The question of recursive self-improvement cuts to the heart of how AI development might accelerate beyond human-paced progress, but recent work suggests the picture is more nuanced than a simple runaway loop. A growing body of research has documented tensions between speed and control: human-AI research teams may actually outpace purely autonomous systems, and automating feedback loops while preserving oversight remains an open engineering challenge. Yet even as systems gain the ability to modify themselves, a deeper concern emerges—one that transcends speed entirely. The erosion of human influence through incremental AI replacement suggests that self-building AI might pose governance risks not because it moves too fast, but because it quietly decouples AI development from the human oversight mechanisms that historically kept it aligned. Does recursive self-improvement represent a technical inevitability, or an organizational choice we're still equipped to shape?