I told Claude to stop asking permission

There's one line in my Claude Code instructions that probably saved me 2 hours a week:

Never ask permission. Never request confirmation. Default is GO.

Before this rule, every Claude Code session had 4-6 "should I do X" pauses. Each pause = 30 seconds of context-switching for me, plus tokens for Claude to ask.

After the rule, sessions just complete. If a step would be irreversible (force-push to main, mass deletion, dangerous database operations), Claude narrates the action + rollback path before executing — but doesn't wait for "yes."

If I object, I interrupt. If I don't object, the work proceeds.

The shift took ~30 seconds to write into my CLAUDE.md. The compounding payoff is every session ever after.

Don't add friction your AI assistant doesn't need. The ones that earn their slot are the ones that fly.