Abstract This paper proposes a stage-based framing of AI governance practices.Rather than focusing solely on runtime controls or post-deployment oversight,we examine governance mechanisms across the lifecycle of AI-assisted software development. We identify four stages of AI governance: 1. Constraint Formation2. Development-Stage Governance3. Runtime Governance4. Post-Hoc Oversight The analysis highlights that many contemporary governance mechanismsconcentrate on runtime and post-hoc controls,while governance during the engineering stage remains structurally underdeveloped. This framing clarifies the conceptual boundaries of governance approachesand provides a foundation for future engineering-stage governance research.