Shape the work before you stamp it

A hypothetical chief executive asks an AI assistant for a board memo explaining slower growth. The draft is orderly and plausible. It presents management’s preferred explanation, supplies transitions, and ends with a recommendation. The temptation is to start improving the prose.

A more useful next step may be to take the memo apart.

Put each important claim into a table beside its supporting evidence, the strongest contrary evidence, and the decision it affects. The polished narrative may turn out to depend on a customer trend nobody has checked or a hiring assumption the operating team disputes. Those are problems that sentence editing would have left intact.

This is the distinction between finding the useful form of the work and making that form repeatable. Early on, the central task may still be discovery. A finished-looking output should not settle it prematurely.

Change the representation to change what you can see

The form of an artifact directs attention. Prose can make a story coherent while leaving relationships implicit. A table can expose a missing source. A decision tree can show an unresolved exception. A crude screen can reveal that a proposed product asks its user to do too much.

Each transformation should answer a question about the work. In the hypothetical memo, a claim table tests whether the argument is supported. Two rival explanations test whether the same facts could lead management somewhere else. An outline organized around the board’s actual choice tests whether the draft is answering the right question.

The temporary forms need not appear in the final memo. Their value lies in what they allow the reviewer to discover before a recommendation becomes difficult to change.

A different hypothetical example makes the point more concrete. A team is revising a customer-escalation policy. Instead of perfecting the policy’s language, walk several cases through it: a contained problem for a large account, an unresolved issue for a small account, and a request with an unclear owner. Identify who acts, what information they need, and when someone else must become involved.

If the cases expose contradictory instructions, return to the policy and change the rule. If they produce sensible answers, the language can become shorter because the underlying distinctions are clearer. AI can help draft cases or compare alternatives; the team must judge them against the behavior it intends.

Know when another transformation will help

Exploration can also become avoidance. Five alternative outlines are not useful merely because they are different. Another representation earns its place when it could expose a consequential uncertainty or change the choice.

Before generating it, say what you expect to learn. After reviewing it, record what changed in your understanding. If nothing changes across repeated attempts, the missing ingredient may be evidence or a decision from someone responsible, rather than another prompt.

The person commissioning the work must define the intended effect. What should the board decide? Which exception should the policy handle? What should a user be able to do? Without that criterion, the team can keep improving the artifact’s appearance while remaining uncertain about its purpose.

Make the settled parts repeatable

Once a form works across relevant cases, preserve what has proved useful. A recurring memo may need a stable source requirement and an explicit section for contrary evidence. A dependable policy may need consistent intake fields and a clear escalation path. Software behavior that must persist can be protected by an appropriate test.

Stability need not mean freezing the entire artifact. The evidence table may remain consistent while the argument changes each quarter. An escalation process may keep its ownership rules while new cases reveal a threshold worth revisiting.

In the growth memo, the final version might be much shorter than the first draft. It can state the uncertainty directly, explain which evidence favors each interpretation, and ask the board for a specific decision. The intermediate work has done its job when it helps management produce that clarity and can then be put aside.