Lili Zhang
Baidu Comate Front-end Architect, Baidu Front-end CMC Leader
Baidu front-end architect, head of Baidu front-end CMC, head of Baidu technical organisation committee for web direction. 2023 Participated in Baidu's intelligent development tool Comate's IDE end architecture, model enhancement, etc., and has rich experience in the application of large models in software development such as renewal, generation, Q&A and so on.
Topic
Building Scalable Agent Tasks Based on Skills
As large language models and agents continue to evolve, development tasks driven by coding agents are becoming increasingly large-scale and complex. Long-horizon tasks have even emerged as a distinct topic of research and discussion. At the same time, the concept of skills as the core mechanism for extending agent capabilities is becoming more standardized and mature. Against this backdrop, combining skill-based capability extension with workflow design, and introducing the key elements required to construct long-horizon tasks, enables agents to autonomously complete large-scale, high-complexity tasks in a reliable, stable, and high-quality manner. This talk will share insights and practical experience from two perspectives: high-quality skill development and how skills support large-scale task execution. Outline: 1. Evolution of agent capability extension 2. Introduction to agent skills 3. How to develop high-quality skills 4. Key success factors and case studies for large-scale tasks 5. Using Meta Skills to automatically construct large-scale tasks
