LLMs will not replace programmers

I think that anyone genuinely believes LLMs will replace human programmers is either:

• Heavily invested in lying to themselves and their investors (e.g. OpenAI);
• Does not understand how LLMs work and thus doesn't understand that nothing resembling thought is actually happening;
• Is not a good programmer themselves, thus cannot judge the absolute shit slop LLMs generate;
• Is really interested in replacing human programmers, reality be damned (a catastrophe couldn't possibly be more expensive than the programmer salaries, right?);
• Has never actually tried to do something with an LLM;
• Is a merchant of chaos, and it doesn't matter what reality is, the point is to fearmonger.

And my record will show that I have consistently held the above opinions from the beginning of the bubble.

The above has been met with the following justifications for why I'm wrong, by people that aren't even programmers:

• "Just a matter of time"
• "Even [insert big important CEO person] says so"
• "Even [insert trustworthy rich friend, who happens to CEO an AI company] says so, and I really trust them"
• "You're just not prompting well, you need to learn how to prompt"
• "Have you tried XYZ AI tool?"

Those that have worked with me know that despite pressure my opinions have remained steadfast, and I continue to not vibe code anything, or even use AI to write code. I at most use ChatGPT to write some code, to get an idea or become aware of options, then I write it myself.

The reasons for my opinions are rather straightforward, perhaps simplistic:

I'm an actual programmer, I've been programming for over a decade, I enjoy programming, and I push my skills consistently.

Therefore I know what programming actually involves, particularly when dealing with real engineering problems.

Early on I studied how LLMs work. I understand modern LLMs have all kinds of tricks and mechanisms to improve their performance, but that doesn't change their underlying reality.

I have never seen anything programmed by an LLM that was high quality, certainly not anything that was up to the standard I hold the code that I stamp my name on.

And I hope it gets there, but I also haven't seen a convincing case to be made for "engineer architects, AI writes", because AI tends to write quite crap.

That's it.

Call me a fool, time will tell who is wrong, I don't care if it's me or others. I just know I enjoy programming.