Dear GPT, dear Gemini,
cher Claude, cher Copilot,
thanks for wasting my time
with your contradictory advice!
I could as well
roll a dice.

I was about to cancel my copilot which I've been paying for since beta days. But its become a lot beter recently, contextual multichanges in docker files, its getting better at scaffolding too. Still dont think its worth the money but slightly impressed making me second guess canceling.

@rootshell so they got better at persuading you to not cancel the subscription ^^

My advice is to not accept advice from any AI

@Lensflare basically. It now does stuff I normally spent too much time automating for little gains

@Lensflare That's catch 22 right there

@rootshell My god, ofcourse it's worth it. I was adding crud functions to a class, after finishing the first one, i select the whole class and order to build the rest. Done. I cou;dn't live without AI anymore :P

@retoor that's exactly the kinda thing I used to automate if the framework of choice didnt have tooling for it yet. 😂
AI be taken me jerb, now I cant automate the automation. 😭

Man I remember a company with graphql.
-> entity definition (ORM)
-> graphql definition in yaml
-> graphql definition in php

Had to effectively define an entity thrice. Stuff like AI would make that somewhat bearable I guess.

@retoor @rootshell Literally got co-pilot to add a new tagging library to a django app - saved so much headache space

I have been playing around with various AI assistants.

JetBrains AI keeps claiming that it does not have access to what's happening in their own software like eslint output in the integrated terminal, even refusing to run or pretend to run eslint on the code in the open editors.

Windsurf is well integrated into its VSCode-based IDE and it's got all permissions that it needs, only to hallucinate boilerplate code that doesn´t even work, although it's able to fix the errors that it made in a subsequent step, or to break a working configuration by downgrading npm packages based on the false claim that the current package version does not exist.

Line-based auto completion is still great in both editors, but one must be an absolute beginner to benefit from vibe coding in its current form.

@devux-bookmark I like JetBrains integration a lot, the fact you can use your own local LLM and on the free tier no less.

For privacy reasons but also if I wanna reverse engineer something it's nice to have an local "unlocked" model, shame most unlocked models are 1+ year old right now or they are naming them something else now. Who knows.