You can download an IDE with fancy autocomplete and use it without issue, but there of lot of devs who either don’t understand the limitations in the tools, or don’t understand the underlying code well enough to be able to maintain it.

The comparison to DreamWeaver is precisely on point. DreamWeaver didn’t invent a core technology, it was in essence an IDE with fancy autocomplete (the autocomplete was visual rather than textual). You and I could learn DreamWeaver, understand its limitations, and work with it in a way that didn’t produce a mess; but the vast majority of people using it didn’t understand its limitations, and didn’t understand the fundamentals of what it generated, so most DreamWeaver sites ended up being shitty messes.

LLMs for code aren’t a bad tool, they are great tools when used correctly. Just as DreamWeaver can generate absolute nightmares, LLMs can do the same. I’m an advocate for LLMs in development as they are an extremely useful knowledge resource and automation tool, but broad understanding of its limitations by developers is woefully absent. Being a dev in a FAANG(adjacent) company, where (I feel) developer standards should be set at a high bar, the slop I’ve seen in PRs recently is genuinely concerning. We’ve had numerous company-wide workshops on LLMs, none discussed their limitations and what you need to be wary of, things you need to know in order to use the tools properly. The unblinking trust I see principal engineers have in LLMs is mind blowing, to the point where if a question can’t be answered by an LLM, they almost give up entirely.

Sufficed to say, LLMs aren’t the problem; it’s their misuse, over reliance, and misunderstanding that is the problem