Article · AI translation
IT Home reported on October 25 that at the 2026 China Computer Federation conference on October 22, Jiang Guanjun, head of AI algorithms at Alibaba's Qwen division, shared Qwen's technical work on building personal AI agents. He said the true era of personal agents should create additional value and deliver personalized, proactive services, with models, memory and the ecosystem serving as the 3 elements that together build this core capability.
Jiang Guanjun divided the development of AI applications into 3 stages: conversation, agency and assistance. AI applications are now moving rapidly toward the assistance stage. Beyond completing specific tasks, they are beginning to draw on professional expertise and personal context to plan and act independently around users' goals, and to identify needs that users have not yet noticed or may have overlooked.
He said the theoretical upper limit for personal AI agents to solve problems in everyday work, study and life is close to 100%. The key to realizing this potential is to better match system capabilities with paths to solving problems and, through effective coordination among models, memory and ecosystem capabilities, find the best solution for each problem.
On the model side, Qwen places greater emphasis on the reliability of task execution. To that end, Qwen has created a map of user needs covering more than 30 leading industries, identified more than 8000 requirements for atomic model capabilities, and brought in industry experts to help evaluate and improve its models, continually strengthening their ability to understand real user needs and complete complex tasks.
On the memory side, Jiang Guanjun said personal AI agents need to understand users over the long term and recognize how historical information relates to current needs, while avoiding the noise and reasoning costs that come from simply accumulating conversations. For example, AI could draw on an older person's earlier mention of their pension to offer advice on their child's employment prospects that better fits the family's circumstances.
On the ecosystem side, Jiang Guanjun said that as model capabilities and Harness technology continue to develop, the ecosystem's ability to connect and collaborate will become crucial to the development of personal AI agents. Qwen is building an open, model-centered service system that brings together specialized knowledge, tools, industry Skills and third-party Agents; connects services such as e-commerce and local services; and incorporates professional capabilities in areas such as healthcare and education to improve the execution of complex real-world tasks.
Jiang Guanjun said the coordination of these 3 elements ultimately points toward proactive service: “Memory provides personal background, the model combines professional knowledge to identify potential needs and plan actions, and the ecosystem supports execution.” This would let users easily build applications for life, work, study and other settings simply by stating their needs, without having to program.