Invisible data breaks AI agents

Invisible data like exceptions, approvals, and undocumented institutional knowledge poses a greater danger to AI agents than bad data. A metadata audit at a large social media company found most production datasets were orphaned or unowned. This blind spot undermines AI reliability in large institutions.