GPT is good at boilerplate and is very bad at understanding and fleshing out complicated requirements, coming from N stakeholders with varying opinions and personalities. So I don't think GPT is a threat to human engineers in the "essential complexity" space.
Then, how much time do we collectively spend on boilerplate? Stackoverflow + the rich collection of open source libraries and frameworks have been doing a great job minimizing boilerplate for typical architectures. E.g., building a Flask app today vs building with Apache mod_perl in 2000 is like a >90% reduction in boilerplate.
So I don't think LLMs are much of a threat to engineers.
Then, how much time do we collectively spend on boilerplate? Stackoverflow + the rich collection of open source libraries and frameworks have been doing a great job minimizing boilerplate for typical architectures. E.g., building a Flask app today vs building with Apache mod_perl in 2000 is like a >90% reduction in boilerplate.
So I don't think LLMs are much of a threat to engineers.