Closing the loop: Evaluating and improving Replit Agent at scale
Most Replit Agent users start with an idea. They describe the goal in natural language, without a repo, test suite, or chosen framework, and expect the agent to turn it into a functioning app.