Alibaba Qwen outlines Personal Agent framework at CNCC 2026
At CNCC 2026, Alibaba Qwen's AI lead Jiang Guanjun presented a technical framework for Personal Agent, emphasizing the coordination of model, memory, and ecosystem to enable proactive services. He noted that AI applications are moving from dialogue to agent and assistant stages, with Personal Agent potentially solving nearly 100% of daily problems if system capabilities align with problem-solving paths. Qwen has built a user demand map covering 30+ industries, identified 8,000+ atomic model capabilities, and reported a 1.6x increase in average token consumption for complex tasks over the past three months.
Coverage timeline
机器之心新闻资讯
10月22日，在2026中国计算机大会上，阿里巴巴千问事业部AI算法负责人蒋冠军分享了千问打造Personal Agent的技术实践。他认为，真正的Personal Agent阶段，应该带来增量价值的创造和个性化的主动服务，模型、记忆与生态是共同构建这一核心能力的三要素。 蒋冠军将AI应用的发展分为对话、代理和助理三个阶段。当前，AI应用正加速迈向助理阶段，在交付具体任务的基础上，开始结合专业经验和个人上下文，围绕用户目标自主规划与执行，并进一步发现用户尚未注意到或有所疏忽的需求。 他表示，Personal Agent解决日常工作、学习、生活问题的理论上限接近100%，释放这一潜力的关键，在于让系统能力与问题解决路径更好地匹配，通过模型、记忆与生态能力的高效协同，为不同问题找到最优解决路径。 在模型层面，千问更加重视任务执行的稳定性。为此，千问建立了一张覆盖30多个头部行业的用户需求透视图，梳理出8000多项原子模型能力要求，并引入行业专家参与模型评估与优化，持续提升模型理解真实用户需求、完成复杂任务的能力。 在记忆层面，蒋冠军表示，Personal Agent需要长期理解用户，识别历史信息与当前需求的关联，避免简单积累对话带来的噪声和推理成本。例如，AI可以结合老人历史对话提及的养老金情况，为其孩子的就业问题提供更贴合家庭处境的建议。千问通过分析事件关联关系设计记忆体系，并按关联强弱分层存储和调用，提升记忆使用的精准性。 生态层面，蒋冠军认为，随着模型能力和Harness技术的持续发展，生态的连接与协作能力将成为Personal Agent发展的关键。千问正构建以模型为中心的开放服务体系，整合专业知识、工具、行业Skill及第三方Agent，连接电商、本地生活等服务，以及医疗、教育等领域的专业能力，提升复杂现实任务的执行效果。 三要素的协同最终指向主动服务：记忆提供个人背景，模型结合专业知识识别潜在需求、规划行动，生态支撑执行。例如，在健康管理中持续关注康复安排及其对日常生活的影响，在理财过程中提醒用户留意新的风险与机遇。 在千问上，用户对复杂任务和主动服务的需求持续增长，过去三个月复杂任务平均Token消耗量增长1.6倍。最新消息显示，千问推出的个人专属AI助理「千问Go!」上线新功能“我的应用”。用户无需编程，只要说出需求，就能轻松搭建用于生活、工作和学习等场
Related stories
Alibaba's Qwen launches 'Qianwen Go' personal agent in beta, handling tasks like flight delay compensation
Alibaba's Qwen unveiled 'Qianwen Go', a personal agent product in beta testing, which can autonomously handle tasks such as flight delay compensation claims, booking services, and customer service negotiations. Users share experiences of the agent logging into platforms, searching order history, and contacting customer service to complete claims with minimal user intervention.
1 reports · 1 sources
Fudan, Tencent Hunyuan, Tsinghua Release ExplorationBench to Test AI Rule Discovery
Fudan University, Tencent Hunyuan, and Tsinghua University released ExplorationBench, a benchmark measuring how AI systems discover rules in unfamiliar environments. It includes two altered 'alien worlds'—the programming language AlienCode and the logic proof system AlienLogic—where models must learn rules through trial and error. Results showed that without environment feedback, scores remained low (max 11.0%), while autonomous exploration led to high scores (up to 89.0%).
2 reports · 1 sources
AgentForesight: 7B Model Audits Multi-Agent Systems Online to Predict Failures
Researchers from Rutgers, UT Austin, and Purdue propose AgentForesight, a 7B model that audits multi-agent systems during task execution to identify critical errors before they cascade into failures. On the AFTraj-2K benchmark, it achieves an Exact-F1 of 66.44 for key error steps in failed trajectories, outperforming the strongest general model baseline by 19.88 points, with a false positive rate of only 2.37% on successful trajectories.
Alibaba's Qwen Office builds e-commerce version, targets pre-Double 11 launch
Alibaba's Qwen Office team is developing an e-commerce-specific version of its AI office product, with nearly 100 staff involved, targeting release before the Double 11 presale. The version, accessible via web or client, offers a dedicated 'e-commerce operations' interface with five sections including Double 11 zone, business analysis, product management, content generation, and marketing promotion. It is the first industry-specific version of Qwen Office, leveraging the team's general agent capabilities and e-commerce team's domain insights.
1 reports · 1 sources
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe