Hi Richard,

I do use "AI coding tools" to help with some of the work - but what is easily 
lost in the hype HYPE  HYPE HYPE that is really the basis of all reporting on 
AI tools is that these are often useful to get you started (and really not 
always at that), but they are creating a lot more problems than they solve when 
they start introducing code that you don't fully understand, yet that someone 
has to maintain. So any LLM generated code needs extra review and more careful 
evaluation. Whenever I consider adding such code, I point this out in the PR 
and ask for the other maintainers to pay extra attention.

And the problems that are slowing us down (especially with QLM) are very poor 
situations for LLMs in the first place, as those are trained by all the code 
the companies were able to steal, all the discussions that they were able to 
scrape from public forums, etc. And since QLM isn't widely used in open source 
projects, and there are extremely few examples of QML Qt 5 -> Qt 6 migrations 
(trust me, I've searched extensively), the LLM is just guessing (or making 
sh... stuff up)

/D


> On Feb 16, 2026, at 08:29, Richard De Witt <[email protected]> wrote:
> 
> Hey all,
> 
> I’m just a knuckle dragging electrician that has used Subsurface for about 10 
> years. I am not a SWE, or even a hobby coder. However, my job exposes me to a 
> lot of ai tools, including some coding ai.
> 
> Some of these challenges seem to be prime for using ai tools to solve and 
> debug some of the issue in compartmentalized blocks. There are a couple of 
> experiences I have had where the tool helps test and debug faster than our 
> team does. 
> 
> Is that an avenue being explored?

_______________________________________________
subsurface mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to