You could say the same thing about every improvement to programming workflows (version control systems, context aware editors, CI/CD, test frameworks, better languages/language improvements, package managers, Q&A repositories like StackOverflow, build systems, and so on). Whether you really consider increasing individual output dark/bad/ominous is up to you but if you apply that outlook historically you'll have had decades of negative outlook on the thought the job is going to become scarce while the number of high paying software jobs continued to increase despite efficiency improvements.
In a more direct way: Making one programmer able to output what two programmers can do is almost always a gain for everyone involved. Making one programmer able to output what 10,000 programmers can do is a sign the field is being replaced. I don't think we'll get anywhere near concerns of the latter with LLMs.
Agreed -- I purposefully avoided saying "AI" instead of "LLMs" because LLMs aren't all that's required for AI to replace programmers. And my guess is that it will be a long time until AI replaces programmers.
I like that. I would darken it a bit though:
"LLMs won't replace programmers, but a programmer with an LLMs might replace two programmers without LLMs."