TENCENT(HK:00700) Sou Hu Cai Jing·2025-12-26 07:25Core Insights - The article discusses the growing importance of "Agent Infrastructure" (Agent Infra) as a critical factor for the successful deployment of AI agents in business environments, highlighting the challenges faced by traditional cloud computing infrastructures in supporting the unique characteristics of agents [2][3][7]. Group 1: Market Potential and Challenges - The global agent market is projected to reach $285 billion by 2028, with 15% of daily business decisions made autonomously by agents and 33% of enterprise software incorporating agent capabilities [2]. - In China, the AI agent software market is expected to exceed 5 billion RMB in 2024 and grow to 852 billion RMB by 2028, with a compound annual growth rate of 72.7% from 2023 to 2028 [2]. Group 2: Paradigm Shift in AI Applications - Traditional AI applications focus on "determinism," while agents introduce uncertainty, complexity, and autonomy, making their behavior less predictable [3]. - The emergence of agents necessitates a shift in how businesses approach AI, requiring them to manage and control the inherent uncertainties of agent behavior [3][4]. Group 3: Technical Evolution and Infrastructure Needs - The article emphasizes the need for a new infrastructure tailored for agents, termed "Agent Infra," to address the limitations of traditional cloud computing in handling high-frequency, lightweight, and real-time workloads [7][23]. - Major cloud providers are competing to develop infrastructures that offer higher elasticity, lower latency, stronger security, and longer session management for agents [8]. Group 4: Sandbox Technology - The sandbox is identified as a crucial component of Agent Infra, providing a controlled execution environment that ensures security and isolation for agents [10]. - Traditional sandbox technologies are deemed inadequate due to their slow startup times, prompting the development of new solutions like Tencent Cloud's Cube, which can deliver a secure sandbox in approximately 100 milliseconds [11][14]. Group 5: Comprehensive Agent Runtime Solutions - Tencent Cloud has introduced the Agent Runtime solution, which integrates various core modules for managing the entire lifecycle of agents, including execution engines, cloud sandboxes, and context services [19][20]. - The execution engine acts as a central hub for intelligent scheduling and supports long-running sessions, which is essential for complex agent tasks [20]. Group 6: Future Directions and Challenges - The industry is still in the early stages of developing a comprehensive Agent Infra paradigm, with current solutions primarily focused on making agents operational rather than optimizing their performance [23][26]. - Future advancements will need to address challenges such as evaluation frameworks for agent performance, data management, and memory/context management to enhance agent intelligence and control [24][25].