Vibe Coding: Best Practices for Flow, Fun, and Results
Vibe coding is the name for letting the ai write your code. The magic lies in AI’s ability to infer intent from vague input—say almost nothing, and it often gives you something surprisingly right. This works because you’re granting it wide degrees of freedom; as long as you’re flexible on the outcome, the model can generate plausible results from minimal cues. But that magic is fragile: on follow-up turns, or when you have a much tighter definition of correct, the inferred intent often changes, and the model applies a different set of constraints then you wanted or what it did previously, causing a cliff, or misery. Turns out this is the same as junior developers.
The key to making vibe coding as productive as possible is having spefications, and verification.
The fix: Specifications
The key to success is specifying your many implicit constraints. And just like in real life, you don’t know them all up, so build up the rules gradually. But I bet the learnings apply across projects so wwrite them out explicitly and then you can pply them to many projects.
For a comprehensive example of codified development practices and conventions, see chop-conventions, which demonstrates how to document and standardize development workflows.
Business logic: PRD/User Stories
Which changes go where: Architecture
The “taste” of code: Clean Code Best Practices
The “readablity” of code: Language specific best practices
The “environment” - Tool chains, CI/CD, Commit Hooks, etc
This includes your development environment setup, tooling configurations, and automated workflows. Having consistent, well-documented development environments is crucial for productive vibe coding.
Check out the dev-setup practices for examples of how to standardize development environments and tooling across projects.