What's happening with LLMs isn't fundamentally different than what has happened as various higher levels of abstraction were introduced since I started writing code in 1983.

Back then, you plugged in a computer and got, typically a C:> prompt, this strange thing nobody had seen before. There was a big hype cycle, and the basic thinking of clients was infused with tremendous FOMO. I was immediately hired mostly because I wasn't afraid of computers. Often the attitude was almost like, "here's a wheelbarrow full of money and blank slate, write me some -- what do you call it -- software for my business so I can participate in this. No, I don't know what it has to do, you figure it out". I mean it was literally low hanging fruit everywhere.

Fast forward 43 years, and relative to THAT, most ordinary mortals are very computer literate, they often feel they have paid too much $$ for too long to people like me, and they think they fully understand the problem space and how to manage devs, while mostly neither of those things are really true.

Still, I have managed to specialize and select the right clients such that I still have tremendous freedom to use my best judgment while being very well compensated. In fact my present client has the attitude, "of course, don't use AI, why would you do that?" although admittedly I'm afraid to fully understand how they came by such a sanguine, hype-proof attitude.

My guess is that I'm one of a half dozen senior people in a pool of about a half dozen industry players in a very specialized fintech field, and I've built whole systems of the type they want twice before, my reputation precedes me, and they want a steady hand to bring those capabilities to them as well -- so whatever I want, is fine by them. Also they play the long game very well.

But even I feel obligated to periodically educate them about LLMs (I refuse to use the term "AIs") and my thinking about them, to sort of inoculate them from the almost inconceivably epic levels of hype around the topic. Fortunately they see the results from my labors that they are looking for, and don't want to disturb that.

And even I have a quantized open weight model hogging RAM on my local hardware where I can figure out if it's useful for anything (I'm looking at unit testing and prototyping mostly). This also gets around my environmental, security / privacy and ethical concerns about using cloud models.

But like many commenters here, I need to be able to substantially "own" / stand behind / understand my code or the meaning just leaches out of it for me.

Ultimately I don't feel the least bit of competition from LLMs in terms of understanding the problem spaces that I have a truly intimate knowledge / experience of, because what I bring to the table is UNDERSTANDING, which is something no LLM has. It only has probabilistic associations and chutzpah.

I think this is true for a lot of us, but it's just going to take a few more months or (dog help us!) a couple of years for the hype cycle to die down and for people to realize there's mostly no real value-added after you pay for all the token costs and data center downsides; rather, we've just been moving the pain points around. The music will quit playing, the investment dollars will dry up, and much (not all) of this will be yet another Cautionary Tale of hubris and overreach.