. Even though I know how to code in some other language, I just couldn't write much C at all and it was painful and slow. And too much time passed between attempts that I forgot most of it in between. Now I can easily make simple C apps, I just explain the underlying logic, with example of how I would do it in my preferred language and piece by piece it quickly comes together and I don't have to remember if the for loop needs brackets of parenthesis or brackets nor if the line terminator is colon or semi colon.
The problem is that you’re still not learning, then. Maybe that’s your goal, and if so, no worries, but AI is currently a hammer that everyone seems to be falling over themselves finding nails for.
All I can do is sigh and shake my head. This bubble will burst, and AI will still take decades to get to the point people think it is already at.
And yet, higher ups continue to lay off more devs because AI "is the future".
In my experience, most of the tech layoffs have been non-devs. PMs and Designers have been the hardest hit and often their roles are being eliminated.
I mean, I'm a dev who got laid off almost a year ago and still can't find anything. I know tons of others who are in similar positions. So...
The writer has a clear bias and a lack of a technical background (writing for Techies.com doesn't count) .
You don't have to look hard to find devs saving time and learning something with AI coding assistants. There are plenty of them in this thread. This is just an opinion piece by someone who read a single study.
if you are already competent and you are aware that it doesn't necessarily give you correct information, the it is really helpful. I know enough to sense when it is making shit up. Also it is, for some scenarios, faster and easier then looking at a documentation. I like it personally. But it will not replace competent developers anytime soon.
This opinion is a breath of fresh air compared to the rest of tech journalism screaming "AI software engineer" after each new model release.
Places GPT-based "AI" next to flying cars
Flying cars exist, they’re just not cost effective. AFAICT there’s no GPT that is proficient at coding yet.
As far as I know, right now the main problem with flying cars is that they are nowhere near as idiot-proof as a normal car, and don't really solve any transportation problem since most countries' air regulations agencies would require them to exclusively take off and land in airports... Where you can usually find tons of planes that can go much further (and are much more cost effective, as you pointed out)
Every now and then, GitHub Copilot saves me a few seconds suggesting some very basic solution that I am usually in the midst of creating. Is it worth the investment? No, at least not yet. It hasn't once "beaten" me or offered an improved solution. It (more frequently than not) requires the developer to understand and modify what it proposes for its suggestions to be useful. Is is a useful tool? Sure, just not worth the price yet, and obviously not perfect. But, where I'm working is testing it out, so I'll keep utilizing it.
lol Uplevel's """full report""" saying devs using Copilot create 41% more bugs has 2 pages and reads like a promotional material.
you can download it with a 10 minute email if you really want to see for yourself.
just some meaningless numbers.
It's just fancier spell check and boilerplate generator
It's great as essentially a StackOverflow that I can talk to in real time. But as with SO, I've still got to figure out what pieces are legit and where they go.
AI search results made stack overflow answers harder to find now lol
It's definitely exploded but content farms were a problem even before 2022. There's a reason google results starting with "reddit" / "stack overflow" were trending so hard.
I honestly stopped using it after a week
Who are those guys they keep asking this question over and over ? And how are they not able to use such a simple tool to increase their productivity ?
I partly disagree, complex algorithms are indeed a no, but for learning a new language it is awesome.
Currently learning Rust and although it cannot solve everything, it does guide you with suggestions and usable code fragments.
Highly recommended.
Is there anything it provided you so far that was better than the guidance from the Rust compiler errors themselves? Every error ends with "run this command for a tutorial on why this error happened and how to fix it" type of info. A lot of times the error will directly tell you how to fix it too.
I agree, although some messages are still cryptic for a newbie like me, but thats maybe more the person on the chair than the compiler 😇.
I'd estimate copilot to be correct in only 10% of the time, solving a situation like that. Most of the time the solution suggested is also wrong, but just differently.
Having said that: sometimes (small chance, 1% maybe) the solution is spot on.
AI mainly helps with the initial syntax and on language constructs and for that it is awesome.
Currently learning Rust and although it cannot solve everything, it does guide you with suggestions and usable code fragments.
as does the compiler and the rust book
Yep, by definition generative AI gets worse the more specific you get. If you need common templates though, it’s almost as good as today’s google.
… which is not a high bar.
I get more benefit from a good IDE that helps me track libraries, cars, functions, grammar checks my code, offers a pop-up with params and options....
I don't needcode I would grade as a D- from an AI. Most of what I write comes from my code closet anyway. I have skeleton code for so much, and I trust my old code more than AIs new code
Garbage in garbage out is how they all work if you give it a well defined prompt you can get exactly what you want out of it most of the time but if you just say fix this problem it’ll just fix the problem ignoring everything else
You're holding it wrong
Wait till this guy learns how a theremin work.
It has some uses.
But I'm waiting for a good self hosted model and to have a more powerful gpu to properly run it.
I’m shocked. There must be an error in this analysis. /s
Maybe engage an AI coding assistant to massage the data analysis lol
I like to use suggestions to feel superior when trash talking the generated code
Devs that are punching above their class, however, probably get great benefit from it. I would think it’s also an OK learning tool, except for how inaccurate it can be sometimes.
I use it occasionally. Recently I used it to convert a written specification in a document to a java object. And it was like 95% correct - but having to manually double check everything and fix the errors eliminated much of the time savings.
However that's a very ideal use case. Most often I forget it exists.
I use it a fair bit. Mind, it's something like formating a giant json stdout into something I want to read...
I also do find it's useful for sketching out an outline In pseudo code.
Claude is my coding mentor. Wouldn't want to work without it.
I run code snippets by three or four LLMs and the consensus is never there. Claude has been the worst for me.
Which one has been best? I’m only a hobbyist, but I’ve found Claude to be my favorite, and the best UI by a mile.
I use it as second last resort, and in those times, it did worked out. I had to test, verify, and make changes. Even so, I avoid using them.
I truly don't understand the tendency of people to hate these kinds of tools. Honestly seems like an ego thing to me.
Having to deal with pull requests defecated by “developers” who blindly copy code from chatgpt is a particularly annoying and depressing waste of time.
At least back when they blindly copied code from stack overflow they had to read through the answers and comments and try to figure out which one fit their use case better and why, and maybe learn something... now they just assume the LLM is right (despite the fact that they asked the wrong question and even if they had asked the right one it'd've given the wrong answer) and call it a day; no brain activity or learning whatsoever.
Typical lack of nuance on the Internet, sadly. Everything has to be Bad or Good. Black or White. AI is either The best thing ever™ or The worst thing ever™. No room for anything in between. Considering negative news generates more clicks, you can see why the media tend to take the latter approach.
I also think much of the hate is just people jumping on the AI = bad band-wagon. Does it have issues? Absolutely. Is it perfect? Far from it. But the constant negativity has gotten tired. There's a lot of fascinating discussion to be had around AI, especially in the art world, but God forbid you suggest it's anything but responsible for the total collapse of civilisation as we know it...
Its really weird.
I want to believe people arent this dumb but i also dont want to be crazy for suggesting such nonsensical sentiment is manufactured. Such is life in the disinformation age.
Like what are we going to do, tell all Countries and fraudsters to stop using ai because it turns out its too much of a hassle?
Also, when a tool increases your productivity but your salary and paid time off don't increase, it's a tool that only benefits the overlords and as such deserves to be hated.
To be honest ChatGPT pretty much killed the fun of programming.
What?
Programming was like a challenge, you have a problem and you need to solve it. You look into the internet, stack overflow, test different chunks of codes, reading documentation, etc. nowadays is simply splitting one problem into pieces, and then copy pasting.
Judging this article by it's title (refuse to click). Calling BS. ChatGPT has been a game changer for me personally
I'm fine with searching stack exchange. It's much more useful. More info, more options, more understanding.