I fully agree. I didn’t see your comment before I wrote my own would have saved me a lot of trouble lol. I always mention a personal anecdote in regards to this topic of discussion…
AI will never be able to make a bad coder good. AI will only ever make a good coder write code to the standard they usually write code at, but around 20x faster.
Story time 
In Push Back, my teammate claimed that he could code the entire robot/auton using Claude Code in approx. 20 minutes. When he did, he never actually compiled or tested it, and told all of us (in a very self-assured manner) that he had fully finished the auton code.
Meanwhile, I (lead programmer) architected the program and wrote the algorithm myself. I then used AI to only write the code, and never really left it to make any descisions on its own. Prettty much all I used Claude for was to save me the wrist pain of typing for 5 hours straight (although the algorithm was pretty long so i did that anyways lol).
My teammate’s code has 23 compilation errors and failed at Nationals.
My code executed decently well, and I was able to EXPLAIN how my code worked to the Judges.
(For context, this all happened the night before Nationals, and I stayed up till 5 am coding)
We won Think award, even though we placed dead last in the tournament.
PS. here is the code which got us Think Award at Nats and qualified us to Worlds: