Because it's a tool which must be used properly. I've encountered senior engineers who, while great on their own complain that AI isn't good at code gen. When I talk to them about it they're using terrible free models and not putting in effort to understand how LLM's work. Agents can now write thousands of lines of code across large codebases following specs closely on levels that are simply impossible for teams of humans to do.

Humans simply cannot code as well as an LLM/Agent in most cases. It's like fighting a bear, and if you think you can beat an adult brown bear you're probably wrong.