Human Approval
COMPASS is designed so the Agent runs the flow and people judge the outcome—data writes require authorized human confirmation.
What you can do
Human approval is the safety gate for AI collaboration. The copilot can query, analyze, and propose changes, but cell writes, batch adjustments, and schedule releases need your explicit confirmation in the UI.
- Spreadsheet writes: Agent-proposed field changes require batch or row confirmation;
- Schedule release: submit and lock in shared workspaces by authorized users;
- Conflict resolution: uncommitted changes and conflict prompts need human decisions;
- Audit trail: key confirmations can align with permissions and collaboration workflows.
Recommended confirmation flow
With AI: select a range → Add to chat → read suggestions and explanations → confirm writes → validate in Gantt/capacity → lock or submit.
- Pilot on a small scope first—single row or vendor—before expanding;
- In shared workspaces, watch the status bar for uncommitted changes;
- Lock key results after confirmation to prevent accidental edits.
Human-in-the-loop principle
Algorithm and AI output assists decisions—it does not replace shop-floor judgment. COMPASS uses permissions, audit, and human approval to embed AI safely on the production floor.