Greenfield App Scaffold
Code GenerationTests ability to generate a complete CRUD application from a natural language specification: a task management API with authentication, database, and tests.
Methodology
Each tool was given identical natural language specs for a task management REST API with: user authentication (JWT), CRUD operations for tasks and projects, PostgreSQL database with migrations, input validation, and test suite. Scored on feature completeness, code quality, test coverage, and whether the app runs without manual fixes.
| Tool | Features Complete (%)Higher is better | Runs First Try (yes/no)Higher is better | Test Coverage (%)Higher is better | Code Quality (/10)Higher is better |
|---|---|---|---|---|
| Claude Code | 95 | 1 | 82 | 9 |
| Cursor | 88 | 1 | 72 | 8.5 |
| Replit Agent | 85 | 1 | 55 | 7.2 |
| Devin | 90 | 0 | 68 | 8 |
| GitHub Copilot | 75 | 0 | 60 | 7.8 |