Comment by NitpickLawyer

> Commented [9]: This is fundamentally untrue. An LLM can certainly spit out thousands of lines of code, but "opens the app itself" is definitely up for question, as is "clicks the buttons" considering how unreliable basically every computer - use LLM is. "It iterates like a developer would, fixing and refining until it's satisfied" is just a bald - faced lie. What're you talking about? This is not what these models do, nor what Codex or Claude code does. This is a clever and sinister way to write, because it abuses the soft edges of the truth - while coding LLMs can test products, or scan/fix some bugs, this suggests they A) do this autonomously without human input, B) they do this correctly every time (or ever!), C) that there is some sort of internal "standard" they follow and D) that all of this just happens without any human involvement

---

Ummm. Yeah, no. This actually works. No idea why bozos who obviously don't use the tools write about how the tools don't do this or that. Yes they do. I know because I use them. Today's best agentic harnesses can absolutely do all of the above. Not perfect by any means, not every time, but enough to be useful to me. As some people say "stop larping". If you don't know how a tool works, or what it can do, why the hell would you comment on something so authoritatively? This is very bad.

(I'll make a note that the original article was written by a 100% certified grifter. I happened to be online on llocallama when that whole debacle happened. He's a quack. No doubt about it. But from the quote I just pasted, so is the commenter. Qwaks commenting on qwacks. This is so futile)