I find it very strange the way people sometimes talk about AI in general. It's in a manner usually devoid of criticism, rooted in the firm belief that it's the next "Big Thing", a technological jump akin to when we replaced assembly with higher, more abstract constructs (as the article puts it). Beyond that, those that don't believe this trend are themselves "stubborn", unable to glimpse at an obvious truth that is staring them down in the face, being themselves faded to be left behind while the righteous believers reap the benefits borne out of their superior insights regarding technology. I myself, clearly, find this proposition hard to believe.<p>While I won't knock the utility of LLMs (they obviously have a wide range of applications), this "ride or die" attitude seems to verge on religious zealotry. You don't need AI to code and, without reasonable proof of a real productivity increase (which the author only hints at, citing some internal studies that show a minimum increase of 30%, but without actually presenting the studies themselves), I will continue to doubt its ability at generating programs, even in the simplified, glue-it-together manner that the authors puts it.<p>One must remember what these LLMs truly are: statistical machines with no real contextual knowledge. It can't, therefore, be aware of the particulars of each implementation. What it generates needs to be analyzed and likely mangled by developers. So it begs the question, why not just code it directly? Perhaps it could lower the barrier of entry, after all, if I cannot code, at least the AI can give me a head start. But the novice developer won't even be able to discern between good and bad code, being forced to simply trust (and hope) that the generated program is not only correct, but that it fits well in the context of her application.<p>Further, what the machine can generate is wholly dependent on what is available for consumption (most of it stolen). Novelty, even in small amounts, will inevitably stump it. It already just performs educated guesswork, but, without basis, it's now only guesswork minus the education.<p>This is not to say that AI is useless, far from that, but is frequent that I see it be given this almost divine-like status. As if it were this inevitable, unstoppable thing, that will forever grow until it achieves perfection, to the woe the skeptics and the joy of the believers.<p>In my humble view, it remains a useful tool in a variety of situations, a occasionally better alternative to Stack Overflow for one, but I would never trust it to write code, at least good code that is.