Often the business requirements are non-sensical, contradicting, or pointless. Someone has to actually figure out what they mean without pissing off steakholders.

Remember, this isn't actually AI, it's machine learning, and the learning part is a misnomer.

> Often the business requirements are non-sensical, contradicting, or pointless. Someone has to actually figure out what they mean without pissing off steakholders.

That reminded me very much of this skit "The Expert". https://www.youtube.com/watch?v=BKorP55Aqvg&ab_channel=Lauri...

I'm pretty sure I get paid to take everybody's collective best guess at what requirements should be (sometimes with way too many people involved giving their concerns/opinions, or just not being able to get anybody's attention and everybody rubber stamping it/half-assing it), put it in QA/CAT/UAT, let somebody else bang on it, then we all figure out in our effort to try to not miss anything, we missed a bunch of massive design flaws. Then we go back to the beginning or scramble to fix it really fast, and the whole design repeats itself.

At least that's what it feels like I get paid to do...

Ha, sounds familiar. I doubt AI will be able to automate that sophisticated process ;)

I suppose AI needs to be able to reason and we are still a long long time away from AGI according to experts.

Btw, what’s the CAT step?

client accessible testing/user accessible testing

another environment before prod but not internal only so our “clients” can hit it for testing

Thanks. Where I've worked we only had Dev > QA > UAT > Prod - so not heard of CAT before.

>Their job will be to create the solution/product even if they're not writing code.

Yep.

You can make a good argument that today, any technical profession doesn't have to know as much because of existing software in most any given sector, whether its CAD software with built in stress/CFD analysis for mechanical parts, or frameworks for actual software development that are larger building blocks then pure code.

In the same way, future software engineers will likely be using generative software based on models like GPT that can take plaintext English and translate it into code. There will still be knowledge required of which model to run, what the parameters do, how to tweak the output, e.t.c

Of course, in the further future (although exponentially less time), those engineers will no longer be needed because there will be general models trained on all of their work as well. But really though, at that point, we wouldn't be that far from AGI.