Briefs, prompts, and the iteration loop
> Great outputs come from clear jobs, constraints, and ruthless iteration — not magic words.
Canonical: https://www.brainyxai.co.za/education/claude/briefs-and-prompts
Markdown: https://www.brainyxai.co.za/md/education/claude/briefs-and-prompts.md
Course: Claude Operator Track
Lesson: 2 of 9
Minutes: 16
Author: Brainyx AI
What you will be able to do
- Write a brief with goal, audience, inputs, constraints, and output shape
- Iterate with specific critiques instead of “try again”
- Separate drafting from reviewing (two different prompts)
The brief beats the “prompt hack”
Stop collecting viral prompt templates. Operators win with a brief:
1. Goal — what decision or artifact this unlocks
2. Audience — who reads it and what they already know
3. Inputs — paste or attach the raw material (emails, specs, data sample)
4. Constraints — length, tone, must-include, must-avoid, tools allowed
5. Output shape — bullets, table, JSON, PR description, email draft
Example skeleton you can reuse:
```text
Goal: …
Audience: …
Inputs: [paste]
Constraints: …
Output: …
Done means: …
```
Iteration is the skill
First drafts are reconnaissance. Second prompts should be surgical:
- “Cut section 2; expand risks with concrete mitigations.”
- “Rewrite for a CFO; no jargon; add a 3-line ROI paragraph.”
- “Mark every claim that needs a source with [CITE].”
Avoid: “Make it better.” That trains you to accept vague work.
Two roles, two chats (or two turns)
1. Drafter — generate options
2. Reviewer — attack gaps, invent nothing, list what’s unverified
Many teams keep these as separate Project instructions or saved prompts.
Context packing
- Put the source of truth above the ask when it matters
- Prefer attachments / Projects over re-pasting the same 20 pages every time
- Say what to ignore (“Ignore marketing fluff; use only the API section”)
Official prompt engineering overview: [Anthropic docs](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview).
Brief lab
1. Pick a real work artifact (proposal outline, support macro, weekly update).
2. Write a full five-part brief and run it once — save the output.
3. Send three surgical iteration prompts; keep only the final version you’d actually send.
Checkpoints
- My prompts state done-means criteria
- I iterate with specifics, not vibes
A worked brief, end to end
Abstract advice about briefs rarely changes behaviour. Here is the same request written twice.
Weak version: "Write a proposal for a client who wants automation."
Operator version:
```text
Goal: A two-page proposal that gets a scoping call booked.
Audience: Operations director, non-technical, sceptical of AI hype.
Inputs: [paste discovery call notes + current process description]
Constraints: No pricing. No jargon. Name the risk they raised about staff resistance.
Output: Problem, proposed approach, what we need from them, next step.
Done means: They could forward it to a CFO without editing.
```
The second version is not longer because longer is better. It is longer because each line removes a decision the model would otherwise guess.
Iteration that compounds
Treat the first output as reconnaissance. Your second message should name the defect and the fix:
- "Section two repeats section one. Cut it and expand the risk paragraph with two concrete mitigations."
- "Rewrite for a CFO. Remove every adjective that is not load-bearing."
- "Mark every factual claim that needs a source with [CITE] so I can verify before sending."
"Make it better" is the tell of an operator who has not decided what better means. If you cannot name the defect, you are not ready to iterate — reread the output and find it.
Splitting drafter from reviewer
Asking one prompt to both generate and critique produces polite self-assessment. Split the roles:
The drafter gets your brief. The reviewer gets a separate instruction: "You did not write this. List every unsupported claim, every place the audience would disagree, and everything missing that a sceptical reader would ask for. Invent nothing."
Teams that formalise this pair — as two saved prompts, or two Project instruction blocks — cut revision cycles noticeably.
Where this pays off in SA operations
Proposal and tender drafting is the obvious win: structured documents, repeated shape, high time cost. Support macros are second. Both benefit from briefs because the audience and constraints are stable, so you write the brief once and reuse it for months.
If you find yourself pasting the same background into every brief, that is the signal to move to Projects — the next lesson.
Mini-FAQ
A: Build a brief library — reusable goal, audience, and constraint blocks for recurring jobs. Collections of clever phrasings age badly; job definitions do not.
A: Two to four surgical rounds for most business artefacts. If you are past six, the brief was wrong, not the model.
A: Clarity improves output. Politeness is free but not the mechanism.
Next lesson
Continue to [Projects and durable context](https://www.brainyxai.co.za/education/claude/projects-and-context).
Official reference: https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview
Course hub: https://www.brainyxai.co.za/education/claude · Previous: https://www.brainyxai.co.za/education/claude/meet-claude · Next: https://www.brainyxai.co.za/education/claude/projects-and-context
Return to the course hub, or explore Brainyx AI services to put the skill into a live system. Markdown: /md/education/claude/briefs-and-prompts.md
Book a consultation · joshua@brainyxai.co.za · Markdown mirrors