Related to this - as agents get more powerful, governance becomes critical. We're building DeltaOps (delta-ops-mvp.vercel.app) - human-in-the-loop for AI agents. Label a GitHub issue → agent works → you approve → it merges. Would love feedback from others thinking about agent safety!