@simon maybe...someday...

In my observations LLM assistance is still somewhat uneven in its effectiveness...it is not "trained" on all languages or platforms evenly. It is basically useless for programming industrial automation devices for example..which is probably a good thing TBH.

I do acknowledge that in many cases LLM assistance makes the easy part easier, that being the creation of code that compiles or executes. However it often makes the hard part harder, the hard part being testing for functional correctness and meeting requirements. For example when a junior coworker used LLM assistance to try to build scripts to automate some software deployment the resulting output was valid code that did things but appeared to be based on using an OS release from 12 years ago, and it took longer to make corrections than it would have to just do it from scratch.

An experienced devops person who is disciplined enough to review the LLM output and can spot the weirdnesses can use it, but newer people...?