FA engineer runs spec, test, report writing via Claude Desktop

This is part of a series on AI coding in FA environments without terminals or Git, explaining why the author chose Claude Desktop over Claude Code. His work is dominated by documents — specs, test procedures, reports — taking nearly half his work hours, so non-coding tasks are where AI helps most.

He shares specific methods. For spec structuring, he hands AI bullet-point notes and asks for a skeleton covering input/output, state transitions, and abnormal behavior, which surfaces missed points like abnormal cases or initialization. For ambiguous specs, he asks AI to generate questions, such as timeout needs or reset conditions on recovery. Test procedures are derived from specs, and he emphasizes including a re-execution step after Execute drop, which is easily missed. Reports use a strict division: the author writes facts, AI formats, and he checks for added content, especially fabricated numbers or dates.

He also notes AI's weakness with audience targeting and the risk of formulaic writing. On information management, documents contain client names and equipment details, so he separates personal and business folders, stressing that line-drawing is more critical for documents than for code.