My understanding is that AI/LLMs can generate code that works, but often that code is more bloated/convoluted than code written by a decent coder, making it less efficient and more difficult to maintain.

This was more true last year but developments in both the frontier models and the harnesses used to control and direct those models have gotten better to the point where I’d say the code is better then most devs. People on here don’t want to admit it but if you did a “coding turing test” between Claude and some random dev coding by hand I’d bet most of the time a dev would say the AI output is better.