Prompt Design and Feature Comparison in AI Coding Assistants

다국어 초록 (Multilingual Abstract)

Large language models have catalyzed significant changes in software development through AI-powered coding assistants. This study examines prompt design strategies and features of three widely-used tools: GitHub Copilot, ChatGPT, and Amazon CodeWhisperer. Through systematic benchmarking using algorithmic problems, API integration tasks, and debugging scenarios, we quantify performance differences across accuracy (71-92%), executability (71-94%), and security vulnerability rates (3-12%). GitHub Copilot demonstrates strongest algorithmic performance (92%), while CodeWhisperer excels in API integration (88%) with lowest security vulnerability rate (3%). We provide evidence-based tool selection guidelines and a practical prompt engineering checklist for developers and educators.