Programme discipline, applied to AI work.
Short essays. Each one starts from a case study on this site and asks what it means for building and running AI tools.
- Your AI reviewer needs a light test
Three checks that all trust the same record are really one check. That's true of a telecom circuit inventory, and it's true of an AI evaluation.
Drawn from: The Light Test · Redline: contract review that shows its work
- Separate what varies from what doesn't
Before deciding where AI goes in a piece of work, watch the work. Most of it usually doesn't change, and the part that doesn't change doesn't need a model.
Drawn from: The Macro · Redline: contract review that shows its work
- Design so the mistake is hard to make
Asking a reviewer to catch every AI error is the same as asking an engineer never to skip a step. Both fail. Mistake-proof the process instead.
Drawn from: Poka-Yoke · Vendor incentives · Redline: contract review that shows its work