DRAG: Drafting, Research, Analysis, Grunt work

Think of a kenduri kahwin in the kampung. The night before, the helpers are cutting onions, stirring the gulai and washing the big pots. The tuan rumah does none of that. They decide the menu, check the dishes are right, and greet every guest at the door.

Nobody thinks the tuan rumah is lazy. The work is shared out sensibly: helpers do the heavy, repetitive jobs, and the host keeps the jobs that need their name and their judgement.

DRAG shares out your office work the same way. AI is your rewang for four kinds of work: Drafting, Research, Analysis and Grunt work. You stay the tuan rumah: you decide, approve, and talk to people when it matters.

The idea in 30 seconds

Drafting

Minutes, surat rasmi, press statements, brief outlines. AI writes the first version.

The idea in 30 seconds0:00 / 0:31

Key takeaways

If you remember nothing else, remember these.

Drafting: AI writes first versions; you edit and sign.
Research: AI reads and summarises the sources you supply; you check them.
Analysis: AI sorts and compares; you judge what the pattern means.
Grunt work: AI does the tedious formatting; you spot-check it.
Decisions, approvals, sensitive mediation calls and personal data never go into the DRAG bins.

The rewang does the work, the tuan rumah decides

In OSEMN you learned to get data you can trust. DRAG is the next question: now that your facts are in order, which parts of the job do you hand to AI? On the course map, DRAG sits in the “Analyze & Design” phase, right after OSEMN, and it closes the path.

Here is the difference in practice. Tap between the two requests.

Handle the monthly KRT report for my district.

What you typically get back

The AI guesses the format, invents numbers it does not have, and writes conclusions about KRTs it knows nothing about. You cannot tell what is real, so you redo it.

You are an assistant to a district unity officer.
Below are 12 anonymised KRT activity summaries for September, and last month's report as a template.
Grunt work: put the figures into the template table. Analysis: flag any KRT with fewer activities than in August. Drafting: write a three-sentence summary paragraph. Leave the "Officer's recommendation" section empty for me.

What you typically get back

Each part of the job is named and bounded. You get a filled table, a short list of KRTs to look at, and a draft paragraph. The recommendation, the part with your name on it, is left for you.

The good request does not ask the AI to “do the report”. It splits the report into DRAG jobs and keeps the judgement for you.

D for Drafting: the first version, not the final one

Drafting is writing the first version of something you will edit. In KPN work that means minutes of meetings, surat rasmi, press statements, social media posts, and the outline of a ministerial brief. AI is fast at a first draft. It is not responsible for what the final version says. You are.

Drafting works best when you give the AI three things: your rough material (notes, points, facts), a sample of the format you need, and the language. For a press statement in BM, English, 中文 and தமிழ், ask for all four, then have a fluent colleague check every language before anything is published.

Drafting: minutes from rough notes

You are a secretariat officer who writes minutes for KRT committee meetings.

Below are my rough notes from the KRT Taman Seri Contoh committee meeting held last Thursday. Names are replaced with roles (Pengerusi, Setiausaha, AJK 1, AJK 2). I also paste the minutes from the previous meeting as the format to follow.

Write the minutes in the same format: attendance, matters arising, decisions, and an action table with columns Action, Owner and Due date. Do not add any decision that is not in my notes. Mark anything unclear as [CHECK].

Before I send this, list every decision and every action owner you included, each with the line of my notes it came from.

The last line is the check step. It makes the AI show where each decision came from, so you can see at once if it invented one.

R for Research: only from sources you give it

Research here means reading and summarising documents for you: a long circular, two programme concept papers to compare, background notes for a briefing. The rule is simple. Use only the sources you give the AI, and check them.

Why so strict? An AI tool can produce a confident sentence that has no source at all. This is called a . If you ask it “what does the latest circular say?” without giving it the circular, it may make one up. If you paste the circular and say “use only this”, you can check every point against the page.

Research: summarise a circular you supply

You are a policy officer who writes short briefing notes for a director.

I have pasted a 30-page circular below. Use only this document. Do not use outside knowledge.

Write a one-page background note with three headings: What changes, Who is affected, and Dates to remember. Under each point, give the section number where you found it.

After the note, list any point you could not find a section number for, and mark it UNSURE.

A for Analysis: patterns, not verdicts

Analysis is sorting, counting and comparing to find patterns. Typical KPN examples: grouping public complaints by theme, comparing monthly activity data from KRTs, or spotting common answers in a community survey. AI is quick at this, and it does not get tired at complaint number 180.

But a pattern is not a conclusion. If complaints about noise rose in one area, the AI can show you that. Whether it signals tension, a festival season or one busy construction site is your call, because you know the ground. Tap each part of this analysis request to see what it does.

Dissect a prompt

Analysis: complaints by theme

Tap a coloured part of the prompt to see what it does.

You are a data assistant in a district unity office. Below are 200 public complaints from the last quarter. Names, MyKad numbers, addresses and phone numbers have been removed. Group them into no more than 6 themes. Give a table with Theme, Count and two example complaint numbers per theme. Then list any complaint you could not place in a theme.

Analysis: compare monthly KRT data

You are a data assistant for a district unity officer.

The table below shows the number of activities for 12 fictional KRTs (KRT A to KRT L), January to June.

Compare the months for each KRT. List any KRT whose activities dropped by half or more from one month to the next, with the two months and the two numbers. Do not explain why; just show the change.

Finish by adding up each KRT's six-month total, so I can check your reading of the table against my own totals.

G for Grunt work: the tedious jobs

Grunt work is the boring, rule-based work that eats your afternoon: reformatting a table, renaming 60 files to one pattern, filling a template from a list, turning a Word list into Excel columns, translating table headers, or checking that every report has the fields it must have. No judgement is involved, which is exactly why AI suits it.

Grunt work is also where small errors hide, because nobody reads boring output closely. So the check step is a count or a spot-check.

Grunt work: check required fields

You are an admin assistant who checks reports against a checklist.

I am pasting 25 KRT activity reports, numbered R01 to R25. Every report must have: date, venue, attendance count, organiser, and at least one photo caption.

Make a table with one row per report and one column per required field. Write Yes or Missing in each cell. Do not fix or fill anything in.

State how many rows your table has. It must be 25. Then I will open three reports myself and compare.

What stays with you

The fifth bin matters most. Some work never goes into D, R, A or G:

Decisions and approvals: approving a grant claim, choosing which programme goes ahead, signing a letter.
Sensitive mediation calls: speaking to two sides in a community dispute, or judging when a situation needs escalation.
Anything with personal data: complainants’ names, MyKad numbers, addresses, phone numbers.

AI can still help around these. It can draft a neutral opening for a mediation session, or check a claim has all its documents. But the decision, the conversation and the data stay with a person.

Sort it yourself

Your turn. Eleven tasks from a KPN week, all fictional. Tap the bin each one belongs in. Watch for the “Keep human” ones; they are the tasks people most often hand over by mistake.

The whole path, in one picture

You have now walked the whole path. Each step answers one question, and DRAG is the last one: what does AI take on?

AIM structures the brief: Actor, Input, Mission.
Domain research gives the facts: the Act, the sources and what the research must deliver.
Verb, Object, Classifier turns those facts into findings anyone can trace.
Bloom and Simpson say what the work must know and do.
Work activities define the job: who does what, with what, to which standard.
OODA gives the decision loop: Observe, Orient, Decide, Act.
OSEMN gives trusted data: Obtain, Scrub, Explore, Model, iNterpret.
DRAG decides what AI takes on, and what stays with you.You are here

The eight steps of the path. Each one hands something to the next. Tap a name to revisit that step.

Put together, it reads like one sentence. AIM structures the brief. Domain research and Verb, Object, Classifier give you the facts, written so anyone can trace them. Bloom and Simpson say what must be known and done. Work activities define the job. OODA gives you the decision loop. OSEMN gives you trusted data. And DRAG decides which parts of the job AI takes on.

AI drafts, you decide

If you remember one line from this path, make it this: AI drafts, you decide. Every prompt on this page ends with a check step for that reason. The AI can draft the minutes, but the chair signs them. It can sort the complaints, but you decide what the pattern means. It can flag a KRT whose activity dropped, but you pick up the phone.

That is not a limit on AI. It is how the work stays yours, and how you stay accountable for it.

Check your understanding

You have reached the end of the learning path. Go back to any step in the picture above when you need it, and keep the rule close: AI drafts, you decide.

Nice work!

Lesson complete

You made it to the end. Take the quiz above if you have not yet, for bonus XP.

0 XP
0 day streak