How to Train Your AI Coding Assistant to Understand Your Style in 30 Minutes

If you're like me, you've probably experienced the frustration of your AI coding assistant not quite getting your style. You write clean, concise code, but it keeps generating overly verbose solutions. Or maybe it doesn’t fully grasp your preferred naming conventions. Whatever the issue, customizing your AI assistant can save you time and boost your productivity. Luckily, you can train it to understand your coding style in just 30 minutes. Here’s how.

Prerequisites: What You Need Before You Start

1. An AI Coding Assistant: Choose from tools like GitHub Copilot, Tabnine, or Codeium. For this guide, I’ll focus on GitHub Copilot, which integrates well with various IDEs.
2. An IDE: Make sure you have Visual Studio Code or another compatible IDE installed.
3. Basic Coding Knowledge: Familiarity with the programming language you're using (e.g., Python, JavaScript).

Step 1: Set Up Your Environment (5 minutes)

1. Install the AI Coding Assistant: If you haven't already, install GitHub Copilot from the marketplace in your IDE.
2. Create a New Project: Open a new project or file where you’ll write a few examples of your preferred coding style.

Step 2: Write Sample Code (10 minutes)

To train your AI assistant, you need to provide examples of your coding style. Here’s what to do:

1. Create a Style Guide: Write down your preferences, such as:

   • Naming conventions (e.g., camelCase vs. snake_case)
   • Commenting style
   • Function structure (e.g., single-responsibility principle)

2. Write Sample Functions: Create 3-5 functions that represent your coding style. For instance, if you prefer concise functions, write a few short ones with clear names and comments.

   def calculate_area(radius):
       """Calculate the area of a circle given its radius."""
       return 3.14 * radius ** 2
   

3. Use Diverse Examples: Include different scenarios, such as error handling, data processing, and API calls.

Step 3: Train Your AI Assistant (10 minutes)

1. Use Your Samples to Provide Feedback: After writing your sample code, start coding a new function. As you write, let the AI generate suggestions. When it suggests something that doesn’t align with your style, correct it.

2. Utilize Feedback Features: In GitHub Copilot, you can accept, reject, or modify suggestions. Take a moment to explain why a suggestion isn’t suitable for your style. This feedback helps the model learn.

3. Iterate: Repeat this process for a few different functions. Over time, the assistant should start generating code that aligns more closely with your style.

Step 4: Fine-Tune Your Assistant (5 minutes)

1. Reinforce Good Suggestions: As you code, give positive feedback for suggestions that fit your style. This reinforces the behavior you want.
2. Adjust Settings: Some AI tools allow you to adjust parameters like creativity or verbosity. Tweak these settings to see if it helps the assistant align with your style better.

Troubleshooting: What Could Go Wrong

• Inconsistent Suggestions: If the assistant still gives poor suggestions, revisit your sample code. Perhaps you need to add more examples or clarify your preferences.
• Longer Suggestions: If the AI keeps generating lengthy code, consider adjusting the verbosity settings if available or providing more concise examples for reference.

What’s Next: Building on Your Progress

Now that your AI coding assistant has a better understanding of your style, consider taking it a step further:

1. Integrate More Tools: If you're using GitHub Copilot, consider exploring alternatives like Tabnine or Codeium for comparison.
2. Create a Comprehensive Style Guide: Document your coding standards for future reference and to train other tools.
3. Join Coding Communities: Engage with other developers who are customizing their AI tools. You might discover new techniques or tips.

Conclusion: Start Here

Training your AI coding assistant to understand your style can transform your coding experience and productivity. Start by setting up your environment, writing sample code, and providing feedback to the AI. It takes only 30 minutes, but the benefits can last for much longer.

What We Actually Use

In our experience at Built This Week, we primarily use GitHub Copilot because of its robust integration with our IDE and its ability to learn from our coding style over time. However, we occasionally test out Tabnine for its unique features and performance.