# Agent actions use risk tiers and domain ownership

Agent operation in the lakehouse is governed by two boundaries: action risk and domain ownership. Risk tiers distinguish observation, reversible workspace actions, controlled operational actions, approval-gated production changes, and forbidden or break-glass actions; domain ownership determines where an agent is allowed to act. This prevents the data-agent mesh from treating autonomy as a global permission while still leaving room for increasingly capable agents inside well-owned data product boundaries.