Replit building complete applications from natural language with frontend, backend and UI is the full-stack generation claim worth testing

The iterative feature building workflow, describing the app and having the AI build out features progressively through conversation, is the development model that changes the non-technical founder experience. Adding authentication, then a database, then a dashboard, through description rather than through code changes who can produce a functional application independently.

The proactive feature suggestions based on the prompt being offered as the build progresses is the product development assistance that surfaces considerations the user may not have thought of. An AI that says "you'll probably want user authentication for this" before you have to ask is a different development partner from one that only implements what you explicitly specify.

The dynamic dashboards, to-do lists and planners with drag-and-drop functionality being demonstrated as typical outputs shows the product range that conversational generation currently handles reliably.

At what point in complexity or specificity does the natural language generation start producing output that requires significant manual correction rather than minor refinement?