Brief before prompt.

Most AI-assisted work starts with a request for output: write the page, generate the identity, build the component, make it feel premium. The machine responds quickly because the request is shaped like a task.

But a task is not the same thing as a decision. When the decision is missing, the output becomes the centre of gravity. We start editing what the model made before asking whether it was the right thing to make.

A good brief gives the machine a boundary. A good judgment decides whether the boundary was worth drawing.

Before opening a prompt box, I try to write four things down. They do not need to be elegant. They need to make the hidden problem visible.

Four things to decide first

1. 01

   Decision

   What should be easier to decide after this work exists? If the answer is only “have a better interface,” the problem is still too vague.

2. 02

   Context

   Who is looking at this, what do they already know, and what are they trying to protect? Context is where generic output starts to lose its authority.

3. 03

   Constraints

   Name the limits: time, language, platform, budget, content, technical reality. A constraint is not an inconvenience; it is part of the material.

4. 04

   Evidence

   What would count as a useful result? A clearer sentence, a shorter path, a real test, a decision the team can repeat. Without evidence, “good” becomes a mood.

Once those four things are visible, AI becomes more useful. It can widen the options, translate a thought into another language, expose a missing edge case, or give a rough first pass enough shape to react to.

It can also be confidently wrong. The fastest way to lose the thread is to treat fluent output as proof. I look for the middle: the assumptions, the skipped states, the generic choices, and the places where the answer sounds finished before the problem is.

The human part is not adding a personal signature at the end. It is deciding what deserves another pass, what should be removed, and when the result is good enough to put in front of another person.

A good build review ends with fewer possibilities than it started with. That is not a failure of imagination. It is evidence that judgment has done its work.