I haven’t personally dealt with large-scale failures here, but with more automation and AI writing state directly, it feels like reversibility matters more. Curious whether teams actively design for rollback, or mostly rely on backups and manual fixes.