Your prediction is... missing so much detail of how that prediction actually happens that it is pointless. This is my big dislike re. the discussion of LLMs and the effect of AI more broadly. Unless you bother to make an effort in going deeper why post it? Theres no value. The same stuff has been posted for months and even years at this point.
replyWhen GPT 3.5 was released, it could handle maybe a 500 LOC codebase. Experienced engineers were calling it cute, but zero threat to actual programmers.
replyThen it became thousands.
Now models can handle and operate on code bases with hundreds of thousands LOC, even low MLOC.
So in just 3.5 years we've gone from LLMs being cute toys, to being powerful enough to actually replace junior engineers. Even if we hit a new AI winter tomorrow, the proverbial damage is already done.