Dragon Riding: The Art of Vibe Coding.

Programming is changing. We’re moving away from the era of manual, line-by-line character entry and into something much more… visceral. Some call it “Vibe Coding,” but to me, it’s Dragonning or Dragon Riding.

The name started as a joke. A friend and I were talking about the old way of coding versus this new, agentic, AI-led approach. We joked that people are still coding manually and expecting to keep their jobs blissfully unaware of the dragons!

All this while my friend was watching his AI agent tear through a complex refactor, he actually started playing Rise of the Dragon from the Tamil movie Dragon (by Pradeep Ranganathan). From that moment on, we weren’t just “using AI”—we were Dragonning.

But riding a dragon requires a different set of skills than building a horse-drawn carriage. If you want to master the art of Vibe Coding, here are the core principles of Dragon Riding.

1. Plan Before You Code

Don’t just ask your agent to jump straight into code generation. That’s like jumping on a dragon’s back without a saddle—you’re going to get burned.

Instead, treat the planning phase as a collaborative dialogue:

• Work together: Develop an implementation plan with your agent.
• Refine and review: Ask questions. Challenge the agent’s assumptions. Let the agent challenge yours.
• Seek clarity: Don’t stop until the plan is unambiguous and crystal clear.
• Only then move to implementation: A solid plan is the difference between a successful flight and a crash landing.

Pro Tip: Experiment with different models. Some models are exceptional artists (coding), while others are brilliant architects (planning). Find the pairing that works best for your workflow.

2. Commit Early, Commit Often

Frequent commits are your safety net. In the world of Dragon Riding.

• Commit before every feature: Before you let the dragon take off on a new task, make sure your current progress is safely stored.
• The “Git Stash” Rule: If a session takes too long or the agent starts hallucinating “slop,” don’t try to fix it by layering more complexity on top of errors. Just git stash and start fresh from a clean state.
• Promoting the Winner: If it took five rounds of “No, that didn’t work, try this” to finally get it right, don’t keep that messy history. git stash one last time, then re-prompt the agent with only the specific approach that worked. This keeps your codebase clean and your dragon focused.

3. Voice is Your Friend

Your voice is often faster and more expressive than your fingers. Voice prompting allows for a more natural, conversational flow that matches the speed of your thoughts.

• Try built-in IDE features: Many modern editors have native voice support.
• Third-party tools: There are loads. I use “Super Whisper”.
• Think aloud: Describe your logic as you think it. Let the agent catch the nuances of your intent that might get lost in a typed command.

4. Context is King

A dragon can’t fly in a vacuum. The more context you provide, the more precise and helpful your AI partner will be.

Don’t just send snippets. Provide the “why” behind the “what.” Think aloud with your transcription running and send that rich context to the agent. When the agent understands your architecture, your goals, and your “vibe,” the results are magic.

5. Documentation as Communication

Documentation isn’t just for human users anymore; it’s a primary communication channel for your agent.

• Sync as you go: Keep your READMEs and documentation in sync with your code changes.
• Ask the agent to write: Agents are surprisingly good at documenting the code they just wrote. Use them to maintain a high-quality knowledge base.
• Future-Proofing: Well-documented code is easier for both humans and AI to work with six months down the line.

6. Rules of Engagement

Every dragon needs boundaries. Establish clear guidelines for how you want your agent to work:

• Define your preferred patterns and conventions.
• Set boundaries for the kinds of solutions you want (e.g., “avoid external libraries if possible” or “prioritize performance over readability in this module”).
• Explicit rules create consistency across different sessions and different agents.

7. Mentor Your Dragon

Think of your agent not as a tool, but as a talented junior engineer you’re mentoring.

• Train it: Guide it toward your preferred architectural patterns.
• Feedback Loops: When it does something well, acknowledge it. When it fails, explain why.
• Codify the Wins: Once you’ve established a pattern that works, have the dragon record it in its rules directory. This ensures the lesson sticks for the next flight.

Conclusion

Dragon Riding isn’t just about efficiency; it’s about the feeling of being in sync with a powerful partner. It’s about taming the complexity of modern software development with intent, context, and a bit of a cinematic flair.

So put on that song, hold on to the reins and get ready to fly. Happy Dragonning!