Three ways to see it

1. 1

   Most professionals in Pakistan have, by now, opened ChatGPT or Claude or Gemini and asked it something. The result was either impressive or disappointing, and there did not seem to be a pattern. There is a pattern. The pattern has a name: prompt engineering. Before we can teach it, we have to be honest about what a prompt actually is, because almost every office in Pakistan, from a Lahore advocate's chamber to a federal secretariat in Islamabad, is using the word loosely. A prompt is not just the question you type. A prompt is the entire input the model sees in one turn: the system instructions you may not have written, the conversation history that piled up before your question, any documents you pasted, any tools the model has been given access to, and finally the words you typed. The model treats all of it as one block of text. If you understand this, you have already moved ahead of ninety percent of the people in your office.

2. 2

   Way one to think about a prompt: it is a set of instructions for a junior office boy who is extremely well-read but has no judgement of his own. Picture a fresh intern who has memorised every law book, every annual report, every newspaper, but who has never once watched what your office actually does. If you say "draft a letter", he will draft something, but the shape of it will be whatever shape he has seen the most. If you say "draft a notice under Section 138 of the Negotiable Instruments Act to a defaulting customer named Asif Khan, signed by me, in a polite but firm tone, on plain letterhead", you will get something close to what you wanted. The intern is not stupid. He is exact. A prompt is an exact instruction to an exact intern.

3. 3

   Way two: a prompt is context. Most failed prompts in Pakistani offices fail not because the instruction was unclear but because the model did not know enough background to produce something useful. "Write a project update" is unclear because the model does not know which project, for whom, in what tone, of what length, against which milestones. The same model that produces nonsense for that prompt will produce a publishable update if you paste in last week's update, the milestones the project is tracking against, and the audience that will read it. Context is not decoration. Context is the largest single lever in prompt quality.

The why-tree