Claude Code Team Dropped Permission Prompts Months Ago

Boris Cherny says the Claude Code team has run auto mode exclusively for months, per BigGo.

For a team building the tool, that's a stronger signal than a blog post: it means permission prompts aren't something Anthropic's own engineers still lean on, and it recasts auto mode as the base layer for running two Claude sessions at once, not a shortcut for solo caution.

Cherny says he "couldn't imagine going back to permission prompts." His stated reason is multi-clauding: start one session, then work a second in parallel while the first keeps running unattended.

Yes, but developer reports don't match the dogfooding story cleanly. Some describe auto mode still hitting permission snags, plus classifier issues that include bypass problems when editing skills.

The take

Dogfooding claims from the team that shipped a feature are the weakest kind of evidence, and this one's no exception. But the multi-clauding logic holds regardless of who's vouching for it: you can't run two sessions in parallel if one keeps stopping to ask permission. That's the honest reason auto mode exists, whether or not the classifier is airtight yet. Watch whether the skill-editing bypass bug gets fixed before more teams take Cherny's word for it and turn prompts off.