Vibe Like a Pro: Getting the Most Out of Your AI Coding Assistant
Copilot and Cursor turn engineers into designers. Five habits that get the most out of AI pair-programming — plan first, work small, test as you go, let AI debug, teach it your style.
Tools like GitHub Copilot and Cursor are turning software engineers into designers — freeing us to solve bigger problems instead of getting stuck between keystrokes. Here's how I get the most out of my AI coding assistant.
Plan before you code
Start with a clear idea of the feature from both a user and an architecture perspective. Use a model to help generate this plan, document it in your codebase or ticket, and refer back to it as you build. 🧭
Work in small batches
AI excels at focused tasks. Don't try to write the entire codebase in one go. Break work into small, testable units and commit often. This makes it easy to experiment, fail fast, and roll back if needed. 🔄
Test as you go
Write lightweight tests for completed parts. Focus on user behavior rather than exhaustive coverage — especially since your code is evolving rapidly. 🧪
Let AI debug for you
Don't fall back to line-by-line manual debugging. Give your assistant the context and let it do the heavy lifting. Copy/paste might be your new best friend. 😅
Teach AI to code like you
Maintain your coding standards in the repo and feed them to your assistant before you start. This helps it write code in your style, not just generic best practices. 🎯
How are you integrating AI into your dev workflow? I'm always curious to compare notes.
Related reading
In Automation, Reliability Is the Product
Browser automation looks like a demo problem and is actually a reliability problem. What building Snappit taught me about why self-healing beats clever.
What an Accounting Degree Taught Me About Building Software Companies
I trained as an accountant and became an engineer and founder. The detour wasn’t wasted — it’s my edge. On outcomes, distribution, and finding what’s true.