How an AI coding experiment reshaped one developer's approach
The author recounts an initial failed attempt to use AI to turn an Excel-based assessment tool into an application. A second attempt worked better after describing desired outcomes and user flows instead of expecting the AI to infer intent. The experience prompted lessons about product thinking, architecture, and the changing nature of software development work.
The writer, an agile coach, had relied on spreadsheet-based team agility assessments, including one with more than 120 questions. Seeking a lighter option, they designed a 10-minute, 12-question Team Agility Quick Scan and built it in Excel.
A first attempt to convert that spreadsheet into an app through Replit failed; the author uploaded the Excel file, expected the AI to infer intent, and abandoned the project for nearly a month. On retry, describing desired outcomes, user flow, registration, scoring, and behavior produced a React interface, SQL backend, registration, and working assessment workflows within an hour.
This account may encourage more non-engineers to build software by describing goals rather than code, potentially shifting product development toward broader participation. Teams and small organizations could benefit from lighter assessment tools, while developers may see their roles emphasize judgment, architecture, and user-flow design over routine implementation. It may also raise expectations about AI reliability, since outcomes could depend on clear intent, oversight, and the ability to evaluate generated systems.