Abstract
This paper examines the techno-political transformation behind the rise of autonomous AI agents. Through three recent cases: the “Doubao Phone” interface conflict, Meta’s acquisition of Manus AI, and Alibaba’s task-oriented Qwen agents, we analyze an emerging governance discourse, called the Agentic Ideology. This discourse combines neoliberal empowerment rhetoric with techno-feudal rent extraction, reorganizing human action through agent mediation. We conduct a comparative case analysis structured around three analytical layers: protocol, interface, and governance. At protocol level, we examine Anthropic’s Model Context Protocol (MCP) and its primitives for structuring delegation and decision visibility. At interface level, we compare structured parsing approaches with “pure vision” GUI agents such as UI-TARS. At governance level, we analyze interface conflicts between agents and major platforms. We argue that the shift from Human-Computer Interaction to Agent-Computer Interaction redistributes control within digital infrastructure, and HCI research should address how public interfaces can support accountable human-agent collaboration.