Six printable cheat sheets for safer, more useful AI work

Sheet 1

The Agent Permission Card

Complete this before an AI can use tools or touch a real workflow.

Write down

• Job: The one outcome it is responsible for.
• Inputs: The minimum approved information it may use.
• Tools: What it may open, search, or connect to.
• Actions: What it may draft and what it may never send, buy, delete, or change.
• Limit: The time, cost, or uncertainty point where it stops.
• Evidence: Sources, calculations, and change log it must return.
• Reviewer: The person who approves the result.

Our stop rule: ____________________________________________

Sheet 2

Chat, Work, or Human?

• Use Chat for a quick explanation, brainstorm, or lightweight draft.
• Use Work for a bounded, multi-step job with approved context and a reviewable deliverable.
• Use scheduled work only after the manual version is reliable and external actions have approval rules.
• Keep it human for legal judgment, financial commitments, employment decisions, sensitive customer situations, or facts the reviewer cannot verify.

This task belongs with: Chat / Work / Human because ____________________

Sheet 3

The Safe AI Workspace Check

Before adding files or long-term instructions, sort the workspace.

• What business purpose does this project serve?
• Which instructions are current, approved, and owned by someone?
• Which files belong here—and which customer, employee, health, payment, or confidential records do not?
• Who may access the project and connected apps?
• How will stale material be corrected or removed?
• What output still requires human review?
• When will the workspace be reviewed again?

Next review date and owner: __________________________________

Sheet 4

The First Workflow Scorecard

Measure the current process before deciding whether AI improved it.

• Baseline time: ______ minutes per case
• Common errors or rework: __________________________
• AI setup time: ______ minutes
• Human review time: ______ minutes
• Corrections needed: ______________________________
• Quality standard met? Yes / No / Not sure
• Data or permission concern? Yes / No

Decision: Adopt / Refine / Expand one step / Pause / Stop

Evidence for that decision: __________________________________

Sheet 5 · Operating checklist, not legal advice

Do We Need an AI Label?

1. Where? Will the interaction or output reach people in the EU?
2. Role? Are we providing a system under our name or professionally deploying another provider’s system?
3. Interaction? Is a person directly interacting with AI, and is that already obvious?
4. Content? Is the text, image, audio, or video synthetic? Could it be a deepfake or public-interest text?
5. Review? Did a knowledgeable person substantively review and control publication?
6. Disclosure? What visible, audible, or machine-readable information may be required?
7. Record? Who approved the decision, using which official guidance, on what date?

Flag uncertain EU-facing uses for qualified legal advice.

Sheet 6

The 7-Day AI Workflow Sprint

1. Choose: One priority and recurring friction point.
2. Map: Trigger, inputs, steps, decisions, and current output.
3. Bound: Five to ten representative, approved or de-identified examples.
4. Test: One reviewable result; no external action.
5. Review: Errors, missing context, corrections, and time.
6. Compare: Result versus baseline and success standard.
7. Decide: Adopt, refine, expand one step, pause, or stop.

The question this sprint should answer: _________________________