Google Cloud is advancing the concept of autonomous AI systems—termed "Agentic AI"—for deployment in government agencies under its "Mission Impact" initiative. The approach shifts from AI systems that advise humans to systems that execute administrative decisions independently. Public sector organisations face a control dilemma: automation promises efficiency gains, but autonomous decision-making in benefits processing, licensing, or compliance raises urgent questions about accountability, auditability, and democratic oversight. Early implementations will test whether government IT teams can enforce meaningful constraints on AI agents, or whether the technology's autonomy fundamentally conflicts with administrative responsibility.