Briefing an AI Agent: How to Write a Good Brief

The result you get from an AI agent (Claude, ChatGPT, or a custom agent) is only as clear as the instruction you gave it. A vague instruction like “write me a report” produces a vague — often wrong — result; the agent fills the gaps with its own assumptions. A good brief prevents this: it defines context (for whom, why), constraints (format, length, tone, phrases to avoid), and expected format (what the output should look like) upfront. Without all three, you’re not assigning a task, you’re rolling dice.

The 4 Parts of a Good Brief

Four elements make a brief a brief. Miss one, and the agent guesses the rest — and its guesses usually don’t serve you well.

1. Context: for whom, why

The agent can’t make good decisions without knowing the background of the task. The same “write an email” instruction requires a completely different result for an email to a late supplier versus a thank-you email to a customer.

• Bad: Write an email.
• Good: Write a firm but polite email to a supplier who’s running late, asking when the order will arrive. This is our third reminder.

2. Clear task definition: what to do, what not to do

The “what not to do” part is missing from most briefs — and that’s exactly where the agent tends to expand scope on its own and do too much.

• Bad: Review my site and improve it.
• Good: Check the form fields on the contact page, list any missing fields. Don’t touch the design or colors, just the form logic.

3. Constraints: format, length, tone, phrases to avoid

If you don’t set a constraint, the agent uses its own default style — usually too formal, too long, and full of clichés.

• Bad: Write a promotional text.
• Good: Keep it under 150 words, use first person, avoid unsubstantiated superlatives like “leader,” “pioneer,” “the best.”

4. Success criteria: what does “done right” mean

If you don’t tell the agent what counts as acceptable, you end up manually approving every result yourself — which defeats the purpose of writing a brief in the first place.

• Bad: Prepare a nice presentation.
• Good: Prepare a presentation no longer than 10 slides, one message per slide, with numbers shown in large font. Success criterion: an investor should be able to understand it in 3 minutes.

Bad Brief vs Good Brief: A Real Comparison

Give an agent the sentence “write me a report” and you’ve essentially said: “do whatever comes to mind.” Compare the two briefs below.

Bad brief: “Write me a report about last month’s ad performance.”

Result: the agent doesn’t know which account, which metrics, or who it’s writing for. It’ll likely produce something generic, long, overly detailed, and that doesn’t answer the question you actually wanted answered. You’ll need a second round to fix it — that’s exactly where the time gets lost.

Good brief: “Looking at July’s Google Ads data (this is going to the board), find which campaign is burning budget without generating conversions. Keep the report under half a page, no technical jargon, format it as three bullet points plus one recommendation sentence. Success criterion: a board member should be able to read it and decide in 60 seconds.”

The difference isn’t in line count, it’s in the information inside the brief. The second version gets the same agent to produce the right result, in the same time, in one pass.

Does the Brief Change by Agent Type?

No, the four parts of a brief stay the same — what changes is how much detail is needed. For a one-off Q&A task (e.g., “translate this email to English, use a formal tone”), the four parts fit into two or three sentences.

For a multi-step, autonomous agent task that calls tools on its own and touches files or accounts, the brief needs two more things: boundaries (which files, accounts, or systems the agent can and can’t touch) and approval points (after which step it needs to stop and get your sign-off). I’ve covered the difference between an AI agent and a simple prompt, and how that maps onto a brief, in Prompt vs Skill vs Agent.

What to Watch Out For

To be honest, writing a good brief has its own cost — and overdoing it has its own risk.

• An overly detailed brief wastes time. If you try to pre-write every possible situation, the brief takes longer to write than the task itself. Three or four sentences are usually enough for a simple task.
• The first attempt is rarely perfect. Even with the best brief, the agent will sometimes run with an interpretation you didn’t intend. That’s not a failure, it’s a normal step.
• A few iterations are normal. The method that works in practice: start with a short brief, look at the output, clarify the missing piece in one sentence. Most tasks settle within two or three rounds.

Trying to write the perfect brief up front takes more time than “start good enough, fix fast.”

An Observation From My Own Operation

One day I told an agent “find the campaigns with underperforming budgets,” and the agent went through all 43 accounts I manage one by one, burning hours — and eventually brought back a list sorted by its own interpretation of “underperforming,” ranked by click-through rate instead of revenue. That wasn’t what I was after; I lost time, and so did the agent.

On the second attempt, I changed the brief to: “Only Google Ads account X, last 30 days, list campaigns where cost-per-conversion is 50% above target, sort by revenue, max 10 rows.” The same agent, with the same capability, brought back exactly the table I wanted within minutes. The difference wasn’t in the agent — it was in the brief.

You Can Ask Me to Do This

Setting up a briefing system for AI agents in your company — which template for which task, which approval point, which constraint set — is work that typically takes 2 hours to 2 weeks, remote, billed hourly. You can ask me to do this as part of my Company Prompt Library service.

Frequently Asked Questions

Does writing a brief for an AI agent actually make a difference?

Yes. The same capable agent produces a random result with a vague instruction, and a targeted result with a clear brief. The difference isn’t in the agent’s intelligence, it’s in the clarity of the information you gave it.

Do I need to write all four parts for a simple task?

No. For short, one-off tasks (a translation, a summary), context and constraints usually fold into a single sentence. Writing the four parts out separately matters mainly for multi-step, autonomous tasks.

Can I reuse the same brief over and over?

Yes, turning a brief into a template saves time for recurring tasks. If your weekly report request always wants the same format, clarify the brief once and reuse it every week.

Can a brief be too long?

Yes, it can. Unnecessary detail can cause the agent to miss the constraint that actually matters. The goal is sufficient, not exhaustive — the shortest brief that covers the four parts is usually the best one.