If you've used AI coding tools, how has your experience been?
Very positive
6.2%
Positive
24.6%
Negative
23.1%
Very negative
4.6%
I haven't used them
41.5%
Poll ended at .

@plexus Negative for the case you're probably thinking of (requires too much guidance to get it to generate the right thing, and I'd usually rather spend that mental effort just writing it myself). It turns "do a thing I am good at" into "guide an idiot through doing it" and that is a much less enjoyable task.

The one exception I've found is the subset of problems where I need a one-off script to run once, it's annoying to write myself, and it's reasonable to just run it and review the output. I did it today for "get a list of tables in this database with ON DELETE CASCADE foreign keys to this other table", because I did not feel like reading through Postgres docs for several minutes to generate a list of four items.

@plexus great for languages and systems I am not familiar with, horrible for languages and systems I am familiar with.

@plexus I spend quite sometime once, with ai assistant so it can assist with a powershell language. It was not able to produce even one runnable example for very small tasks. It was continuously suggesting methods calls that didn't exist.

I suppose that there is some bonus here for a language/library design, but as a developer this was a waste of my effort. Reading powershell sdocs was more productive.

@plexus Mixed - I think it's oversold everywhere. It's not the superpower that is punted all over.
Or I'm just living under a rock.

It's great at exploring new areas where I don't know much, but I always start with the mindset that the response/result needs to be verified.

Other times it just gets in the way.

Honestly, I was late to the game and very sceptical, but have started to embrace it where possible.

It's _not_ a silver bullet.

That said, I'm a local-first user. Not trusting own/client information to be sent off onto the internet, so I'm using a local inference engine (llama.cpp) with a model that seems OK (Qwen2.5 Coder 7B).

Tried DeepSeek... and not sure what the rave is about. The Qwen model gives better results (faster).

My knowledge is still limited and I do not claim to be an expert 😀

@plexus I am a hobbyist/beginner/intermediate and it sometimes helps me to „sketch” a code. Then I redesign it and try to make it more idiomatic.