This article shares the author’s firsthand experience developing large-scale projects using AI tools like GPT-5.2 and Opus 4.5. Initially expecting AI to streamline workflows, the author discovered that these AI tools’ context windows are insufficient for processing complete requirement documents and prototype screenshots. This necessitates manual preliminary design and task decomposition. The experience reveals current AI limitations in complex projects, emphasizes the necessity of human intervention, and provides practical insights for AI developers and software engineers to optimize AI-assisted development strategies.
