As others have stated, I don't think we have anything to worry about.

As a SWE you are expected to neatly balance code, its architecture and how it addresses the customers' problems. At best, what I've seen LLMs produce is code monkey level programming (like copy pasting from StackOverflow), but then a human is still needed to tweak it properly.

What would be needed is General AI and that's still some 50 years away (and has been for the past 70 years). The LLMs are a nice sleight of hand and are useful but more often wrong than right, as soon as you delve into details.