Spec-driven development: Using Markdown as a programming language when building with AI

Abstraction: Writing apps entirely in Markdown specs then AI-compiling to production code

Key points:

• Workflow: author app entirely in main.md (Markdown spec), invoke compile.prompt.md to have GitHub Copilot "compile" it to Go; developer rarely edits Go directly
• Four key files: README.md (user docs, imported into spec), main.md (the spec/source), compile.prompt.md (repeatable compile prompt), and main.go (generated artifact)
• Markdown spec includes variables, loops, conditionals, database schema, and GraphQL queries — effectively a declarative programming language in plain English
• lint.prompt.md cleans up the spec for clarity; Copilot treats English as a programming language and normalizes synonyms (pull/get/fetch → one term)
• Main challenge: compiled output grows as spec grows; next step is splitting spec sections into separate Go modules
• Author notes spec-driven workflow improves continuously with each Copilot agentic update; testing remains essential but is not yet integrated

Connections: Github · Github Copilot · Agentic Coding · Prompt Engineering

Source: https://github.blog/ai-and-ml/generative-ai/spec-driven-development-using-markdown-as-a-programming-language-when-building-with-ai/