Prompt Engineering Is Requirements Engineering (www.oreilly.com)

The piece argues that prompt engineering is not a new discipline but a modern instantiation of requirements engineering: the same human communication problems that caused the “software crisis” in the 1960s–90s are now reappearing as teams write prompts for AI. Drawing on the history of software quality (Deming, Juran, Crosby), Agile user stories, and the “No Silver Bullet” thesis, it shows that prompts are simply lightweight, iterative requirements artifacts—placeholders for a conversation that align intent, assumptions, constraints, and acceptance criteria across stakeholders (now including an AI). Tools, templates and prompt libraries can help, but they can’t replace the disciplined work of surfacing and sharing intent. For AI/ML practitioners this reframing has concrete implications: treat prompting as requirements work—specify functional and nonfunctional needs (readability, maintainability), provide targeted context (surrounding code, test inputs/expected outputs, design constraints), and iterate with tests and validations. Beware two common failure modes: too little context (model hallucinates based on its training distribution) and too much context (model loses focus). Because models won’t reliably ask clarifying questions, teams must design prompts and processes that capture fitness-for-use and conformance-to-requirements, integrate automated checks, and resist the “template trap” that confuses format for shared understanding.