How much of your coding is done by AI coding agents these days?
Displaying poll results.23561 total votes.
Computer Science is the only discipline in which we view adding a new wing to a building as being maintenance -- Jim Horning
Computer Science is the only discipline in which we view adding a new wing to a building as being maintenance -- Jim Horning
I need to know what every line does (Score:2)
I'm basically a lead senior ... (Score:3)
... to my AI metasubscription now. AI does what I ask it to do, I just review the changes and commit. It's like having a personal team of 10-20 experts sitting in a chat just ready to do my bidding. It's not sitting but it doesn't feel like that too often yet.
However it's quite staggering to watch am AI so your job an order of magnitude better than yourself. And that for a bunch of software stacks a human couldn't dream to comprehend. It's also sobering to watch the value-add chains I'm supposed to automate with code being voided entirely by AI. Not only is my job gone, the context with which it makes sense is also rapidly vanishing. You should see the look on the face of the lawyers I work with when the realize how AI does away with them too.
I'm very likely going to leave my current team. I'm in the process of leaving classic Web software development as a day job. ... You guys can't imagine how glad I am not having just software and the Web as my only field of experience and expertise.
Re: (Score:3)
AI doesn't do away with lawyers. In fact, several have tried submitting hallucinated AI reports to the courts and gotten penalized for it.
Doing your basic due diligence is still a thing. And you can't rely on AI to lawyer you out of a bad situation. IDK how you can pass the bar and then fail so hard, on something so basic, as doing proper research before submitting something important to a judge. That's literally what staff and interns are for.
https://www.businessinsider.co... [businessinsider.com]
Re: (Score:2)
Hit and miss (Score:2)
The 2nd attempt was with a very obscure communicatio
Very hit and miss (Score:2)
Simple ten-liners in python or bash are almost idiot proof.
But mathy things like transformations between different kinds of coordinate systems in c or c++ are at best equivalent to a college intern who may or may not have gone on a bender the night before. Wrong sense of the transformation (sometimes switching bacl and forth through the calculation), numerical approximations where analytical form was requested, occasionally just plain wrong, frequently has to be corrected.
May work better if trained on our c
Not good enough for coding (Score:1)
I tried using AI a few times, but the code quality was too low for my taste. In the end, it's faster for me to just write the code than to ask an AI to do it and then to fix its bad code.
However, AI is still a blessing. When I have a specific question, like about frameworks that I know only superficially, it's certainly faster to just ask an AI than to search the web. I don't always get a correct answer, but even with the time wasted trying things that are wrong, overall I still save a lot of time. AI basic
Re: Not good enough for coding (Score:1)
Same. I keep trying it and the new models are getting worse than the older models so far. Going to try more tomorrow but right now it makes so many errors I canâ(TM)t use them for any coding.
I wish it could code for me. Would be fun.
Re: (Score:2)
Conditions for experimenting with them (Score:3)
For me to even think about trying LLMs, I really need one which:
Is not actively dishonest, so it contains no programming for flattery, confidence, automatic agreement or other pretences not explicitly requested by the user.
Does not use any datacentres forced on communities or which are otherwise doing harm to them, or whose electricity and/or water consumption are secret, or which use resources gained by underhanded means or simple theft.
Is not trained on material without the consent of the people who made that material.
Is not being offered on a "cheap now, expensive later" model.
"Be the change you wish to see in the world" and all that.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
I do have to produce code as part of my job, though as is pretty typical in the industry I spend far more time reading code than writing it. But...
Re: (Score:2)
That's an idealist but somewhat pointless list of requirements to keep you from ever trying a tool.
Is not actively dishonest, so it contains no programming...
That's pretty much based on the training set and the prompt. You can prompt it to behave how you like
Does not use any datacentres forced on communities
Not really possible to know where a model was trained so an easy excuse to ignore it.
Is not trained on material without the consent of the people who made that material.
Another impossible requirement. All models will include material from public domain or licensing deals that won't explicitly include the people who made the material
Is not being offered on a "cheap now, expensive later" model.
At least this one is somewhat possible. Obviously it exclude
Learning (Score:1)
I've used some AI thing to learn how to improve my amateur coding skills. I study other peoples code and this is just another part of that.
The professional coding I leave to the professionals. They think, design, refactor and fix projects. If the AI stuff can save you a little time here and there, all the better. I'd want to use local modals if any though.
Same for my professional vocation in audio. I use AI tools where it makes sense and saves time. I often use it for noise suppression, voice resynthesis, t
Cowboy Neal (Score:1)
No choice (Score:1)
Punchcards (Score:2)
Re: (Score:2)
6800, 8080/8085 , Z80, 68HC11, 6502 and the various variants of these models.
Re: (Score:2)
Re: (Score:2)
Goes from calculator sized one with Basic installed though to a PDP11/23
Tons of manuals / books/ etc to go with them
Re: (Score:2)
My cue? My latest AI project had a punch card phase. At one point the entire thing was on Hollerith cards, with the program in PL/C (a dialect of PL/I) and about a thousand cards of data. It had originally been a simple typewritten list derived from a handwritten list. Also in the early 80s it became a database for the first time, ported to bigger and better computers and operating systems later... Probably late in the 90s it got a PERL/CGI front end interface (ultimately based on some code the late James L
This is a fairly shallow question for a poll (Score:1)
Given the health and cost problems caused by AI Datacenters built in residential centers, I personally don't use AI for my own stuff. If I do, I prefer something on a local device. But with how chatbot AI is being force into everything so much now it is getting harder and harder to ignore it. xcode had an AI button that would just post errors not to long ago... prominently displayed so you can't ignore it. It's put into that spot you will accidentally click it.
There are those that say that AI is the ine
we *clap* are *clap* the eleven per-cent! *clap!* (Score:1)
Missing Option (Score:2)
All of it, until the tool we were ordered to use runs out of tokens. Yeah, that's accelerating development.
its "limitations" (Score:1)
AI's going to start doing everyone's coding soon (though I was originally a
Re: its "limitations" (Score:2)
THIS. I had pounded out so many for loops. And the opportunity to solve actually hard algorithm and data structure problems were few and far between. This has given new life to my career by taking out all the stuff I have typed a thousand times. I focus on the unique issues that LLM does not do as well with.
Re: its "limitations" (Score:2)
Currently i have to work with a program that doesn't work and costs 10K/ license/year. I've now created my own version that has global search, instant import, very good UI and instant save instead of a crashing heap of garbage. 20K lines of code, no imports. Sub second performance.
Yeah AI is here to stay and you can pry it out of my cold dead hands.
Missing option (Score:2)
Cowboy Neil writes my code.
Coding isn't my primary, but I do it. Good policy. (Score:2)
I'm not on the engineering team at my company, but I do a little bit of coding as part of my job.
My company has a good AI coding policy: "You are responsible for all code you check in. You are encouraged to use AI responsibly, but you must check it for errors. If you turn in code that is bad, you can't blame the AI."
Plus, while we are trying out a few different AI platforms, we have token limits (I don't know what they are, because in my role I never come close to hitting them - but company-wide I know our
Agents are good at finding bugs, not writing code (Score:2)
I've had no success getting Agents to help write code for SQLite. All the code I've asked Claude to write for me is slop that didn't work or worked poorly and never got committed. On the flip side, I do ask Claude to review the code that I write prior to committing, and it is good at that, often finding serious oversights that I had missed.
This observation, that agents are good at code review but bad at writing original code, seems to carry through into AI-generated bug reports coming into SQLite from