Key Concepts
Context: Everything the agent needs to understand the world it works in: who the audience is, what the brand sounds like, why the task matters, and where examples of good work live. It is the background you would give a new teammate on day one.
Job: The actual task, stated as one pattern with one outcome. It only works if the context is in place.
Output: What the result looks like: the format, length, sections, and where it goes. Decide delivery first, since people tend to describe it last.
Quick Reference
The three-part brief
- Context: audience, brand voice, why the task matters, where good examples live
- Job: one pattern, one outcome. Example: take a topic and produce a brief using what answer engines already cite.
- Output: format, length, sections, and destination
Two additions
- When there is nothing to report: say what to do. Without this, the agent will find something small and report it, because silence feels like failure to a system built to be helpful.
- Constraints: say what the agent should not do. Examples: don’t recommend, don’t rewrite, don’t summarize what stayed the same. What you rule out keeps the result sharp.
The 3 AM test
- Imagine the agent runs at 3 AM and nobody is awake. Is the brief still clear enough?
- Any decision you leave open, the agent will make for you.
Why this matters more than in chat
- In chat you are present: you read, correct, and retry.
- An agent runs without you, so errors are harder to catch and the next step builds on the mistake.
- Inputs can change every run, including ones you have never tried.
Reference build: AEO Content Brief Generator
- Takes a topic and turns it into a brief grounded in what answer engines already cite
Notes
- None of this is code or clever phrasing. It is describing the work clearly.