Fawzooz
عربيCollaborate
Your journey
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
0 of 34 units

You are in Degree 3 · Practical masteryunit 2 of 6Ahead of you: A real piece of professional work, and the page that shows you checked it.

Degree 3 · Unit 3.2

Context engineering

For a few years the term "prompt engineering" was everywhere, as though the secret lay in some magic phrase that had to be uttered correctly. Then the windows widened and the models improved, and it became clear that the phrase was never the secret. The secret is what you put in front of the system: which information, in what order, and what you deliberately hold back from it.

The irony is that the commonest mistake today is excess — throwing everything you have at the system in the belief that more of it must be better. A window full of padding weakens the attention the system can give to the things that matter inside it, and context that contradicts itself produces a hesitant answer. So the rule is this: supply what is needed, not everything you happen to have.

FIG. 13 — Ordering the window
1
First
The role and the governing rules — what must never be forgotten
2
Middle
The long reference material — titled and numbered so it can be pointed to
3
Last
The specific task, the constraints and the standard of success — the last thing read carries the most weight
fawzooz.ai

Four techniques worth memorising

1
Attached examples
Attach two or three earlier outputs you consider good. The system picks up from an example what your description cannot convey: the tone, the level of detail, the way things are ordered. One good example spares you a paragraph of instructions.
2
Thinking step by step
On analytical tasks, ask it to lay out its reasoning before the conclusion. The benefit is double: the answer improves, and you can review its logic rather than only its result.
3
Splitting into stages
Do not ask for a whole report in one commission. Ask for the outline, review it, then ask for it section by section. On complex tasks, quality comes from splitting, not from longer instructions.
4
Deliberate withholding
Strip out sensitive names and figures before you put anything in and replace them with symbols ("Client A", "Amount X"). You get the same analysis with no leak. This is a professional habit, not excessive caution.
fawzooz.ai

Do this

  1. 1 — On the tool. Run the same task twice: once with a written description of the output you want, and once with a real example attached. The difference between the results is the value of the example.

  2. 2 — On the tool. Take a complex task and run it once in a single pass and once in four stages. Which produced work you would put your name to?

  3. 3 — In your field. Add three templates to "my templates": analysis, drafting, and review — each with the five layers and an attached example.

The template library: the most valuable thing you build this year

The people who gain most from this technique are not the ones who memorise it, but the ones who turn it into a file they keep coming back to. After three months you will find that half your work recurs in similar forms, and that one improved template saves you an hour every week.

And order that library by task rather than by tool: "risk analysis", "contract drafting", "report review", "explaining something to a non-specialist". The tools change every year. The tasks do not.

Where to after this unit? You have mastered the conversation. The next unit moves to what does not converse: a system that works on your behalf with no moment-to-moment supervision.