科技观察：“端侧智能体”将走向何方

Core Viewpoint - The development of AI agents, particularly "end-side AI agents," is gaining traction, with significant advancements in their ability to perform tasks autonomously, shifting from mere interaction to active service [1][2]. Group 1: AI Agent Development - The term "end-side AI agents" refers to AI systems that can independently perform reasoning, decision-making, and task execution on devices like smartphones and IoT devices [2][4]. - The year 2025 is anticipated to be a pivotal moment for AI agents, marking the "AI agent year" [2]. - The current trend shows a rising interest in AI agents across foreign media and social platforms from January 2024 to January 2026 [2]. Group 2: Technical Routes - There are two main technical routes for developing end-side AI agents: the GUI (Graphical User Interface) route and the protocol authorization route [4][5]. - The GUI route allows for visual recognition and simulated operations without needing to adapt to third-party applications, providing short-term deployment advantages [4]. - The protocol authorization route, which includes APIs, is currently the mainstream choice, offering better security and traceability but requiring a robust industry ecosystem [4][5]. Group 3: Industry Differences - China adopts a dual-track strategy of GUI simulation and API authorization, aiming for a balance between innovation speed and system security, while the U.S. primarily favors the API route [5]. - The choice of technical routes is influenced by ecological structure, institutional constraints, and market demands, indicating that a single path may not support the long-term evolution of the AI agent ecosystem [5]. Group 4: User Interaction and Concerns - The GUI route has significantly changed smartphone interaction paradigms, enabling users to perform tasks like ordering food or booking tickets with simple voice commands [6]. - Concerns exist regarding user rights and data privacy, as users may not fully understand the permissions granted to AI agents, potentially leading to unauthorized access to sensitive information [8]. - The risks associated with high-level system permissions for AI agents could complicate security measures and increase vulnerability to cyber threats [8]. Group 5: Future Directions and Governance - Experts emphasize the need for a comprehensive approach to AI governance, focusing on permission delegation and the identification of acceptable error rates across different industries [10][11]. - The future of end-side AI agents may involve a shift towards a more integrated ecosystem where AI can seamlessly interact across various applications, potentially revolutionizing user experience [9][10]. - A multi-layered, flexible technical framework is recommended to support diverse access methods, enhance security, and improve cross-entity collaboration [11].