This session combines a practical AI risk framework with a side-by-side demonstration of trained and untrained agents completing the same accessibility-related production tasks. It shows attendees why generic prompting is not enough, and why reliable AI use depends on standards, context, review discipline, and deliberate training.