Practice — Hot · The AI Policy Memo
Skills focus. Turning thirteen weeks of verification discipline into a
policy other people can follow.
Ungraded, open-ended. The final Hot practice of the course; write it like it.
Scenario
Lantern & Pine's leadership enables Copilot and the AI visuals org-wide, and — having watched your work all year — asks you to write the one-page usage policy for the analytics team. Two constraints from the sponsor: it must not be a ban ("we bought it; people will use it"), and it must not be toothless ("'be careful' is not a policy").
Requirements
- Policy by output type. For each of: generated DAX, generated report pages, generated prose summaries, Q&A answers, influencer findings, and forecasts — one rule stating what must be done before the output ships. Each rule must be checkable (a reviewer can verify compliance) and grounded (name the course evidence that justifies it — the gauntlet artifacts, the two averages, the fabricated summary figure).
- The escalation clause: what happens when generated output disagrees with a verified figure. (You know the answer; write it as procedure.)
- The enablement half. A policy that only restricts will be ignored. Name two places the team should lean into the tools (first drafts, synonym-driven Q&A for stakeholders…) and what makes those uses safe.
- Under 300 words, titled, signed. The sponsor reads it, then the whole team does, forever.
Reflection
- Which single rule in your policy would have caught the most gauntlet artifacts?
- The course ends with you writing policy rather than formulas. Why is that the right final exercise — what is the actual product of this course, per its own closing?