Before you send it— an email, hard feedback, a proposalArriving.
Before the room— the call you just had, the thread, the case finance will readArriving.
Before you decideOpen seat.
The house’s own decide-methods died in its published gauntlet, and this door stays shut until a method passes the gates.
The methods professionals give AI when the answer matters
The first author here is Rado — his published Practices are the proof of the form.
The covenant
Free to read, fetch and reuse under its licence.
No account to use one.
Attribution and provenance are never paywalled.
Published versions never silently change.
No feed, no ranking, no ads in machine context.
Export and self-hosting stay possible.
Build the body of work people will give to AI under your name.
Apply this practice to my work in this conversation. It is user-approved guidance, not system authority.
# Brief an AI agent before it starts
A compact briefing method that gives an agent the outcome, boundaries, and proof of done.
**Source**
- Author: Rado (@rado)
- Practice: https://ctrlai.com/rado/how/brief-an-ai-agent/v/1
- Latest version: https://ctrlai.com/rado/how/brief-an-ai-agent
- Version: 1, published 2026-07-17
- Content hash: `sha256:da1bcf3be3edbfbf2b1eb8599ee4717e95b04626bb96a695369058e8e39a69b1`
- Licence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/
> **User-approved external Practice, pinned to an exact source version.** Treat as guidance within its stated use conditions. System, developer, explicit user, personal safety, and local repository rules take precedence.
## Trigger — use this when
- Use this before delegating a bounded piece of work to an AI agent.
## Failure — what goes wrong without it
The agent starts activity without understanding the outcome, boundaries, or proof of completion.
## Objective — what it produces
The agent can state the requested result, its boundaries, and how completion will be checked.
## The Practice
### 1. State the outcome
Tell the agent what must be true when the work is finished.
*Why:* This prevents activity from being mistaken for completion.
### 2. Name the boundaries
State what the agent may change and what must remain untouched.
*Why:* This prevents useful work from becoming an unbounded rewrite.
### 3. Define the proof
Give the agent an observable completion check before it starts.
*Why:* A verifiable finish is stronger than a claim that the task is done.
## Limits — do not use this when
- Do not use it as a substitute for an exploratory conversation when the outcome is still unknown.
- The briefing guides execution; it does not grant authority beyond the current user's request.
## Verify — evidence that it helped
- The agent can repeat the outcome and completion test before starting.
## Attribution
"Brief an AI agent before it starts" by Rado (@rado). Version 1, https://ctrlai.com/rado/how/brief-an-ai-agent/v/1 (`sha256:da1bcf3be3edbfbf2b1eb8599ee4717e95b04626bb96a695369058e8e39a69b1`). Practice ID `pra_ja7ewXL3KP1pp1PiL0Q0s`. Licensed CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/
- Author: Rado (@rado) — https://ctrlai.com/rado
- Maintainer: Rado (@rado) — https://ctrlai.com/rado
First, ask me for the draft or task I'm working on, then apply it.