The conversation around AI in software development has largely fixated on the 'copilot' paradigm: intelligent assistants that augment human programmers, offering suggestions, completing boilerplate, and catching errors. This framing, while accurate for the immediate past, misses...