Your AI-built app works beautifully — on your machine

AI code generation tools operate in a mental model of `localhost:3000`. Every piece of code they produce assumes the dev server is running, hot module replacement is active, and all services are available on the same machine. This creates dozens of invisible dependencies on the development environment that only surface when you try to deploy. The most common: `process.env` variables that exist in your `.env` file but were never configured in your hosting platform, API URLs hardcoded to `localhost`, and database connection strings that point to a local instance.

Build-time versus runtime is a distinction AI tools fundamentally don't handle well. In development, your framework serves everything dynamically — nothing is pre-built. But production deployment for most frameworks means a build step that statically analyzes your code, tree-shakes unused imports, and bundles everything. AI-generated code frequently breaks at this stage because it uses dynamic imports incorrectly, references browser APIs during server-side rendering, or has circular dependencies that the dev server silently resolves but the build tool cannot.

CORS errors are the other silent killer. In development, your frontend and backend run on the same origin or your browser is configured to be permissive. In production, your API is on a different domain or subdomain, and suddenly every fetch request fails with an opaque CORS error. AI tools rarely generate proper CORS configuration because the problem simply doesn't exist in the environment they're modeling.