Practices for Governing Agentic AI Systems