I agree that you shouldn't over rely on AI, that you shouldn't rely on AI to write code when you are learning. However, in most codebases I can't see that happ

I agree that you shouldn't over rely on AI, that you shouldn't rely on AI to write code when you are learning. However, in most codebases I can't see that happening. Once the codebase is complex enough, AI will not work, take more time to use than writing the code yourself or breaks the existing code. The author himself said that AI isn't usable for anything more complex than an University project, yet he complains he lost his skills because he used AI. As far as I see it, in the current stage of AI, it has limited usage. You can use it to start a greenfield project or start a new feature that is independent of the rest of the codebase. It will help with most of the setup and boiler plate. Past that, it will either not be able to do what you need, it will take more time or will break the code. So this pretty much doesn't allow you to modify large code bases or add logic. But you can still use it to generate a function, a method or a service, provided that function, method or service does not require much context and you don't need to modify the rest of the codebase to accommodate it. I see AI as merely an accelerator, not solving problems by itself but helping you solve problems faster, sometimes. I think it is very similar to Intellicode in Visual Studio and other editor tiols, which don't write code but provide method autocompletion, provide small suggestions, provide syntax highlighting, provide formatting and warn you when you make syntax errors. Was I using a text editor instead of a good IDE, my speed of development would be slower.