Junwei Shi

Architect of Baidu Comate

Leads the system architecture design of the knowledge-enhanced and code-context engine, exploring the integration of large models in areas such as code generation, semantic retrieval, and adaptive reasoning. Focuses on research in code model training, knowledge augmentation mechanisms, and agent collaboration, with a commitment to advancing intelligent programming technologies and driving their deep adoption in enterprise-level R&D scenarios.

Topic

Coding Agents Reshaping Software Development

The Role of Coding Agents in Modern Software Development With the rise of intelligent development tools, coding agents are playing an increasingly vital role in today’s software engineering landscape. Baidu’s ERNIE Comate, powered by its code context engine, has significantly enhanced agents’ capabilities in code comprehension and task execution. It excels in use cases such as code search, defect localization, and intelligent code generation. Practice has shown that integrating high-quality pre-generated knowledge can further improve the overall performance and user experience of intelligent coding applications. This talk focuses on the construction and practical application of the code context engine, covering key components such as code retrieval tools and pre-generated documentation for codebases. --- ### Outline I. The Evolution and Underlying Logic of AI Coding A review of how large language models have evolved in the programming domain — from simple code completion to intelligent code generation — exploring the core principles and design logic behind this transformation. II. From Model to Agent: The Case of ERNIE Comate An introduction to the core capabilities of coding agents, including code understanding, generation, and collaboration; an in-depth look at Comate’s architecture across model design, tooling, and context management. III. Enterprise-Level Challenges and Context Engine Design Addressing challenges such as large-scale codebases, information redundancy, and limited context windows, this section details Comate’s self-developed reasoning-based context engine and how it efficiently handles complex enterprise scenarios. IV. Practical Cases: How Agents Empower Real Development Through real-world case studies, this section demonstrates how coding agents operate in actual development workflows, and how the context engine improves accuracy, speed, and stability. V. Deployment and Practice: The Optimal Path for Enterprise Adoption Sharing best practices from Comate’s enterprise deployment, including management tools and measurement systems designed to help organizations evaluate and enhance AI coding productivity effectively.

© boolan.com 博览 版权所有

沪ICP备15014563号

沪公网安备31011502003949号