How I brief an agent
What if your chat went silent and you could actually work in parallel?
- The goal
- The rules
- The proof of delivery
- The expectation
- The assets
Most people talk to an agent the way they talk to a colleague at the coffee machine: a stream of context, a hope, a "you know what I mean". A colleague fills the gaps with three years of shared history. The agent fills them with the average of the internet and its recent context.
A brief seemed like overkill, and it turned out to be the thing the agents treat as their golden path. It gave me better continuity between agents than any prompt I ever wrote. And while the brief gets mapped and sketched, I can work on a detailed feature scope, or review other work and enter new issues. Never interrupting its flow, only improving its context.
A brief gets you the thing. The difference is in five truths, and they are the same five for a person and for an agent.
The brief is an open door to iteration. It does not define what is behind it.
Why it is hard
Prompts make it look fast, but the waste of context and long-term memory makes them slower.
What changed
Two rounds instead of six+ on a typical build or creative project.
The five truths
- The goal. What finished looks like, the high-level sentence a stranger could check.
- The rules. What must not change, what must not be touched, who must not be named.
- The proof of delivery. How you will confirm: a sketch, an artifact, a test, a screenshot, a number.
- The expectation. What the minimum is, and what the stretch goal is.
- The assets. What is provided and what is needed: data, assets, decisions.
The first week feels slow. You are writing down what you used to keep in your head. By the second week, briefs are what let you keep up across several projects and several agents. It is not only their memory. It is ours too.