Relying on AI to write and read code may also remove the practice through which developers learn architectural judgment. That is the argument of this September 22 essay: the author says the cost of poor architecture can take months, even years, to show up as software that is hard to maintain, and that no reliable metrics capture it.
The essay connects that delayed feedback to learning. Developers build judgment by making decisions, encountering their consequences and correcting mistakes, so the author argues that people who hand those choices to AI never develop it. This is an argument from engineering experience, rather than a measured comparison of AI-assisted and human-written projects.
