XNews

Vibe Coding Engineering Best Practices

Two practitioners share vibe coding experiences. One longtime developer describes fully delegating code to AI when working in an unfamiliar domain (Swift + AppKit) after initial mistrust; they co-decide architecture with AI, perform only high-level review and testing, and found quality sufficient to let AI self-maintain. The second shares a list of best practices: maintain clean architecture and layering, add automated tests (with regression coverage for bug fixes), remove accumulated features along with their code, use GItHub Actions for CI/CD runs on clean cloud machines before release, have AI perform auto-validation/correction while humans validate at critical stages, and convert repetitive tasks into skills to avoid re-explaining work. Together they illustrate a shift from coding to architecture oversight and quality gating.

Key insights

In unfamiliar domains, delegating code to AI and focusing on architecture and testing can be productive

A seasoned engineer found that letting AI write Swift/AppKit code—an area they were not expert in—resulted in good-quality output after high-level review and functional testing. Architecture and module design were collaboratively defined with AI, and the engineer stopped close code inspection when safety and performance were verified, enabling AI-led maintenance.

Vibe coding best practices center on structure, automation, and reusable skills

Key practices include: keep proper architecture and layering for maintainability; use automated testing and extend tests with bug fixes to prevent regression; clean up deprecated features and associated code; enforce CI/CD with GItHub Actions running full test suites on clean cloud environments; let AI perform auto-validation/correction with human gate reviews only at critical stages; and package frequently repeated workflows as skills to avoid re-explaining context to AI.

Source posts