Spec-driven development with AI: Get started with a new open-source toolkit (github.blog)

Spec Kit introduces a spec-driven development approach that redefines how AI coding agents like GitHub Copilot, Claude Code, and Gemini CLI are used in software projects. Instead of treating these agents as search engines to generate vague code snippets, Spec Kit centers the engineering process around a living, executable specification. This spec captures user intent clearly and evolves alongside the project, guiding the coding agent through four structured phases: specifying user goals, planning technical architecture, breaking down tasks, and iterative implementation with continuous human validation at each step. This methodology addresses longstanding challenges in AI-assisted coding where generic prompts often lead to incomplete or misaligned code, especially in mission-critical or legacy projects. By anchoring the AI's output to a precise, evolving spec and a detailed technical plan, Spec Kit enhances reliability, ensures compliance with organizational constraints, and enables safe incremental development. It is particularly impactful for greenfield projects, complex feature additions, and legacy system modernization by separating “what” to build from “how” to build it, allowing rapid iteration without costly rewrites. Beyond improving AI coding accuracy, Spec Kit marks a shift in software development philosophy: intent becomes the source of truth rather than the code itself. As specs turn executable, this process promises to harness AI’s pattern recognition strengths more effectively. Open sourced to foster community adoption and refinement, Spec Kit aims to integrate with developer workflows more seamlessly and scale across organizations, empowering developers to translate human creativity into robust, maintainable software.