IT Home On October 25, at the 2026 China Computer Conference on October 22, Jiang Guanjun, the head of AI algorithms at Alibaba Qianwen Division, shared the technical implementation process of Qianwen in creating personal AI agents. He believes that in the true stage of personal agents, incremental value creation and personalized proactive services should be brought about. Models, memory, and an ecosystem are the three key elements that jointly build this core capability.
Jiang Guanjun divides the development of AI applications into three stages: dialogue, agent, and assistant. Currently, AI applications are accelerating towards the assistant stage. Based on delivering specific tasks, they begin to combine professional experience and personal context to plan and execute actions autonomously around user goals, and further identify needs that users have not noticed or overlooked.
He said that the theoretical limit of an individual AI agent in solving daily work, learning, and life problems is close to 100%. The key to realizing this potential is to align the system’s capabilities with problem-solving approaches. Through the efficient coordination of models, memory, and ecosystem capabilities, the optimal solution for different problems can be found.
At the model level, Qianwen places greater emphasis on the stability of task execution. To this end, Qianwen has established a perspective diagram of user requirements covering over 30 leading industries, identified more than 8,000 atomic model capability requirements, and involved industry experts in model evaluation and optimization, continuously improving the model’s ability to understand real user needs and complete complex tasks.
At the memory level, Jiang Guanjun said that individual AI agents need to understand users over time, identify the connections between historical information and current needs, and avoid the noise and reasoning costs associated with simply accumulating conversations. For example, AI can combine the pension-related discussions in past conversations of elderly people to provide more relevant advice on their children’s employment issues based on their family circumstances.
At the ecological level, Jiang Guanjun believes that with the continuous development of model capabilities and Harness technology, the ability to connect and collaborate within the ecosystem will become crucial for the development of individual AI agents. Qianwen is building an open service system centered around models, integrating professional knowledge, tools, industry skills, and third-party agents. It connects services such as e-commerce and local life, as well as professional capabilities in fields like healthcare and education, to improve the execution of complex real-world tasks.
Jiang Guanjun said that the collaboration of the above three elements ultimately leads to active service. That is, “Memory provides personal background, the model combines professional knowledge to identify potential needs and plan actions, and the ecosystem supports execution.” Users don’t need to program; they just need to express their needs, and applications for life, work, and learning can be easily created.