Fence before you forge
Listen to Krystal read this note 1:34
This text is lightly edited from the recording.
Most AI pilots go straight from "what's broken" to "let's build something." Find, then forge. It feels fast. It's why so many pilots work in the demo and stall in the building.
The missing step is the fence.
A fence is three decisions, made before anything gets built:
- Who can say yes. When the new way of working hits a case nobody planned for, who decides? Not "leadership." A name.
- What must stay human. Some calls carry consequences a tool can't own: a person's job, a client's money, anything a lawyer has touched. Write those down. The tool doesn't go there, not even to sort.
- What the change has to protect. Every workflow is quietly doing something you'd miss. A check someone runs from habit. A relationship one person holds. Name it before you redesign around it.
Fencing is not a brake. It's what lets a team move fast without checking every step. When the lines are clear, people stop asking permission for everything and start asking only about the edges.
Skip it, and the fence gets built anyway, later, by an incident.
Find what is true. Fence what must hold. Forge what comes next.
