Builder briefs before prompts
Prompting has a reputation problem: half the internet treats it like incantation — magic words that unlock the machine. It isn't. A prompt is a brief for a tool, the same builder brief you met in chapter zero, sharpened here for serious work.
The five knobs are still the foundation:
- Role — who should the model act as for this task?
- Task — what exact job should it do?
- Context — what does it need to know before answering?
- Format — what shape should the output take?
- Examples — what does good look like?
For agentic coding tools and AI project workspaces, add a sixth:
- Constraints — what must it use, avoid, preserve, test, or not touch?
Bad prompt:
make this better
Builder brief:
Role: product editor for beginner AI education
Task: rewrite this lesson intro so it is less scary for people new to AI
Context: the learner has never used an agentic coding tool seriously
Format: 3 short paragraphs and 4 bullet points
Examples: use "tool" not "helper"; code is "moving parts"
Constraints: do not mention Python in the first paragraph; do not change file paths
The brief isn't better because it's longer — length is not a virtue. It's better because it removes the guessing. Most failed AI sessions are missing one of the knobs: the model guesses, you dislike the guess, and suddenly the chat is a negotiation with a mistake you created upstream. (You'll lose that negotiation politely, one turn at a time.)
Builders don't argue with vague outputs. They improve the brief.