Harnessing AI for Coding (A Beginner’s Guide)

“Vibe Coding” essentially describes leveraging a large language model (LLM) to translate your development vision into actual code. Expanding on this, it involves entrusting the AI with much of the intricate decision-making related to software architecture and logic, alongside handling the repetitive coding chores. This delegation allows you, the user, to channel your mental energy towards the more creative and enjoyable aspects of building software.

To begin, secure access to an AI service provider. This tutorial will use Google AI Studio as an example, demonstrating how to obtain a Gemini API key. Google provides powerful models like Gemini 2.5 Pro and Flash, often accessible via a generous free usage tier to start.

Next, ensure you have Visual Studio Code installed on your system. If not, download and set it up.

Within VS Code, navigate to the Extensions marketplace (found in the left sidebar), search for “Roo Code,” and install it. (For comprehensive documentation, visit: https://docs.roocode.com/)

Connect your API key by configuring the Roo Code extension. Click the Roo Code icon (which resembles a kangaroo) in the sidebar. Access its settings menu, select your AI provider, input your API key, and choose a specific model. For initial basic coding tasks, Gemini 2.5 Flash is recommended due to its performance and current favorable free tier. For more complex requirements, consider Gemini 2.5 Pro or Claude 3.7 Sonnet (note: Claude requires a different provider than Google).

For an optimized workspace layout in VS Code, consider repositioning the Roo Code window, perhaps by dragging its icon to the right side of your display.

Prepare your development environment by creating a new directory for your project on your computer. A highly recommended, though optional, step is to initialize a Git repository within this directory (using git init via the terminal or VS Code’s integrated features) and potentially link it to GitHub. Alternatively, you can postpone the Git setup and seek assistance from the AI later.

Further customize Roo Code settings if desired. This might include creating a .roorules directory inside your project folder to house custom instructions for the AI (refer to the Roo Code documentation for specifics). A beneficial starting set of rules can be found here: https://gist.github.com/ruvnet/b6cae644453de36403f991a0bfc61635. While not mandatory, this is strongly advised. Consult the full Roo Code documentation for complete details.

Initiate the coding phase. Switch Roo Code to its “Orchestrator” mode or the equivalent state designed for planning and execution. Present a high-level description of your desired application (e.g., “Create a simple Flask Python web server that serves a single HTML page”). Review the plan proposed by the AI and provide feedback as necessary, but initially prevent it from generating code. First, instruct it to generate an architectural plan and save this document for future reference. This plan serves as a guide, helping the AI adhere to the agreed-upon design path. Once the plan is saved, ideally as a markdown file, review it, and then allow Roo Code’s Orchestrator mode to proceed with generating the code step-by-step.

And that covers the fundamental steps to start using AI assistance in coding through Vibe Coding! Hopefully, you found this introduction helpful. Feel free to explore my projects on GitHub at https://github.com/eoko-dev/.

  • eoko