Tasks, not job titles: how to X-ray a job

Key facts

• The ILO's 2025 index puts one in four workers worldwide in a job with some exposure to generative AI, and says jobs changing is the most likely result.^[1]
• Clerical work is the most exposed group in both the ILO's 2023 and 2025 studies.^[2][1]
• Employers surveyed by the WEF in 2025 said 47% of work tasks are done mainly by people today, and expected a roughly even split between people, technology and both by 2030.^[3]
• In Anthropic's November 2025 data, 52% of Claude conversations helped a person work and 45% handed a task over.^[4]
• e-MASCO, Malaysia's official occupation list, gives the tasks for each job code.^[5]

A job title such as "engineer" or "nurse" hides what people actually do all day. Every job is a bundle of tasks: answering calls, checking a drawing, explaining a result to a worried parent. AI tends to change some tasks long before it changes a whole job, so to see how a career might change, look inside it.

The big studies work this way too. The International Labour Organization (ILO) scores thousands of single tasks and adds them up by occupation.^[1] Anthropic, the company that makes the Claude AI model, matches Claude conversations to a US task list called O*NET.^[6]

How to X-ray a job

1. Find a task list. Start with e-MASCO, Malaysia's official occupation list from the Ministry of Human Resources, which gives the tasks (tugas) for each occupation code.^[5] The US O*NET list adds detail.^[7] Best of all, ask someone who does the job what fills their week.
2. Write eight to ten tasks as a verb and an object: "schedule appointments", "explain a test result". If a task has an "and" in it, split it. Mark any task you guessed rather than saw, read or were told.
3. Label each task. Two labels are fine.

   • AI can assist: AI drafts, checks or speeds it up; a person still decides.
   • Might be automated: software could do most of it with little human input.
   • Still needs a physical or social setting: hands, presence, trust or judgement in the room.
   • New work that may appear: work that exists because AI is now used.
4. Give a reason and say how sure you are: low, medium or high. Some of your labels should be low. That is honest, not weak.
5. Support one label and challenge one assumption. Back your surest label with your best evidence. Then pick your least sure label and ask: what would change my mind?
6. Check with a practitioner. Which tasks take most of their week? What is missing? Which label would they change?

On the Day 2 Job X-ray worksheet the same four labels are called augmented, automated, left alone and new.

Worked example: a clinic's medical secretary

Illustrative only: these labels are our reasoning for a classroom exercise, not a forecast. A practising clinic administrator has not yet checked them. Tasks are adapted from e-MASCO 3344 Setiausaha Perubatan and O*NET 43-6013.00.^[5][7] The ILO's 2025 index places medical secretaries in gradient 3, "significant exposure", with an average task score of 0.53 on a 0 to 1 scale.^[1]

Support one label. Claim forms are structured and repetitive, the kind of clerical task the ILO rates as most exposed.^[1]

Challenge one assumption. "Answering calls will be automated" assumes callers are calm and speak one language. Now change the setting: the caller is elderly and describing chest pain. That label moves to "needs a social setting", because someone has to judge the urgency and stay calm on the line. Try other settings too: many visitors speak different languages, or have no smartphone.

What the research says, and how sure it is

• Most jobs are more likely to change than vanish. In 2023 the ILO judged that generative AI's main effect is likely to be augmenting work rather than fully automating occupations.^[2] Its 2025 update says transformation of jobs is the most likely impact, because most occupations include tasks that need human input.^[1] These are estimates of potential exposure, not counts of jobs lost.
• Clerical work is the most exposed. In the 2023 study, 24% of clerical tasks were rated highly exposed and another 58% medium.^[2] In 2025 clerical occupations still had the highest exposure, and some highly digital professional and technical jobs had become more exposed.^[1]
• About one worker in four. The 2025 index puts one in four workers worldwide in an occupation with some generative AI exposure, and 3.3% of global employment in the highest group (4.7% of women's jobs, 2.4% of men's).^[1]
• Employers expect a shift, but it is their guess. In the World Economic Forum survey published in January 2025, employers estimated that 47% of work tasks are done mainly by people, 22% mainly by technology and 30% by both, and expected a roughly even three-way split by 2030.^[3]
• What people actually use AI for. In Anthropic's February 2025 data, 57% of Claude conversations helped a person do the work (augmentation) and 43% handed a task over (automation).^[6] In November 2025 data the split was 52% and 45%.^[4] This is one company's tool and not a representative sample of all AI use: it shows usage, not jobs.^[8]
• Early signs for young US workers. Stanford researchers reported in August 2026 that employment of US workers aged 22 to 25 in highly AI-exposed jobs was about 19% below where it would otherwise be, mostly because fewer were hired. They found no sign of job losses across the whole economy, and say the pattern is descriptive, not proof that AI caused it.^[9]

None of these studies is about Malaysia, and none can tell you what your own career will look like in ten years.

Different assumptions, different futures

Anthropic's economic scenario explorer, published in September 2026, lets you set five assumptions: how many tasks AI can do, how widely it is used, how much it does on its own, how much more productive it makes people, and how long displaced workers take to find new work. It then shows a possible US economy in 2030.^[10] Anthropic calls it a simplification and says real outcomes may differ materially. Use it to see that one changed assumption changes the whole story. It is not a forecast for Malaysia or of what you will earn here.

For facilitators

Use a practitioner-checked task list where you have one; learners without pre-work use the medical secretary example. Look for reasoning, not a right answer, and record the supported label and the challenged assumption in "The future" section of the Future Map. Stop any verdict that a job is "safe" or "finished". If you show the scenario explorer, say plainly that it models the US.

What this means for you

• Judge a career by its tasks. Two people with the same job title can spend their weeks very differently. The tasks you would enjoy matter more than the title.
• Exposure is not a verdict. A high exposure score means some tasks may change. It does not mean the job disappears, and a low score does not mean nothing will change.
• Your labels are guesses until someone who does the job checks them. A practitioner's correction is better evidence than any table, including ours.
• The tasks AI cannot easily take tend to involve hands, presence, trust and judgement. Notice which of those you are good at and enjoy.

Your next step

1. Pick one job you are curious about, find its code on e-MASCO(opens in a new tab) and write eight to ten of its tasks as a verb and an object. Label each one and mark how sure you are.
2. Ask someone who does that job, a relative or a practitioner on Day 2, which task takes most of their week and which of your labels they would change.