From Using to Mastering: Real-World Challenges of AI Coding in Production

In a recent InfoQ AICon livestream, moderated by Lin Xiangxin from NetEase Games, experts Chen Jie from NetDragon Websoft, Bai Jiachen from PingCAP, and Li Weining from HSBC Technology explored how Coding Agents are reshaping software engineering. They noted that AI Coding increases business demand rather than reducing headcount, and that legacy code transformation is a major need. Quality assurance shifts from line-by-line review to risk-based review and automated gates, with developers taking more responsibility. The experts discussed the use of MCP, Skills, and Rules, noting that Skills can cause context bloat and require governance. Agent permissions were a key concern, with emphasis on isolation, minimal privilege, and human gates for high-risk actions. They also highlighted the importance of knowledge systems, evaluation gates, and changing organizational culture to adopt AI-native practices.