Coding agents, particularly LLMs like Claude Code, can effectively serve as first-time user testers for command-line interfaces (CLIs) and websites. Their training on extensive software data allows them to have expectations about how interfaces should function. When they encounter confusion, it often signals that the interface is too complex for new users. This approach can help catch usability issues early, as traditional user testing can be slow and costly, often leading engineers to skip it altogether. In practical examples, the author used Claude Code to test a CLI designed for complex file formats. Initially, the agent struggled significantly, misinterpreting flags and making errors in command execution. After reviewing its feedback, the author simplified the interface, reducing the steps needed to complete a task from about 20 to 11. A similar testing method was applied to a web interface, where the agent faced difficulties navigating similar-looking buttons and ambiguous configurations. Feedback included suggestions like removing redundant buttons for clarity. The article proposes using agents as a form of regression testing for user intuition. By measuring how many steps an agent takes to complete a task, engineers can gauge an interface's intuitiveness. For example, a test could check that an agent successfully completes a sign-up task in 15 steps or fewer. While agents can’t replace human feedback due to their limitations in understanding aesthetics and potential technical glitches, they provide a cost-effective method for identifying usability issues before real users engage with the software.