6 Guidelines for Governing AI

For the initial ten years of my career, I worked as a product manager and data analyst. I wrote database queries, built statistical models, and created data pipelines. Later, I took on leading enterprise AI transformation at Lowe’s. My role now involves not selling AI, but using it to provide valuable insights when customers need them. In retail and other customer-facing industries, AI can help people with everyday questions and guide them to relevant products or services.

The work has shifted from building AI systems to governing them. This is called the "governor shift." Business operators need to define the intent, principles, and boundaries of AI systems, rather than just executing tasks. Many organizations have invested in generative AI, but few have seen measurable profit and loss impact. The majority are stuck in an "administrator trap," constantly transferring data between systems, but not able to make informed decisions.

Instead of following rules, businesses should write "principles" in priority order. These principles help the AI system settle conflicts automatically. The business must also define decision rights, stating who or what may make specific calls. Governance should be written into the code as machine-readable instructions. This includes a constitution, doctrine, and playbook, ensuring the AI agent follows these instructions automatically.

A trust thermostat is recommended instead of a trust switch. Every decision an AI agent makes should have a confidence score measured against the principles. If the score is above a certain threshold, the system proceeds alone; if not, a human decides. This approach maintains transparency, auditability, and explainability, creating a "glass box" for the AI system. Before governing, it's essential to provide the system with a complete context, as it cannot "read" human posters or understand the full picture.