About this Video

• Video Title: My AI Coding Workflow + Free APIs: How I DO AI CODING! (RooCode, Gemini 2.5, T3 Stack, Requesty)
• Channel: AICodeKing
• Speakers: AICodeKing
• Duration: 00:13:20

Introduction

This video details the speaker's AI coding workflow, outlining the tools and models used to transform an idea into a functional application. The speaker focuses on efficiency and cost-effectiveness, highlighting specific strategies and tools.

Key Takeaways

• Workflow Overview: The speaker's workflow begins with brainstorming and creating a basic mockup (Figma or Excalidraw). ChatGPT is then used to refine the UI.
• Model Selection: Gemini 2.5 Pro (both free and paid tiers) and DeepSeek V3 are the primary LLMs used, with Requesty managing API keys and model switching for cost optimization. The speaker leverages Requesty's features for logging, fallback models, and custom prompts to enhance efficiency and reduce costs.
• RooCode and T3 Stack: RooCode is used for project setup, and the T3 stack is preferred for Next.js app development due to its built-in type safety and TRPC.
• Tool Integration: Requesty's seamless integration with RooCode, including features like logging, custom prompts, and fallback models, is highlighted as a key efficiency booster.
• Coding Knowledge: The speaker emphasizes the necessity of fundamental programming knowledge, even when using AI tools.

The AICodeKing's video details their AI-powered coding workflow, emphasizing efficiency and cost-effectiveness. The process begins with brainstorming an app idea, followed by creating a simple mockup using Figma or Excalidraw. This mockup is then fed to ChatGPT to generate a more polished UI design.

The core of the workflow revolves around two primary large language models (LLMs): Gemini 2.5 Pro (both its free and paid versions) and DeepSeek V3. The choice between them depends on the task's complexity, with DeepSeek V3 favored for simpler tasks to save costs. However, managing multiple model API keys and credits is streamlined using Requesty, a platform the speaker highly recommends. Requesty provides a $1 sign-up credit and additional bonuses for larger top-ups, which are added incentives to its other functionalities.

Requesty's key features exploited by the speaker are: seamless integration with RooCode (the code generation tool used); a detailed dashboard providing insights into model performance and token consumption; and the ability to set API key-based policies including logging, fallback models (allowing automatic switching between models if the primary one fails), and custom prompts. The custom prompt feature allows modification of the system prompt for each API key, leading to considerable token savings and optimized code generation. The speaker uses a specific custom prompt (the "GOU coder" prompt) within Requesty to significantly reduce token usage in RooCode.

For project initialization, the speaker primarily uses the T3 stack for Next.js apps, appreciating its type safety and TRPC integration. For simpler mobile apps, Expo is preferred. Once the project is set up, initial prompts focus on cleaning up the boilerplate code, removing unnecessary pages, and renaming the project appropriately.

Further development involves iterative prompting of the chosen LLMs to build the application, interspersed with manual coding for tasks beyond the AI's capabilities. The speaker stresses that while AI tools are valuable, a foundational understanding of the programming language is crucial to validate the AI's output and handle complex tasks. Superbase is the preferred database, easily integrated via RooCode, and the speaker uses Kodium for code completion. Finally, the speaker avoids tools like Windsurf or Cursor due to limitations with model performance and configuration options. The video concludes with a call to action, encouraging viewers to share their AI workflows and subscribe to the channel.