Using the model on real work

A framework that only describes behavior is a vocabulary. This one is meant to be used before the work, not only after it.

About 20–30 minutes · 13 named steps · research detail is optional

The findings above concern patterns across people. This step turns the model into decisions for one piece of work. Before opening AI, write three short answers. They determine which principles matter, which modes fit, and what evidence you will require.

Each question below includes a student example and a professional example so the prompt is concrete rather than ceremonial.

Before you start

1.What am I trying to keep?
Name the capability or judgment that must still belong to you after the artifact is finished. Student example: “I need to be able to solve a new inventory problem without AI, so I will use Tutor and Collaborative Problem-Solver before asking for production.” Professional example: “I own the recommendation, so AI may format the memo but I must define the criteria, compare the options, and defend the conclusion.” If there is no capability to retain, deliberate Production Assistant use may be exactly right.
2.What would have to be true for this to be wrong?
Name the failure mode and the independent evidence that could reveal it before seeing the AI answer. Student example: “The solution fails if the denominator uses monthly rather than annual demand; I will recompute it from the source table.” Professional example: “The recommendation fails if the customer sample excludes cancellations; I will check the extraction rule and base rates.” This turns Verification from an afterthought into a planned control.
3.Whose question is this?
State the decision, stakeholder, constraints, and success criterion in your own words. Student example: “The assignment asks which explanation the evidence supports, not which paragraph sounds strongest.” Professional example: “The decision is whether to pilot the process for one region under a six-week and $20,000 constraint.” If the AI supplied the frame, use Problem Setter to test it before solving the wrong problem efficiently.

Afterwards, ask one more question: could I explain and defend this to someone who will push back without reopening the AI output? If not, decide whether the task only required an artifact or whether you need another Tutor, Challenger, Collaborative, or Verification pass.

A student who can submit a polished answer but cannot solve a parallel problem has evidence of assisted performance, not learning. A manager who cannot name the data, assumptions, and decision rule has a recommendation, not yet a defensible judgment.

For classrooms and teams

The same three questions scale, with one change: the answers have to be written down before the work, because after the work everyone remembers having intended the good version.

A policy that says "use AI responsibly" is unenforceable and slightly insulting. A policy that says which parts of the work must remain the person's, what must be checked and against what, and who owns the final claim is a policy about authorship, friction and calibrated trust, and people can actually follow it.

What the model does not license: judging a person from a mode count. A high Oracle share is not evidence of a lazy student, and a high Challenger share is not evidence of a good one. Modes are situated in tasks.

For high-stakes leadership decisions

Use AI to widen options, expose assumptions, and test a recommendation before the decision meeting. Record the decision owner, the evidence that was checked, the unresolved uncertainty, and which claims still depend on judgment. A polished executive brief is not the control; a traceable decision process is.

Leaders should also decide what their teams must still be able to explain or do without the tool. That protects capability while allowing deliberate delegation of bounded production work.

One honest limit

Learning, judgment, agency, and voice are outcomes the engagement can plausibly strengthen or weaken, with different levels of evidence. Flourishing appears in the model diagram as a theoretical long-run outcome. It has not been demonstrated as an effect of AIEM. Return to the outcomes diagram or compare it with the fitted evidence.

What survived the reading

A short check on the parts of the model people most often invert — the tier grouping, what a mode is, and what a single conversation can and cannot show.

Test what you’ve learned

Twelve short items, about five minutes. Auto-scored where possible.

Restoring your knowledge check…

Not quite the whole model yet

You have landed on the last step, but some of the walkthrough has not been read. The steps below are the ones still outstanding.

Next step

Where to go from here

If the framework helped you locate yourself, the most useful next move depends on what kind of practice you want.

Map your engagement profile (5 min) →

Join the 30–60 minute study Open the companion book