A mental model for building trust in coding agents through feedforward guides, feedback sensors, and iterative harness engineering.