Unfortunately @billylanchantin more or less nailed the topic’s nuances and reasoning so I am left feeling almost useless trying to contribute. 
I am not an economist and I will not pretend that I am. But I don’t quite agree that us the programmers are naive or blind to it. Our systems thinking does help us analyze things on a decent level.
To that end, AI companies are actually painting targets on their backs; them taking away livelihoods makes them actively hated and gradually worked against by other, also quite big, players – do you think big chains of grocery stores want to see 12% revenue reduction? They will start doing something.
I feel like, in the bigger picture, we are gradually entering a phase where mega-capitalists will start turning on each other. The market is not infinite and many of the AI companies are discovering this in a very painful way (most of them operate at a loss).
Sorry, that was a bit of a digression. But still related.
On to the topic itself, more directly: AI will not replace barely any programmer longer-term. The fact that the ruling class wants that to be true and that they are firing many people is just the effect of a collective delusion – one that already is beginning to fade because many vibe-coded apps cannot be easily extended by AI, or when they are, they have loads of bugs.
The thing that most people miss about the influence of AI / LLMs is this:
It has nearly peaked and it cannot do much more than it already does.
The next play are sub-agents and complex agentic workflows. I like those because they play to the LLM’s strengths: smaller-scope tasks. So the sub-agents workflows have a real chance of actually replacing a programmer.
But who gives them the detailed plans? The clueless project managers that are hand-waving requirements and are leaving you, the programmer, to define the project’s scope and priorities, i.e. doing their job for them? How will those people be able to prompt the multi-agent LLM systems?
I remain very skeptical on that point.
Now one can say that because I have a vested interest in our profession not disappearing and as such my takes cannot be trusted. While that would be somewhat fair to say, I believe it would be misguided because in my experience most people cannot see long-tail effects.
So let us try to hypothesize a few long-tail effects?
- Loads of vibe-coded apps that are near-impossible to extend. Let us assume the absolute very best-case scenario for the owner / ruling class: that the LLMs would still be able to maintain their own slop. But in order to do so, they would require more and more tokens and thus higher rent. I guarantee you that most capitalists will not like that and will do something about it.
- Good chunk of people will give up on programming. Whether that’s good or bad I don’t want to theorize.
- …However, I posit that at one point companies will want more programmers again. I am fairly sure the demand might never swing back to what it was back in f.ex. 2019 but there will be some increased demand. Which means that the remaining programmers would be, theoretically, able to command pre-COVID salaries by the mere virtue of the laws of supply and demand.
- Jobs are being lost even now but I believe this is only a few-years trend. These people will not just lay down and die. They’ll find another place in the economy. Which brings me to:
- Reducing the general populace’s buying power will not go unpunished. Think of the AI companies as a powerful and large but disorganized army. Their first blitz-krieg attacks had a crushing success but they cannot be everywhere and their limitations are already acutely felt by many. The “resistance” is already organizing.
TL;DR we are near the end of a bubble. I am not sure how doom-and-gloom the scenario would be; maybe a few blips on the stock market or maybe a full-blown crash leading to many other effects – no idea. But the current unsustainable pace cannot be kept for much longer.
And oh yeah, ChatGPT will include ads very soon. Surprising nobody in the IT area.
Even shorter TL;DR – the amount of programmers will get reduced but not by as much as predicted by many doom-sayers, and technical expertise will be even more valuable in some mere 2 - 5 years from now. Just endure the storm.
).
