Prompting Is Programming

A software engineer’s essay argues that effective AI prompting is essentially programming, explaining why engineers have seen dramatic productivity gains from tools like Claude Opus 4.5 while other professionals have not. The author says he largely stopped writing code within weeks of Opus 4.5’s release, a shift echoed by nearly every engineer he knows, whereas marketers, analysts and managers mostly use AI for emails and meeting summaries and then spend saved time correcting plausible but flawed output. He attributes the gap less to code-heavy training data or verifiability than to the mental discipline of programming: defining inputs, specifying exact outputs, breaking problems into verifiable steps and anticipating edge cases. Non-programmers often omit a clear starting state and goal, as in a request to “put together a report on Q3 sales” without specifying data sources, breakdowns, baselines or formats. The piece matters because it suggests AI’s promised org-wide productivity gains depend on teaching workers to specify problems like engineers, not just handing them better models.