Frameworks · AI Don't Fix Stupidity

Governance Tiers

Governance Tiers 1 to 3 classify how consequential a capability's decisions are and how much human judgement must stay in the loop.

Tiers

  1. Low (meeting summaries, data analysis)
  2. Medium (churn scoring, pricing recommendations)
  3. High (credit, hiring, regulatory decisions)

Tiers classify the decision; the Tool Risk Ladder classifies the tool. They are independent and both are assessed for every capability, and "confusing them is the single most common architecture mistake". What moves a capability up is consequence and reversibility, not money: "A high-value recommendation a human checks is Tier 2. A low-value decision about a person that nobody checks is Tier 3." The book also retires "deploy at 95% accuracy" and sets a transparency floor: explainable, confirmable, auditable, bounded, escalatable.

Use it on Monday

"Governance you add later is governance you pay for twice."