{
  "id": 9385279,
  "title": "Claude Thinks, GitHub Copilot Executes: How We Structured AI-Assisted Development on a Real Project",
  "url": "https://urgent.news/2026/09/23/claude-thinks-github-copilot-executes-how-we-structured-ai-assisted",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-23T17:57:46.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/mikaelkrief2/claude-thinks-github-copilot-executes-how-we-structured-ai-assisted-development-on-a-real-project-dpo"
  },
  "original_language": "en",
  "account": "This article outlines a structured approach to developing software using AI assistance from GitHub Copilot. The methodology begins with the AI tool Claude, which acts as a \"thinking partner\" to refine feature ideas, create UI mockups, challenge architectural decisions, and maintain consistent coding conventions. Claude works at the level of intent and strategy, while GitHub Copilot, acting as an \"execution engine,\" generates the actual code following the precise prompts provided by Claude.\n\nA key principle of this methodology is to separate the two roles of Claude and Copilot. Claude never directly touches the codebase, while Copilot only generates code after receiving instructions from Claude. This clear boundary prevents mixing architectural decisions with code generation and maintains consistency in the development process.\n\nThe article emphasizes the importance of structuring prompts in a specific format that covers one functional scope and technical layer at a time. Each prompt is stored as a *.prompt.md file, version-controlled in Git, and triggered via a dedicated button in VS Code to keep the prompts and their execution traceable.\n\nTo ensure visual consistency, the methodology introduces versioned UI reference files that define the design system for each application module. These reference files are used as constraints in the prompts, ensuring that the generated code adheres to the established design guidelines. The article also describes how the Figma design tool can be connected directly to Copilot through a custom MCP server, eliminating the need for manual transcription of design specifications into prompts.\n\nOverall, the described approach provides a repeatable, structured process for AI-assisted software development that leverages the strengths of both Claude and Copilot while maintaining clear boundaries and version control for prompts.",
  "summary": "A hands-on experience report on a structured, repeatable, AI-assisted development approach We built a full-stack web application — .NET backend, Vue 3 frontend, PostgreSQL database, hosted on Azure. The platform handles complex business flows: payments, electronic invoicing, AI-based candidate scoring, and automated multilingual translations. From the start, we made one defining choice: not to…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}