Qwen outlines personal agent framework built on models, memory and ecosystem

Qwen's head of AI algorithms Jiang Guanjun described a three-part architecture for personal agents at CNCC 2026 in Chengdu: the model, tiered memory and an ecosystem of services. The team mapped user needs across 30+ industries into more than 8,000 atomic model capability requirements. The talk is the most detailed public account yet of the thinking behind Qwen Go!, which opened for sign-ups this month.

• Qwen's needs map covers 30+ industries and 8,000 model capability requirements
• Memory stores events by relation strength and tiers instead of piling up chat history
• The ecosystem links e-commerce, local services, healthcare and education
• Jiang split AI apps into chat, agent and assistant stages, with the industry entering the third