Chen Yingping
Head of Alibaba 1688 Intelligent Innovation Team
Chen Yingping is the Head of the Intelligent Innovation Team at Alibaba 1688, with 15 years of experience in internet R&D and extensive experience in technology management. He has long focused on high-traffic e-commerce systems and AI application engineering, and previously led the incubation of the 1688 AIGC content generation platform and the 1688 official procurement assistant plugin from the ground up. He currently leads the team in developing Agent applications and engineering architectures for e-commerce operations, advancing Agents from search and decision support toward cross-platform business execution. His work focuses on long-running complex tasks, controlled execution, and evaluation-driven evolution, exploring a production-ready path for autonomous business operation Agents.
Topic
Enabling Agents to Run 24/7 in Real-World Stores: Architecture and Practice of Autonomous Operations Agents for Distribution
In distribution operations, tasks such as product selection, listing, supplier switching, and loss mitigation often span multiple platforms and long time horizons, while store conditions continuously change. A single conversational Agent is therefore difficult to rely on for stable, long-running execution. Based on experience running Agents 24/7 in real-world stores, this talk introduces a layered architecture of scheduled orchestration, Agent-based decision-making, and controlled Skill execution, along with engineering approaches for anomaly recovery, multi-store management, and end-to-end observability. It will also explore how to build real-world operational trajectories into a Shopkeeper Bench, connecting the full loop of candidate Skill generation, deployment evaluation, and regression rollback. Finally, the talk will review key engineering trade-offs encountered when evolving from single-store experiments and distributed deployments to cross-platform, multi-store operations. Outline From Business Automation to Autonomous Operations Agents Long-running tasks and dynamic states in distribution operations Layered design of scheduling, Agent decision-making, and Skill execution Running Agents Long-Term in Real-World Stores Anomaly detection, state validation, and automated recovery From independent single-store deployments to unified multi-store operation and observability Bench-Driven Skill Iteration Operational trajectories, Skill contracts, and the Shopkeeper Bench Deployment, evaluation, and regression rollback of candidate Skills Cross-Platform Migration and Engineering Lessons Migrating operational workflows from Pinduoduo to Taobao Validated capabilities, application boundaries, and future challenges Audience Takeaways Attendees will gain a practical framework for building Agents for real-world, long-running business operations: how to define the boundary between Agents and traditional automation; how to achieve controlled execution through layered scheduling, Agent decision-making, and Skills; how to support multi-store operation, anomaly recovery, and end-to-end observability; and how to use real-world operational trajectories and Bench-based evaluation to continuously assess, improve, and roll back Skills.
