Guide

How to Brief an AI Team: Templates and a Worked Example

By the Space Office team · July 6, 2026 · 7 min read

To brief an AI team, state five things in about 100–150 words: the deliverable, who it's for, the voice with one example, the hard constraints, and the deadline. Then send it to the manager — not the individual specialists — and treat every revision note as an edit to your next brief. That's the whole method. The rest of this guide shows it working on a real request, line by line.

The brief is the interface

When you hire people, your management happens in meetings, Slack threads, and hallway corrections. With an AI team there is no hallway. Everything you would have managed in person has to live in the brief — which sounds like a burden until you notice it's actually the deal of the decade: write 150 good words once, and a whole team executes against them without a single status meeting.

That makes brief-writing the highest-leverage fifteen minutes in your week. It's also a skill almost nobody has practiced, because delegating to humans let us get away with vagueness — a person fills the gaps by asking questions at the coffee machine. An AI team fills them by guessing.

Autopsy of a vague brief

Here's a brief we've all written: "Write some posts about our launch." Ten words, feels efficient, and it forces the team to guess four times. Each guess it gets wrong becomes a revision round you pay for with your own review time.

1. Who is this for? (the team guesses)

Posts for developers read nothing like posts for buyers. Left unstated, the team picks a generic middle — and generic is the one audience nobody belongs to.

2. What does done look like? (the team guesses)

"Some posts" could be three tweets or a 2,000-word announcement. A brief that names the deliverable — format, count, length — removes the most expensive guess first.

3. What voice — and says who? (the team guesses)

"On-brand" means nothing without an exhibit. One link to a piece you love and one line about why does more than three paragraphs of adjectives.

4. By when, and what's fixed? (the team guesses)

Deadlines, banned claims, required links, the price you do or don't want mentioned. Constraints aren't bureaucracy; they're the walls that stop good work from wandering.

The same request, rewritten

A brief a team can actually run

"We launch scheduled tasks on July 20. Deliverable: one 900-word announcement post and six social captions (4 X, 2 LinkedIn). Audience: solo founders who already use us — assume they know the product. Voice: plain and confident, like our 'delegate work' post; no hype words. Must include: the /how-it-works link and the flat-$100 pricing. Must not: promise integrations we haven't shipped. Draft by July 14."

Sixty-nine words. Every sentence closes a guess: the event, the exact deliverables, the reader, the voice with an exhibit, two hard constraints, a date. Nothing in it required marketing skill — only decisions you'd already made and hadn't written down.

What a brief must nail, by job type

Different work fails on different missing details. This table is the pre-flight check: for each job, the two or three details that most often force a revision when omitted — and who on the Space Office roster picks the work up.

Brief the manager, not the specialists

You don't send five briefs to five agents. You send one brief to Hydrogen, the AI project manager, and it does what a good manager does: splits the work into sub-tasks, assigns the right specialists, and reviews everything before it reaches you. Give it the launch brief above and it hands Lithium the announcement, Beryllium the header graphic, Boron the SEO pass, and Neon the captions — coordinated against the same deadline. One brief in, one finished package out is the difference between owning a team and operating a toolbox.

Include what the model can't know; skip what it can

The most common briefing mistake is spending words on things the team already does — "write clearly," "use correct grammar" — and none on things only you know. Google engineering leader Addy Osmani makes the same point about writing specs for AI agents: include only the knowledge the model doesn't have. Your customer, your voice, your constraints, and your definition of done are the entire brief. Style rules the model already follows are padding.

A brief isn't instructions for how to work. It's a record of the decisions only you could make.

The revision-round math

At a flat $100/month, an extra revision round costs Space Office customers no extra money — what it costs is your attention. Say each round of review takes about 25 minutes (illustrative). A vague brief that averages 3 rounds per task, across 10 tasks a month, is 10 × 3 × 25 = 750 minutes — about 12.5 hours of your month spent re-explaining yourself. A tight brief that averages 1 round is 10 × 25 = 250 minutes, or roughly 4 hours. That's over 8 hours a month returned by the fifteen minutes you spend writing briefs properly. The subscription was never the scarce resource; your attention is.

Feedback is the second half of the brief

No brief survives first contact perfectly, and it doesn't need to — Hydrogen QAs every output before delivery, and in internal testing it caught roughly 4 of 5 quality issues across 240 sample tasks. Your job is the remaining fifth: when a draft misses, say what was wrong in one sentence and move that sentence into your next brief. "Too formal — we say 'you,' not 'users'" is feedback today and a brief line forever. Founders who do this watch their revision average fall month over month.

A skeleton you can reuse

You don't need a new structure per task — you need the same five lines answered freshly each time. Keep this skeleton wherever you write briefs, and fill it in order:

1. 1Deliverable: exactly what comes back — format, count, length.
2. 2Audience: who reads it, and what they already know.
3. 3Voice: one example you admire, and one line on why.
4. 4Constraints: deadlines, required links, banned claims, fixed facts like pricing.
5. 5Done means: the one-sentence test you'll apply when you review it.

Five lines takes most founders under fifteen minutes once the first one exists — the voice example and standing constraints barely change between briefs, so each new brief is mostly line one and line five. The first brief is the slowest one you'll ever write.

When to skip the brief and do it yourself

If you can't state the outcome in a sentence, the task isn't ready to delegate — to an AI team or to anyone. Strategy pivots, pricing calls, taste decisions on brand, anything where you'd recognize the answer only when you see it: do those yourself, once, and delegate the repeatable version that emerges. We go deeper on choosing what to hand off in our guide to delegating work to AI agents.

Write the sentence, get the team

Every founder wants leverage, and most look for it in tools. It's closer than that: a team of 30 specialists is sitting behind a text box, and the quality of what comes back is set by the quality of the 150 words you put in. A team is only ever as good as the sentence you hand it — so hand it a good one.