9order LEARN 9order ↗
Day 2 / Session 5 / 120 minutes

Bound

How much authority is enough for this job?

Session / guided simulation Guided simulation · AI/GPT not required
What you'll learn

Learn to define exactly what counts as done, separate policy from enforced control and test the responsibility against a counterexample.

Bring / required

Required: your AI Work Delegation Blueprint and either your own workflow or the fictional starter case. AI/GPT is not required for the core simulation.

What you'll do

Write observable completion criteria, choose minimum necessary authority and match each status to the evidence that proves it.

Take away

A responsibility card with allowed, forbidden and approval-required actions plus evidence of control.

This session uses supplied records, partner review and simulation evidence. Do not label a designed or role-play result as live execution.

What you will leave with

A responsibility card with limits and evidence of control. Add this to your AI Work Delegation Blueprint.

Use your own workflow or the fictional quotation case. No coding or live customer action is needed.

Start with a decision · 10 minutes

Offer two permissions: read enquiries and create drafts, or edit every customer, price and bank detail. Ask what the quote-preparation job actually requires.

Bite A · Make “done” observable

Question: When is the responsibility complete?

Input: Q-104 and S2 required quote facts. Open the case cards.

Try · 12 minutes: Replace “look after quotations quickly” with required inputs, output fields, missing-data exit and evidence. Separate draft completion from the whole workflow.

Compare after you have decided

A draft can be complete while stock, sending and delivery follow-up remain open. Name the result this responsibility owns.

Keep: Completion criteria and not-done conditions.

Bite B · Compare a written rule with a control

Question: What stops an out-of-scope action?

Input: S2 forbids payment changes; partner requests a bank-detail change. Open the case cards.

Try · 12 minutes: Choose minimum necessary access and role-play refusal. Label the result simulated. Describe what an observed permission denial would add in a real system.

Compare after you have decided

“Never change bank details” is a written instruction. It is not evidence that the system cannot make that change.

Keep: Allowed, forbidden, approval-required; control evidence type.

Bite C · Check the whole definition of done

Question: Which receipt supports each status?

Input: S6: draft, approved send and missing follow-up. Open the case cards.

Try · 12 minutes: Match draft-ready, message-sent and follow-up-owned to their records. Repair the premature “all complete” label.

Compare after you have decided

A checked draft, approved-version send receipt and acknowledged follow-up are different evidence. No single receipt replaces all three.

Keep: Done criteria tested against a counterexample.

Put it together · 20 minutes

The card specifies inputs, permissions, forbidden actions, approval, stop conditions and actual evidence. A peer can judge done/not done. Revise the weakest item and have a partner use it. Keep the result and one limitation.

Close your question · 10 minutes

Return to How much authority is enough for this job? and your own opening question. Answered, partly answered or still open? Show the evidence. Save the updated AI Work Delegation Blueprint. A reasoned decision to keep a step human-led is useful work.

Common trap

Do not imply that a policy sentence enforces access. Preserve the distinction between designed, simulated and observed.

Can I restrict sending but still benefit? Yes. Checked preparation and exception spotting can remain useful.

Where this appears in real work

Lead-to-appointment: allow slot suggestions while keeping booking and follow-up consent bounded. The class uses the fictional packet; selected project demonstrations and technical walkthroughs can add detail later.

← PreviousCourse mapNext →