My Wife Wanted Dior. I Spent $600 on Claude Code to Vibe-Code a 2M-Line Database Instead.

James Luan, the VP of Engineering at Zilliz, shares an experience where he spent $600 on Claude Code for an AI experiment instead of buying his wife a Dior bag for their anniversary, leading to two lessons: the importance of personal relationships and the limitations of AI on complex tasks. As a maintainer of Milvus, an open-source vector database, Luan attempted to use Claude Code for cross-platform compilation, a task his team had previously avoided due to its complexity. Despite initial progress, he faced numerous setbacks as fixes for one platform caused failures in others, leading to a realization that the problem lay not with the AI's intelligence but with the process. By adopting a constraints-first approach, writing tests before implementation, and utilizing parallel computing, he successfully completed the task in two days. However, he acknowledged the ongoing challenges of translating this workflow to a team and maintaining personal relationships while working on intensive AI projects. Luan emphasizes the need for engineers who enjoy problem-solving and are methodical in their approach, inviting interested individuals to join his team at Milvus.