Skip to content

Latest commit

 

History

History
86 lines (61 loc) · 3.49 KB

File metadata and controls

86 lines (61 loc) · 3.49 KB

Vibe Engineering: Working with AI Coding Agents

What is Vibe Engineering?

Vibe engineering represents the professional end of the AI-assisted development spectrum—where seasoned engineers accelerate their work with LLMs while staying proudly accountable for the software they produce.

This contrasts with vibe coding: the fast, loose, and entirely prompt-driven approach with no attention to how code actually works.

The Rise of Coding Agents

Modern coding agents (Claude Code, OpenAI's Codex CLI, Gemini CLI) can iterate on code, actively testing and modifying it until achieving specified goals. Experienced engineers are now running multiple agents in parallel, tackling several problems simultaneously.

Best Practices for Agent-Assisted Development

1. Automated Testing

  • Robust test suites let agents fly
  • Without tests, agents may claim success without verification
  • Test-first development is particularly effective with iterative agents

2. Planning in Advance

  • Start with high-level plans
  • Iterate on the plan first, then hand off to the agent
  • Clear specifications lead to better results

3. Comprehensive Documentation

  • Agents can only keep a subset of the codebase in context
  • Good docs enable API usage without reading implementation
  • Write documentation first; let agents implement from it

4. Good Version Control Habits

  • Essential for understanding and undoing agent changes
  • LLMs excel at Git navigation and tools like git bisect
  • Leverage their competence with version control

5. Effective Automation

  • CI/CD, automated formatting, linting, preview environments
  • LLMs make writing automation scripts easier
  • Automation helps agents repeat tasks consistently

6. Code Review Culture

  • Fast, productive code review is critical
  • Treat agent output like any other code submission
  • Don't be afraid to reject and request changes

7. Management-Like Skills

  • Provide clear instructions and necessary context
  • Give actionable feedback
  • No need to worry about offending the agent!

8. Manual QA Excellence

  • Beyond automated tests, strong manual testing skills are essential
  • Predict and dig into edge cases
  • Preview environments are crucial for safe review

9. Research Skills

  • Identify the best approaches before implementation
  • Prove concepts before unleashing agents
  • Research remains a key blocker/enabler

10. Knowing When to Use AI

  • Maintain intuition for when LLMs can be applied
  • This constantly evolves as tools improve
  • Not everything should be delegated

11. Updated Estimation Skills

  • AI changes project timelines unpredictably
  • Estimations now depend on new, evolving factors
  • Senior engineers must recalibrate their intuition

Key Insight

AI tools amplify existing expertise. The more skills and experience you have as a software engineer, the faster and better your results with LLMs and coding agents.

Vibe engineering requires operating at the top of your game: researching approaches, deciding architecture, writing specs, defining success criteria, designing agentic loops, planning QA, and managing your "digital interns."

Why "Vibe Engineering"?

  • Establishes clear distinction from dismissive "vibe coding"
  • Signals a harder, more sophisticated approach
  • Embraces the absurdity of the current AI moment
  • The mismatch between "vibes" and "engineering" is intentionally mischievous

Remember: These tools are powerful, but you remain accountable for the software produced.