Completely the opposite experience here! I am a tech lead with decades of experience with various programming languages.

When it comes to producing code with an llm, most noobs get stuck producing spaghetti and rolling over. It is so bad that I have to go prompt-fix their randomly generated architecture, de-duplicate, vectorize and simplify.

If they lack domain knowledge on top of being a noob it is a complete disaster. I saw llm code pick a bad default (0) for a denominator and then "fix" that by replacing with epsilon.

It isn't the end, it is a new beginning. And I'm excited.