智能体工程（Agentic Engineering）的核心内涵并非全新的工程范式，而是传统软件工程应该早已采纳的规范做法。[1] 这些实践包括使用问题追踪器、编写文档字符串、撰写描述性的代码审查请求、保持文档更新、编写有意义的测试、自动化代码检查以及进行设计评审和记录会议内容等。[1] 正因为人工智能智能体的出现，这些被长期忽视的工程规范才重新获得关注。
文章作者引入了詹姆斯·C·斯科特提出的"mētis"概念来阐释问题的根源。[1] 编码规范是容易被形式化描述的人类构造，但工程原则需要根据具体情况本地化和灵活调整。[1] 如果基础工程实践不健全，仅仅引入AI工具也无法改善局面。作者指出，如果智能体工程遇到困难，应当反思公司的常规软件工程是否同样困难。[1] 智能体缺乏应对复杂情况的本地化能力，将AI加到一个功能失调的系统上无法解决问题。[1]
A recent discussion highlights that agentic engineering, despite its contemporary framing, fundamentally represents best practices that software engineering should have adopted long ago.[1] The core practices now being emphasized—including issue tracking, documentation, code reviews, and automated checks—constitute standard software engineering disciplines that have gained renewed attention primarily due to the emergence of AI agents.[1]
The argument underscores a critical dependency: without proper foundational engineering practices in place, introducing AI tools alone cannot resolve systemic problems.[1] As one perspective notes, "if agentic engineering is just normal engineering with robots added, then anything that makes engineering harder will make agentic engineering much harder."[1] This reflects a broader principle that agents lack the contextual judgment—what philosopher James C. Scott termed "mētis"—necessary to navigate complex situations that fall outside standardized procedures.[1]
Organizations experiencing difficulties with agentic engineering implementations should first examine whether their conventional software engineering practices themselves are functioning effectively, since "adding AI to a dysfunctional system doesn't solve the problem."[1] The underlying message is that successful agentic engineering requires addressing foundational engineering discipline before overlaying automation technologies.