A central AI team is becoming a bottleneck as business units request similar projects. How can a federated operating model help without abandoning governance?

1. Require the central team to manually perform every user interaction forever.
2. Provide shared standards and reusable capabilities while enabling qualified domain teams to own use cases.
3. Let every unit ignore data and risk policies entirely.
4. Eliminate all shared knowledge so each unit starts from nothing.