The article examines the frustrations and reliability problems developers face when working with AI coding agents, focusing mainly on Cursor’s Composer 2.5. It notes that some issues may be model-specific but have also appeared in other models such as GPT-5.5 and Opus 4.6.