I’m just curious about the spread in our community.
Answers are anonymous.
- Yes (Professional)
- Yes (Hobbyist)
- No (Professional)
- No (Hobbyist)
0
voters
1 Like
I try to use it and generally think, I would have been faster, if I just did it my self without LLM.
Maybe if I would use a team of agents with different roles but I loath the upfront setup work.
Also I’m more of a sysadmin then a coder.
3 Likes
i’m thinking about it. started chatting with LLM a bit. LLM can give better responses than people sometimes, but i’d much rather prefer working alongside someone
so far, chats take a ton of effort to structure. some questions require none, but the issues i’m hung up on don’t have great answers. LLM tends to respond with lists, a few superficial justifications for each item, another list… it takes a ton of time to get past all the basic advice. it can’t really process your network’s specific needs. there’s quite a bit of misleading information, but it’s usually fairly correct.
gemini in search has been helpful sometimes. whether search or chat, the burden pretty quickly shifts to you to do the work
i don’t like the whole idea of agent-based development. using it correctly would be highly varied, depending on the project/context. if i ended up with 1,000 projects… great? right?
1 Like
Nope, not even once 
5 Likes
Yeah, it’s fantastic to generate a lot of code very fast – until you realize, that every line of code is a liability…
2 Likes
It took me a while to get to a satisfactory state (started to chat first, tried github copilot, then pycharm integrated AI tools.. not satisfied.. until I tried the right harness, the sort of framework that drives the LLM coding.. now it’s fantastic, wouldn’t go back).
The harness I use most is Claude Code (you can use it with free LLM models, not only with Anthropic ones). The key is to clearly specify what you expect in one or several well-designed Markdown files, then it’s just magic..
..if you want to get something done quickly for a certain task that you don’t want / have time to code yourself. SO many ‘little’ tools that now have a web interface, cool practical functionality etc. that I would never have spent the time doing myself.
I could go on an on about that.. but I think my point is: give a good harness a try! There are many alternatives. I use Claude Code because it’s the one with the most material out there to not start from scratch and learn how things work the hard way.
1 Like
I voted No/No.
No (Professionally) *
My employer expects me to use AI. So I do. However, I use it with a twist:
- I implement the new features or fix bugs by hand.
- Before my colleagues have a look at my code, I have (any) model review my code.
I’ve noticed that for almost all tickets that I implement, I already know what I want to implement. The hard part is already done, the architecture is clear. I just write code top-to-bottom and I’m done. Reading code I’ve written myself is also quite easy, since I saw at least every line of code twice: once in my mind, once on screen. And probably a third time while refactoring some names.
But if a LLM/agent creates the code, everything grinds to a halt. The initial review is easy. And then, a requirement changes, since the PM (or we devs!) suddenly realize that the feature is not enough and a tiny[1] change could help. You’re LLM-locked at that point: you don’t know the code by heart, so you ask the LLM to change it, which forces you to review almost all of it again. In total, this takes more time than just writing it myself.
So, why the * in the header? Because when it’s a one shot script, I may use a vibe coded 80 line script. It’s easy to read, I can grab a coffee, and I’m not going to be the maintainer of that script for the next 12 months. Instead, I will be the one who will ./temp-script.sh && rm ./temp-script.sh to ensure that it’s really done-and-gone.
No (Hobbyist) *
I’m too stingy to pay for AI, too privacy oriented and paranoid to use free online ones, and my PC is not strong enough for AI. So no, I’m not using AI for any hobby project.
But I may (*) occasionally use company resources to have my open source hobby projects reviewed by a LLM. My private .emacs.d got a lot of LLM review lately. Lots of hallucinations, too, but I found it nice to see some blind spots that I would usually miss.
Still, coding is for me a way to relax, so I’m not going to hand that over to a machine 
.
Yes! (Professionally, non-code)
You asked about Code. Nope for code. But a big “Yes!” for many text things at work. Jira tickets, Confluence searches, text editing.
Keep in mind that “big yes” still boils down to less than 1h/week. But I may be an outlier. I only work part-time and still get my things done 
.
actually tiny, not “PM tiny” ↩︎
1 Like
At this point, I wonder whether you’d even be able to change your Emacs configuration without an LLM 
1 Like
Funny, how you see the same problems with LLMs.
Good for reviews, not really worth the money, vibe coding creating more work in the end.
The only use I found in terms of code generation is to generate the boilerplate for Ansible playbooks, Icinga/Nagios plugins and unit tests for the plugins. As you wrote we’re at around 100 lines of code and I still get intern quality that needs review. So no real time savings but a way around the white canvas syndrome 
I also us it to proof read my business communication as approved by my employer.
1 Like
I wouldn’t call it coding, but I am finally getting further along my coding learning journey then ever. Claude is my tutor and does help me fix the inevitable tool chain woos that aren’t covered in learn to code books.
1 Like
What kind of controls does an employer put in place regarding tickets?
What effect does this have on how users add metadata to tickets? (tags, categories, references, etc) I.e. do users not add as much context if they expect AI to provide that to other users of a system?
I find it to be a good reference-tool. If I’m also in need of an implementation it does well for variety of things, but after all there is no free lunch and the machine’s performance can drop fast, although it does seem to get better every release cycle.
1 Like