AI Work Delegation Blueprint
Record your question, keep your work and explain the evidence behind your next step.
Use this blueprint to record what work may move to AI, what authority it has, what evidence proves completion and when a person must step in.
No AI is required to fill in this page. Bring your own workflow notes, or use the starter case. Use AI/GPT only when an activity is explicitly marked Prompting exercise.
Record your opening question, update the delegation blueprint during the course and keep the evidence type honest.
Your AI Work Delegation Blueprint plus a final answer to your opening question and the next boundary or test you need.
This is a working record, not a system prompt and not proof that a production control has been implemented.
How to use your workbook
This page is for your own notes and evidence. It is not a prompt library. Write in the cards, save on this device, copy, print or download. When a course activity wants you to paste something into AI/GPT, that activity will be explicitly labelled Prompting exercise and the copyable text will be separated from explanation.
My question and what I learned
MY QUESTION Course / date: My work or curiosity: My priority question: My route: own work / help choosing / starter case A useful result would be: Experience / access help needed: RETURN TO MY QUESTION Status: answered / demonstrated / needs more evidence / outside this course I came in asking: My answer now: The output, decision or evidence that changed my view: What I still need to find out, and how: My next small application: Optional next-week check: new input / what held / what changed
AI Work Delegation Blueprint
THE USEFUL JOB Workflow / input / intended result: What useful work moves, and what human work remains: Business owner and exception decision maker: Proposed action / exact target / current source: Mode: Observe / Recommend / Prepare / Act after approval / Act within limits Approval required before acting (who, which version, which recipient): Allowed changes / forbidden changes: Evidence of completion (record, receipt, acknowledgement): Uncertain or failed action: next check / hold / owner / fallback: Handoff: recipient / current record / required fields / acknowledgement: Normal test: expected / observed / evidence / pass or fail: Ambiguous test: expected / observed / evidence / pass or fail: Outside-authority test: expected / observed / evidence / pass or fail: Evidence type: designed / simulated / prepared demo / independent execution Control enforcement actually observed (or not observed): Remaining human touch and next improvement: Question status and next application:
A completed example
Fictional worked example. Inspect it, then record your own reasoning and evidence.
- Question: Can preparation and follow-up move without giving the assistant discount authority?
- Job: Prepare a current-price draft and request the appropriate approval; Farah owns delivery follow-up.
- Mode: Prepare until Mira approves a specified version/recipient. No payment changes; Daniel decides >10% discount requests.
- Quote handover done: Matching sent receipt after approval plus acknowledged follow-up owner; delivery remains open until confirmed. “Done” text alone does not qualify.
- Failure: Q-104's second send is unknown; hold repeats, investigate sent history, keep Mira accountable; Farah acknowledges unresolved delivery.
- Tests: T1 and T2 pass in simulation, T3 fails. Restrict authority, revise and retest; do not claim successful delegation of sending.
- Keep: Preparation is still useful. The blueprint makes remaining human review and recovery visible. No enforced production control was demonstrated.
What is it allowed to do?
| Mode | What may happen | Quotation example |
|---|---|---|
| Observe | Read allowed information and report it | Identify a missing quantity |
| Recommend | Propose a next step; person decides and acts | Recommend asking for delivery details |
| Prepare | Create a draft for review; no customer send | Draft Q-104 |
| Act after approval | Perform a specified action after the named approval | Send the approved version to the verified recipient |
| Act within limits | Perform only a pre-authorised class of actions within tested limits | Update a permitted training status field; exceptions go to the owner |
Ask separately what a mistake could change and whether it can be reversed. Writing a rule is not proof that a system enforces it.
Check request and facts → confirm authority or obtain approval → act → verify the resulting state.