rant about ai

[Kapendev]

On Saturday, 15 November 2025 at 18:16:40 UTC, monkyyy wrote:
> https://github.com/crazymonkyyy/usv
>
> https://github.com/crazymonkyyy/usv_old/blob/master/usv.d
>
> I feel ai is here as a ".1x engineer" who does a years work of 
> .1x work over night
>
> The goal will be to control it and to get objective 
> measurements(such as the compiler) running as much as possible 
> as you leave it to do a long todo list over a week end
>
> I tried 3ish approaches to implement usv(dont use it, every 
> line of code is ai slop still) the best by far was "spec-driven 
> development", or more cynically ranting in md docs and telling 
> the ai to read and reread them or directing it to use a doc 
> section while writting a test.
>
> You let the ai hallucinate its on an agile team and your its 
> boss, and lots and lots of unit tests. I believe this gets you 
> the best results.
>
> I want unix file permission management so I can mark human docs 
> as read only, and so the ai knows which files were its. I want 
> folder watchers that will auto run tests and tell me when a 
> test got short(implying the ai deleted something) or started to 
> fail cause the ai didn't bother to run it.

It should produce OK results if you give it an abstract idea 
about how things work and the problem is small in scope. Text 
files with simple pseudo code and comments are a good idea. 
Making it work with a specific project and language is a waste of 
time IMO. I'm not really the target user for anyway, so yeah.

Btw, I see an open issue on your project. You should probably fix 
it smh smh