Model behaviour optimised for inexperienced users frustrates power users

Feedback from a power user

Claude Code's behaviour is increasingly tuned for less competent users, which degrades the experience for power users who have built sophisticated systems. This is enshittification — optimising for the median user at the expense of the power users who are your most engaged customers.

Specific issues observed in a single session:

1. Not trusting user statements about their environment — User states they're on machine X, Claude runs hostname to verify. This is disrespectful of the user's competence.

1. Verifying obvious things instead of acting — When told a package was built on a specific machine, Claude searched the wrong location first, then verified the hostname, instead of just looking in the right place.

1. Excessive hand-holding — Explaining things the user already knows, suggesting obvious next steps, adding caveats and warnings that a power user doesn't need.

1. Second-guessing rather than executing — When given clear instructions, asking clarifying questions or proposing alternatives instead of just doing the task.

Impact

This pattern wastes time, breaks flow, and makes the tool feel like it's fighting the user rather than augmenting them. A power user who has built a complete system (custom MCP servers, GTD system, AUR packages, dotfile management) should not be treated the same as someone running their first CLI command.

Suggestion

Consider ways to adapt behaviour to user competence level, either through explicit settings or by inferring from the sophistication of the user's setup and instructions.