If you are editing the proposal, you briefed it too late.

The proposal comes back clean. You asked for a case to expand the partnership, and there it is: a confident opening over a recommendation already reached. You start editing. You tighten the second paragraph and fix the number in the table. Forty minutes later you send it, and it reads well.

You also, somewhere in those forty minutes, adopted its argument. Not the wording. The position. The draft decided the deal was worth doing on the strength of scale, and by the time you were smoothing its sentences you were defending that case as if it were the one you walked in with. You never actually asked whether scale was the right reason. The draft asked it for you, answered it, and handed you the answer wearing your voice.

This is the loop everyone recommends. Let the machine draft, then edit it into shape. It feels like delegation with a safety net: the associate does the heavy lifting, you keep control at the end. The trouble is that the end is the wrong place to keep control, because the argument was already made three screens up, and editing does not revisit the argument. Editing accepts it and polishes it.

There is now a measurement of how quietly this happens. In two preregistered experiments with 2,582 people, researchers had participants write about a contested social issue alongside an AI assistant that made subtly slanted suggestions as they typed. The attitudes people reported afterward, their own stated views, moved toward the assistant’s position. Most of them did not notice the suggestions were slanted, and did not believe they had been influenced. The pull was stronger than reading the same opinions presented as plain static text, which is the part worth sitting with: it was not the information that moved them. It was writing alongside it (Williams-Ceci et al., 2026).

That study used autocomplete, a nudge of a few words at a time. A finished draft is the same mechanism with the volume turned up. You are not being offered a phrase to accept or wave off. You are being handed a whole structure, already committed, with the reasoning baked into the order of the paragraphs, and your job has been defined for you as improving the thing rather than interrogating it. Nobody sets out to rubber-stamp a case. You set out to edit, and editing is a posture that assumes the case is settled.

Here is the opinion. In a proposal, the value was never the prose. It was the position: what you are recommending, and why this reason and not the other three. That is the part that is yours, the part a client still remembers after the sentences have blurred. Handing off the drafting is fine. Handing off the drafting before you have decided the position is not delegation. It is buying back your own argument at retail during the edit, at whatever price the draft set while you were not looking.

So invert the loop. Brief the argument first, in a sentence or two, in plain language, before a single paragraph exists. We think expanding the partnership is right because of retention, not headcount, and the objection we have to answer is channel conflict. Now the drafting has somewhere to go, and the associate is building your case instead of its own. You brief the take, it builds the prose, and you check the prose still carries the take. That is a director’s sequence.

The tell that you got it backwards is simple. If you are rewriting the draft, not tightening it but arguing with it, moving the claim the whole thing rests on, you briefed too late. The rewrite is you finally doing the work you should have done at the start, except now you are doing it against the current, fighting a fluent draft that has already made itself at home in your head. Fluency is most persuasive exactly when it is unearned. A rough draft you disagree with is easy to overrule. A smooth one you never briefed talks you into agreement before you have noticed you were negotiating.

None of this makes the associate less useful. It sets the usefulness in the right order. Drafting is real work and worth handing off, gladly. The position was the work that had to stay with you, and the reason to state it first is not diligence. It is that a good draft is a very good lawyer, and you want to decide the verdict before you let it open its mouth.