Brief AI Like You Brief Your Team

Think about the best delegator you have ever worked for. When they handed something off, you knew what it was for, who it was for, what good looked like, and when they wanted to see it again. Now think about the worst: "just take a pass at this," followed three days later by disappointment that you read their mind wrong.

Here is the most underrated fact about working with AI, and the reason leaders are better positioned for it than the technical people they assume have the advantage: getting excellent output from AI is a briefing problem, and you have been briefing for years. The people getting poor results from these tools are not under-skilled technically. They are delegating like the second manager: vague in, vague out, then blaming the intern.

The mental model: a capable stranger, not a search engine

Most people's results are capped by one wrong habit: they type into AI the way they type into a search box. Three words, no context, take what comes.

Replace the model. Treat AI as a highly capable new team member: intelligent, fast, tireless, astonishingly widely read, and completely new to your organization. It knows how the world generally does things and nothing about how you do things: your audience, your standards, your situation, your voice. Everything you would tell a talented stranger on their first day is exactly what it needs, and your brief is that onboarding.

Leaders who treat AI like a search engine get search-engine results. Leaders who brief it like a team member get a collaborator. The gap between those two experiences is not the tool. It is the brief. The model works because it matches what the technology genuinely is, which is worth seeing clearly before you direct it.

Where the analogy breaks. One place the team-member model misleads, and it catches leaders out: a new hire accumulates context. Tell them once that morale is fragile after the restructure and they carry it into every conversation for the rest of their tenure. AI largely starts cold. Some tools have added memory features as of mid-2026, but the behavior is uneven across products and not something to build a habit on, so the working assumption should be that yesterday's explanation is gone and the context travels with the brief rather than with the relationship. This is why a leader who had one excellent exchange is baffled when the same request produces mediocre work a week later. Nothing degraded. The brief simply was not there the second time.

The three-part brief: role, context, deliverable

Every strong brief you have ever given carries three things, and they translate directly.

Assign the role. Tell it who to be before what to do. "Act as a senior HR director preparing for a difficult conversation with a disengaged high performer" produces categorically different work than "write some talking points," because the role sets the lens: the assumptions, the tone, the level of expertise to operate from. Without one, AI defaults to a generalist writing for nobody.

Set the stage. Give the background you would give a person: who the audience is, what the purpose is, what has already happened. "I am presenting to a non-technical board next Thursday; we have just completed a restructure and morale is fragile" is one sentence, and it transforms everything produced after it. Context is the difference between output that is generically correct and output that was made for your situation.

Define the deliverable. Be precise about the shape of done: format, length, tone, what it must include, what it must avoid. You would never tell a designer "make something nice." The same discipline applies, and it costs one more sentence.

Three sentences of brief, sometimes four, is the entire foundation. Most people supply zero of them.

What a real brief looks like

Descriptions of the structure are easy to nod at and hard to act on, so here is one written out in full. A division head has a monthly update to send upward, the same one they send every month, and the month went badly.

Act as a chief of staff who has written monthly updates for a decade for executives who read fast and dislike preamble. I run operations for a 40-person division and report to a COO who has been supportive but is under pressure too. We missed our delivery target by 9% this month, mostly because two people left in April and replacement hiring has been slow. The fix is already underway and I want the memo to read as neither panic nor excuse-making. Draft a one-page update in three short sections, what happened, why, and what I am doing about it, under 400 words, plain declarative sentences, leading with the miss rather than burying it. No corporate optimism language and no apologizing.

Read it again and notice what is in there: the role in the first sentence, the situation and the audience in the middle three, the shape of done at the end. Notice also what is not in there. Nothing technical. No special syntax. No terminology a leader would have to look up. It is what you would say out loud to a competent assistant standing in your doorway, and it takes about ninety seconds to type, or thirty to dictate.

The honest objection is that ninety seconds is longer than the eight seconds most people spend typing "write me a monthly update." It is. It is also shorter than the three rounds of disappointing drafts that the eight-second version reliably produces, and it is the version that ends with something you would actually send.

The reason this brief works is not its length. It is that every sentence removes a decision the tool would otherwise have made badly on your behalf: who is speaking, who is listening, what happened, what tone the situation calls for, and when to stop. Vagueness does not spare AI those decisions. It just hands them over.

When the first output misses: direct, don't restart

The second skill is what you do after the first draft comes back, because a first draft is what it is. Amateurs start over or give up. Leaders redirect, exactly as they would with a person.

• Redirect what is close. "Keep the structure and tone. The opening is too formal; make it direct. Sharpen the recommendation in the second paragraph." Progress preserved, correction targeted.
• Narrow the frame when output is generic. "Do not write this for a general audience. Write it for a CFO who has already approved the budget and needs to communicate the decision, not justify it."
• Set the contrast when it keeps defaulting. Say what you do not want with the same precision as what you do: "No corporate announcement voice. No 'we are excited to share.' One leader, writing honestly to their team."
• Ask for the thinking first on complex work. Before the final product: "list the key tensions and trade-offs in this decision first." Review the reasoning, then build. You would run a big decision with a strategist the same way.

Notice what these four moves have in common: they are all management. Nothing technical has appeared anywhere in this guide, because there is nothing technical to appear.

The quality of what AI gives you is set almost entirely by the quality of what you give it. That has always been true of delegation. This is delegation.

The two disciplines that keep the skill safe

Briefing gets you quality. Two habits keep it trustworthy.

Verify before it travels. AI produces errors with the same confidence as accuracies, so any fact, figure, or claim gets checked before the work leaves your hands, every time, and anything AI-assisted that reaches colleagues or clients gets your eyes first. The brief was yours; the output is yours too. The same discipline decides what gets briefed at all: delegation to AI follows the line between pattern work and judgment work, and a team briefing AI without shared standards is running ungoverned whether anyone meant it or not.

Save what works. When a brief produces excellent output, keep it. You are not just finishing a task; you are building a personal library of delegation instructions, reusable every time the situation recurs. Six months of this and you have something no tool subscription provides: your own operating manual for directing AI in your world.

One task this week

Do not transform your workflow. Pick one recurring piece of writing or analysis, ideally something from the attention work that quietly eats a manager's week, brief it properly, role, stage, deliverable, redirect twice, verify, and save the brief. One task, done well, teaches more than a month of scattered experimenting, and the confidence it builds is the real product.

This skill is the heart of what the AI Stakeholder Challenge calls fluency: directing AI effectively, efficiently, ethically, and safely, built through your own hands on your own real work across the seven days, not watched in a demo. But the first brief needs nothing except the next task on your list and three honest sentences. You have been training for this skill your whole leadership life. Use it.