Vibes Don't Maintain

A restaurant owner with no coding experience used Claude Code to build a polished staff-management app, then asked a software engineer to host and operate it. The app looked convincing and handled its intended workflow, but it lacked basics needed for real use: mobile support, user accounts, language options, settings, error handling, and safeguards for empty, duplicated, or simultaneous edits. Its data storage was fragile, too. The app was a useful prototype for testing an idea, not a finished product. The central argument is that people mistake the visible “happy path” for the bulk of software work, overlooking the protections and failure handling that matter when systems are used in production.

Putting the prototype into daily use could expose corrupted data, broken dependencies, security risks, or scheduling errors caused by daylight saving time. Without specifications, tests, documentation, or recorded design decisions, a future developer would have to reverse-engineer the code before fixing it. AI makes implementation faster, but it does not supply the judgment to anticipate risks or determine what should be built; vague instructions produce systems that cover only obvious scenarios. What makes software maintainable is an accessible record of its requirements and decisions, not merely working code. The piece is not an argument against AI, but against treating a quick demonstration as a durable system - or assuming that producing code is the whole of software engineering.