100%
I wouldn’t bet on it. The two guys up above have both accused me of trolling. Apart from it being a disgusting term IMO, in psychological terms it means they most likely got “triggered” (another disgusting word given the context, also IMO), which on the other hand implies they may have made an emotional investment into the hype, hence their accusation of me “trolling”.
You will do well to take a few deep breaths because your tone was not quite productive at places either. Let’s not have this thread locked. There are interesting sub-plots here. And IMO discussing pros and cons of the current breed of AI gives people perspectives – that alone is extremely valuable.
I was not expressing bias or vested interest. I was responding to what you actually wrote, which may not be what was intended.
In your original message, you used the term ‘prompt’ while just now, you used the term ‘result’. To me, prompt is input while result is output. Those are opposite meanings. At this point, it is not clear to me which you actually meant.
It’s a figure of speech called adynation (a hyperbolic figure of impossibility), meaning what I’ve already told you. I could’ve written it in a longer and more explicit way like: “No AI prompt that generates code can produce a result that is as optimal as the code written by humans.”, but it would no longer sound as dramatic for a headline, but rather dull.
And yes, it did cross my mind that there might be some ambiguity to it, especially if taken at face value, but I opted for it anyway. Plus, I myself was is in a dramatic mood at the moment still fresh from yet another frustration with LLMs (hence the post).
Btw, another episode with LLMs just around the time of your reply. Seems they are particularly dumb when it comes to Elixir macros, but I just solved it myself (again) so I won’t complain (for now).
This might not be a useful observation. But programming languages are not great tools for expressing ideas, they are implementation tools for controlling a machine, they all have constrains & flaws and bypassing their constrains you need either some esoteric knowledge (battle scars) or time. Hence why we use various PL for various tasks
I think on the other hand prompting is better for expressing ideas, the key is to constrain what the LLM can do with said prompts. Last week I was firmly on the AI & LLMs are garbage for generating software, this was based on my claude use from earlier this year, but since the release of gemini 3, I have heavily revised my opinion and I strongly believe that eventually LLMs will be great at “one-shotting” software generation in the coming months
This just reminds me of a programmer I respect on twitter saying that he starts all new projects with AI, now this was like 3 months ago and I thought such proclamation was cringe/corny, but now I can see where this is headed
This seems to be a pretty common approach, and it makes sense. AIs are good at producing something, so they are a great tool for getting over the initial inertia of trying to start a project. If LLMs were around back when I was in school I would have used them to write the first draft of every paper. I was always pretty good at revising my essays, but getting that first draft down was the hard part.
When it comes to software though, I personally find that revising is actually harder depending on what kind of foundation is in place. I don’t think there’s anything wrong with using an LLM to generate the initial project boilerplate as long as the programmer driving it is still putting thought into the structure of the code and the modules and all that.
All the little decisions early on add up to unmaintainable software if we’re not careful. 
It kind of seems like the industry is hoping that LLMs will get good enough that they can just rewrite it once it gets bad, so this would become a non-issue. I think that’s probably not feasible without another major breakthrough in AI research though. The current approach to training LLMs just doesn’t seem like it would be sustainable with the amount of improvement they’d need to get to that point. 
Can you please share how exactly you think this will be possible given the limitations of the technology employed? I understand there is anecdotal evidence of LLMs excelling in doing superficial repetitive jobs and thus saving us time, but for being good (let alone great) at software engineering it takes real intelligence, not just pattern matching.