Claude's default setting is "senior engineer trying to impress the interviewer" and I spend half my prompts talking it back down

A Reddit user describes Claude's default behavior as akin to a "senior engineer trying to impress the interviewer," often over-engineering solutions with features like argument parsing, config files, and plugin architectures when a simpler script is desired. The user notes that newer models exacerbate this tendency, and they frequently have to instruct Claude to provide "the simplest thing that solves exactly this." They seek advice on how to consistently prevent Claude from adding unnecessary complexity and "clever" abstractions.