Personal Taste Is the Moat (wangcong.org)

Recent discussions in the AI community highlight the limitations of AI in software development, particularly regarding its inability to assess the deeper implications of design choices. While AI tools can efficiently review code for correctness, identify bugs, and suggest fixes, they lack the nuanced judgment derived from "taste," a term redefined here to signify the deep understanding gained through experience with both successful and failed designs. An example illustrates this point: a Linux kernel patch, approved by AI, was ultimately rejected by a human maintainer for introducing long-term design flaws that AI couldn't recognize, such as unintended system complexity and the misalignment of fixes with root causes. This phenomenon emphasizes a critical shift in the landscape of software engineering, where the technical competence provided by AI is becoming standardized. As AI takes over routine correctness checks, the essential differentiator in engineering lies in the ability to make informed, value-driven decisions that reflect accumulated wisdom and judgment—elements that AI cannot replicate. In essence, while AI can ensure functional correctness, human taste will determine the longevity and appropriateness of design choices in complex systems. Thus, personal taste has emerged as a crucial "moat" that sets experienced engineers apart in an AI-enhanced era.