Malaysia Advertised 33% More Jobs and Cut 38% More People

Two numbers that should not both be true

In the first half of 2026, Jobstreet by SEEK reported that Malaysian job ads grew 33 percent year on year. The Department of Statistics put unemployment at 3.0 percent in June, near the lowest it has ever been, with labour force participation at 70.9 percent.

On those numbers alone, this is a good year to be looking for work.

Then you open PERKESO's Loss of Employment file. Between January and June 2026, 47,053 Malaysians filed for unemployment benefit. In the same six months of 2025 the figure was 34,005. That is a 38 percent increase in one year, and by mid-July the running total had reached 52,607.

Both of those come from official sources. Neither is a projection. So what is happening?

Look at who got cut

This is the number that reorganised everything else for me.

High-skilled workers make up 30.5 percent of the Malaysian workforce. They account for more than half of all retrenchments. Managers, professionals, executives, technicians. They are being cut at roughly 1.7 times their share of employment, and about 40.85 percent of those losses trace to business closures and downsizing.

So the jobs are not disappearing. They are being swapped. Employers are cutting expensive mid-level office staff and advertising different roles, more operational and more technical. Same country, same six months.

A swap does not show up in an unemployment rate. The person who was cut takes another job at lower pay and quietly rejoins the employed column. Where it does show up is in a different statistic: 35.2 percent skill-related underemployment, and a 49.3 percent wage penalty for the people it happens to. RM2,096 a month instead of RM4,131.

Nobody in this picture shows up as unemployed. They show up as swapped, and paid less.

The gap between the headline and the file

What the other job boards say

One country's data cannot tell you whether this is a Malaysian story or a world story. So I read the others.

Four things fall out of reading them side by side.

1. Malaysia is the outlier

Plus 33 percent while Australia runs minus 5.8 and the United States minus 2.9. Whatever is happening to hiring globally, Malaysia is not in it. Semiconductors, data centres and infrastructure are pulling the other way.

2. The entry-level collapse is a US story, not a world story

Stanford's Digital Economy Lab, using ADP payroll data through June 2026, found US workers aged 22 to 25 in AI-exposed occupations sitting 19 percent below where they would otherwise be. It works through reduced hiring, not firing. That finding is real and it is widely quoted.

It is also not global. India's fresher hiring is up 8 percent, with 0-to-3-year roles up 33 to 42 percent in hospitality, insurance, BPO and real estate. China's problem is 12.7 million graduates hitting one market at once, which is a supply glut rather than an AI effect. Indeed's own researchers note that English-speaking markets recovered software hiring faster than the rest. Be careful importing the American headline into a Malaysian conversation.

3. Everywhere, more people chase each job, even where jobs are growing

Australia: applications per job ad at the highest level on record. Germany: applications per active posting nearly doubled between 2023 and 2026. Two unrelated economies, the same signal. Some of that is AI writing the applications. It means a job-ad count is no longer a reliable measure of how easy it is to actually get hired.

4. Independent markets agree on where the growth is

Australia's only annual growth: engineering, construction, mining and energy, trades and services. The United States: loading and stocking up 11 percent, production and manufacturing up 8 percent. Europe: manufacturing labourers recorded the largest vacancy-rate rise of any group since 2019. And Malaysia's own critical list is full of technicians, welders, mechanics and operators.

Four separate economies, four separate platforms, pointing at the same kind of work. That is what a pattern looks like, as opposed to a forecast.

What Malaysia is actually short of

Malaysia publishes the answer every year and almost nobody in business reads it. The Malaysia Critical Occupations List, MyCOL 2024/2025, from TalentCorp under the Ministry of Human Resources. Eighth edition, 66 occupations across 18 sectors, up from 37 in the previous cycle. Notably, 77 percent of the employers surveyed were SMEs and microenterprises.

Sort those 66 by what kind of work they are and the shape appears.

The biggest single block on the national shortage list is hands. Welders and flame cutters. Aircraft technicians. Motor vehicle mechanics. Electrical mechanics and fitters. Steam engine and boiler operators. Heavy truck and lorry drivers. Crane and hoist operators. Generative AI cannot do one of them.

Nine occupations have appeared in all eight editions since 2015: finance managers, business services managers, ICT managers, industrial and production engineers, mechanical engineers, manufacturing professionals, software developers, IT system administrators, mechanical engineering technicians. Eleven years of the same shortage is not a trend. It is a structural feature of the country.

Five are new this cycle, and four of the five are hands: pharmacists and physiotherapists as the population ages, chemical and physical science technicians, draughtspersons and surveying technicians as infrastructure builds out, and motor vehicle mechanics.

The barbell

Put the retrenchment file next to the shortage list and the labour market stops looking like a ladder. It looks like a barbell. Weight at both ends, thin in the middle.

The World Bank ran a task-based study with ISIS Malaysia and found that 45 percent of Malaysian workers are in jobs with high (28 percent) or medium-high (17 percent) exposure to AI, concentrated among women, younger prime-age workers, clerical occupations and urban employees. Exposure rises from low-skill to mid-skill and then levels off at the top, where creativity, complex judgment and socio-emotional work stay hard to automate.

One caution the World Bank states plainly and I will repeat: exposure is not displacement. Whether AI substitutes for a worker or complements them depends on adoption pace, job redesign and complementary skills. That sentence is the difference between a credible position and a scare campaign.

The month-end report

Here is the same thing at the size of one desk. Picture an accounts department in a small company.

Two people. Both use AI. Both finish the month-end report in half an hour, work that used to take two days.

The first one finishes and sends it up. Fast. Genuinely faster than last year, and pleased about it.

The second one does one more thing before sending. She knows which cell AI usually gets wrong in her line of work, because she has watched it happen. Refunds booked as revenue, that one. She knows which document to pull to check it against. She checks. Then she signs.

Three months later the boss learns to open ChatGPT himself and discovers he can produce that report in half an hour too.

At that moment the first person's contribution has a market price, and the price is falling. The second person's has not changed at all.

What the boss is actually buying

Not the report. He can make the report.

What he is buying is that if the number is wrong and it reaches LHDN, somebody carries it.

AI cannot carry it. AI does not get fined. AI does not get called in to explain. AI does not lie awake at two in the morning. Accountability is not a task, so it is not automatable, and that is not a philosophical point. It is the reason the boss keeps a person in the chair.

Which means the stronger AI gets, the cheaper speed becomes and the more expensive accountability becomes. Those two move in opposite directions, and almost every AI career conversation I hear is about the side that is getting cheaper.

The one question

You do not need a framework for this. You need one question, and you can ask it tonight about any task you do.

When this is wrong, who gets blamed?

Ask it about every task on your desk

If the honest answer is that the AI got it wrong and nobody really carries it, that task is not durable work. It will keep getting cheaper.

If the answer is you, because you checked it, because you know where it usually breaks, because your name is on it, then AI cannot take that from you. Not because you are faster. Because you can be held to it.

What to do this week

1. Run the question over your own week. List the tasks you did. Mark each one: who gets blamed if it is wrong? You will find your job splits into two piles, and the piles are not the same size you expected.
2. Learn where AI breaks in your specific line of work. Not AI in general. Yours. Which number does it get wrong, which claim does it invent, which step does it skip. This is the knowledge that makes you the one who can sign, and it is not in any course because it is specific to your trade.
3. Write it down where it can be checked. A one-page list of the failure points and what you check them against. That page is the difference between saying you are careful and being able to prove it.

None of that requires a new qualification, a bootcamp, or two years of retraining. It requires you to stop selling the half of your work that got cheap and start naming the half that did not.

FAQ