01Start with a moment of workDefine one recurring event, the user handling it, the current pain and the outcome that should improve before choosing an AI model or interface.
02Ground before generatingRetrieve approved, permitted and current sources, match the correct case, then apply explicit rules and approval limits before preparing guidance.
03Make boundaries visibleShow sources, case facts, rules, permissions, refusals, escalation and what the co-pilot may prepare but may not decide or execute.
04Close the operational loopCapture the action, override reason and outcome so knowledge gaps, policy defects and guidance quality can be measured rather than guessed.
