Xie Jibao (Tang San)
Head of Technology for Qoder
Previously worked in the Middleware Technology Department’s High Availability Architecture Team, where he led the design of key systems such as gray release and geo-distributed active-active architecture, and participated multiple times in Double 11 stability assurance, gaining extensive experience in high-availability systems and architecture. He currently leads the technical architecture and product direction of the AI programming tool Qoder, including the development of products such as the CLI and QoderWork.
Topic
Redefining Desktop Productivity: Creating a Full-Scenario Driven Desktop Agent
QoderWork is Alibaba’s first desktop intelligent assistant—not just a tool, but a “smart hub” on your desktop—aimed at redefining the boundaries of human-computer interaction through full-scenario intelligence. Deeply integrated with your operating system, QoderWork breaks the traditional software silo. Whether coding, document editing, data analysis, or creative design, it understands the current context and, leveraging an advanced Agent architecture, can automate repetitive tasks, intelligently orchestrate applications, and transform your workflow from “operating software” to “commanding tasks.” QoderWork is building an ecosystem that shifts productivity from manual operation to full-scenario intelligent guidance. Outline: Product Positioning & Design Philosophy QoderWork is built around three principles: Local-First: Agents run on the user’s machine, with direct access to the local file system, terminal, and browser. Safe & Controllable: Sandbox mechanisms and file protection strategies ensure user data security. Extensible: Skill modules and the MCP protocol continuously expand the Agent’s capabilities. Core Technologies Autonomous Task Planning & Execution Engine: How does an Agent break down a natural language instruction into an executable task chain? Covers task planning mechanisms including TodoList-driven progress management, multi-tool coordination, and self-validation and error correction during execution. Sandbox & File Protection System: Design strategies for file safety (prevent permanent deletion, automatic backups, workspace isolation) and script security review, balancing powerful Agents with user confidence. Skill System & MCP Protocol: Pluggable skill modules enable Agents to handle professional formats like PDF, DOCX, PPTX, and Excel. Live Demo: Agent workflows in real scenarios Lessons Learned & Best Practices Audience Takeaways: Understand the core design challenges of desktop AI Agents. Gain practical approaches for Agent task planning and execution. Learn the extensible architecture using Skill System + MCP Protocol. See real-world validation, not just conceptual demos, and obtain actionable technical decision references.
