A three-run benchmark of Claude Code building the same backend across Encore, Express, Fastify, Hono, and NestJS, measuring not just whether tests pass but whether the code an AI agent ships is fit for production.