Framework 5 · Intelligence Abundance Toolkit

Constitutional Principles vs. Compliance Rules

The best AI governance document I've seen was two pages. The worst was forty-seven. The forty-seven-page version was thorough, compliance-friendly, and completely useless in novel situations, which is exactly where AI governance gets tested.

The Failure Mode This Addresses

What goes wrong

Governance documents too rigid to handle novel situations. AI moves faster than any compliance framework can update. A 47-page document covers every known scenario with specific rules and fails completely when something new comes up. Novel situations are exactly where AI governance gets tested.

Compliance vs. Constitutional

The difference isn't length. It's what the document is designed to do.

Compliance Rules Look Like This

"Employees may not use AI to draft external communications without manager approval. AI-generated content must be reviewed before publication. Customer data may not be entered into unapproved AI tools." These rules cover known scenarios. They fail on novel ones. And every new AI capability creates a new novel scenario.

Constitutional Principles Look Like This

"We don't ship anything we wouldn't put our name on, regardless of how it was produced. Our customers' data is treated with the same care we'd want our own data treated with. We use AI to do better work, not to do less work. The quality bar doesn't move." These principles cover known and unknown scenarios. They get sharper with use.

The Two-Page Constitution Structure

1. Identity Anchor

One paragraph. Who you are and the standard AI doesn't change. This is the sentence every person in your organization can use to evaluate any AI output. Example: "We produce work we'd be proud to put our name on. That standard doesn't change based on who or what produced it."

2. Decision Rights

One page. Three tiers: individual discretion, manager approval, executive approval. Be specific about what goes in each tier. Ambiguity here creates the escalation problems the constitution is designed to prevent.

3. Quality Standards

Three to five bullet points. Not rules about process. Standards about output. What does "good enough" mean? How do you evaluate AI-generated work against the same bar you'd apply to human-generated work?

4. Escalation Path

One paragraph. When someone isn't sure whether something is authorized: one person, one decision point, fast response. If the path takes more than 24 hours, it won't be used.

The Practitioner Note

The test: could every person in your organization use this principle to make a good autonomous decision in a situation the document has never contemplated? If no, it's not a principle. It's a rule that hasn't been written yet. Write your constitution with that test in mind.

Three Questions Before Writing Your Constitution

Use these to assess your current governance posture:

  1. Could every person in your organization use your current AI governance to make a good autonomous decision about a situation the document has never contemplated?
  2. Does your governance answer "what does good AI output look like" or only "what process was followed"?
  3. When someone isn't sure whether something is authorized, what do they do? Is there one clear, fast answer?

The Full Practitioner Tool

The Intelligence Abundance Toolkit includes the Two-Page AI Constitution Template: a fill-in-the-blank format with annotated examples for each section, a constitutional consistency test, and guidance on how to migrate from a compliance document without losing the protections that compliance rules provide.

The Intelligence Abundance Toolkit

Seven frameworks. Seven worksheets. 90-min session guide, 12-month roadmap, and 11 AI prompts. $97.

Get the Toolkit →