Yes, and there's another reason to it as well (old fart ranting here):
When you do suggest that LLMs can be terrific and explain why they CAN be useful, old professional coders will often chime in, chop your head off and absolutely hate on you, every move you make, everything you say with passion.
I'm old so the last time that happened I just said "oh you're right, I didn't think about that, my bad" and they were happy.
Do you think I stopped experimenting with the tools? Not in any way. In fact I run my own local LLMs now on (I feel kinda super lucky) hardware bought the last few years before pricing became absolutely insane, so I have some hardware to do that, glad I did not sell those, but that's getting offtopic, I did warn - I'm an old fart so I do rant a lot.
But I do experiment a lot too.
For anyone who cares, here's the skills you do need:
- Your prompting needs to be god-tier level, it's like in the olden days we used to say "Google it", well most people had no clue WHAT to google, because they often didn't have the initial skills to know what to look for. LLMs are the same, they have various training, you need to know their limitations and ask questions sharply and very focussed, not like you'd ask a human (because they're not), but kind of extremely specific, like you would with a computer.
- Have thousands of hours with AI prompting, experimentation, build and learn to scaffold your code. Build frameworks with your AI, compartmentalize your code so you don't have to ask LLM to re-code everything from scratch every time, saves tokens, saves on context memory (especially important if you run locally). The thousands of hours with using the models will help you become an expert in using it as a tool, surprise - just like any other tool.
Just because you can prompt and get results, doesn't mean you'll get good results, you need to get good at recognizing the failure points, what it responds to well and where it fails, and learn to tighten your scaffolding skills. The more direct, focused you are, the more NOTES you take of your project, such as naming the items in your apps, naming the tasks in your app (accurately, not loosely like this), the more high quality your results will be.
- You also need to be a great project manager and note taker. Take tons of notes, document the hell out of everything you do, and have your LLM document EVERYTHING. Make things tight so an Idiot (me) could understand it even with meager coding skills (I might be humble here, but in my world everyone is better than me, it helps because you anticipate you could be wrong, this is good!).
Learn the above well, be humble, use several approaches, learn from that, and you'll be surprised how dangerous and amazing these tools indeed are. I've honed my local model skills to become so damn good I code actually good games today with the help of ONLY local models, we've reached that point - we were NOT at that point just 9 months ago.
No joke!