The debate around AI coding assistants is missing something important: most of their limitations were predicted decades ago by classical cognitive and organizational theory. Drawing on the work of Nobel laureate Herbert A. Simon, Michael Polanyi, and complex systems research, this paper argues that AI struggles with software engineering for structural—not temporary—reasons. AI optimizes locally, while engineering requires global reasoning, tacit knowledge, and system-level judgment. The real question is no longer whether AI can generate code. It is whether prediction alone can ever replace engineering thinking.