Do you know the best practices to start vibe coding?

Vibe coding empowers non-developers to build software using AI agents. However, without proper foundation and structure, vibe coding can result in unmaintainable code that fails "the bus test" – meaning if you were hit by a bus tomorrow, no one else could understand or maintain your code.

The key to successful vibe coding is starting with the right foundation. This means ensuring your code lives in a proper repository, uses proven architectural patterns, leverages existing AI guidance, and takes every change from an issue to a pull request.

Getting started

❌ Don't

Don't keep your code only on your local machine or in an unmanaged environment.

✅ Do

Create a GitHub repository (e.g. in the SSWConsulting GitHub organization) from the start.

Starting your project in GitHub, specifically within your organization, ensures:

• Version control - Every change is tracked and can be reverted
• Collaboration - Others can contribute and review your code
• Backup - Your code is safely stored in the cloud
• Transparency - Team members can see what you're building
• The bus test - If you're unavailable, others can take over

Use architecture templates

Don't start from scratch! Use one of these battle-tested SSW templates that already include comprehensive AI guidance.

Clean Architecture template

Use the SSW Clean Architecture template which includes:

• Proper separation of concerns (Domain, Application, Infrastructure, Presentation layers)
• Built-in dependency injection
• Comprehensive AGENTS.md files guiding AI on architectural decisions (See more on writing an AGENTS.md file)
• Pre-configured with best practices and patterns

Vertical Slice Architecture template

Alternatively, use the SSW Vertical Slice Architecture template which offers:

• Feature-focused organization (each slice is a complete vertical through all layers)
• Easier to understand for non-developers
• AI guidance files pre-configured
• Less ceremony than Clean Architecture for simpler projects

See the software architecture decision tree for more guidance on choosing the right architecture.

Drive each change from an issue

Take every change through the same small loop:

1. Start from an issue - Describe one small change and its acceptance criteria. See Do you use guardrails when vibe coding with AI?
2. Ask for a plan first - Review it before any code is written, and have a developer check non-trivial plans. See Do you review your AI coding plans? and Do you get your implementation plan reviewed before you write the code?
3. Implement on a branch - Never let the agent change main directly
4. Open a pull request - Check it does what the issue asked, then have a developer review it before merging. See Do you do AI-assisted development by vibe coding one bite at a time?

Pick a coding agent to work through the loop with you:

• Copilot cloud agent - Runs on GitHub.com with no local setup. Point it at the issue, ask for a plan and iterate until it's right, then have it implement on a branch and create the pull request
• CLI and IDE agents such as Claude Code, Codex and Copilot CLI - See Do you use AI CLI tools? and Do you use the best AI Powered IDE?

❌ Figure: Bad example - A vague request with no scope and no way to check the result

✅ Figure: Good example - One small change with clear acceptance criteria and boundaries

Why this approach passes the bus test ✅

By following these practices, your vibe-coded project will be:

• Maintainable - Following proven architecture patterns
• Understandable - Using standard structures that developers recognize
• Documented - The template's structure and AI guidance explain how the code fits together
• Traceable - Every change links back to an issue and a pull request
• Reviewed - A developer checks every change before it reaches main

Learn more

• SSW Vibe Coding Consulting