The K Economy · Techvisory
Evidence base · worker-facing

The K-shaped economy and its impact on jobs

No credible body finds economy-wide AI job destruction. What is measured is redistribution — away from young workers holding codified knowledge, toward experienced and AI-skilled workers. That is a K even with zero net job loss, and for the person on the third rung the cause matters far less than the missing step.

Clerical support exposed
92%
1.65m workers in the highest generative-AI exposure quartile.
Tertiary working below qualification
36.1%
Up from 30.2% in 2015, at a 49.3% wage penalty.
Skilled share of new vacancies
24.6%
Against a 30.2% skilled share of existing employment.
Jobs highly affected, 3–5 yrs
697,000
TalentCorp estimate, stated June 2026.

Exposure by occupational group

Share of each group falling in the highest generative-AI exposure quartile. World Bank and ISIS Malaysia, published July 2025, computed on the 2021 Labour Force Survey.

Share of group in the top AI-exposure quartile

Clerical work is not the lowest-paid work in Malaysia, and it is by far the most exposed. The bottom row is the study's residual bucket — trades, plant and machine operation, agriculture and elementary occupations — not a safety score.

Clerical support workers1,654,000 workers
92%
Technicians & associate professionals988,000 workers
56%
Professionals746,000 workers
35%
Service & sales workers1,089,000 workers
26%
Managers125,000 workers
19%
Trades, plant, agriculture, elementary2,936 workers
0.2%
Read it as technical substitutability, not predicted job loss. Only 5% of Malaysian workers are in roles requiring predominantly high-exposure skills; 66% are low or medium-low.

The inverted U

Exposure does not fall as you climb. It peaks in the middle — by education, by age, and, counter-intuitively, it rises with pay.

By education

Post-secondary / diploma
46%
Tertiary
37%
Secondary
27%

By age

25–34
32%
35–44
29%
15–24
26%
45–54
24%
55–64
17%

By wage quintile

Top 20%
41%
4th
20%
3rd
11%
2nd
7%
Bottom 20%
6%
Who this describes

Diploma-qualified. Aged 25–34. Technical or clerical. Klang Valley. Paid above the median. In Malaysian terms, that is the M40 — and it is the segment the training market sells past: up to executives, down to the welfare track, never to the middle.

Exposure and retrenchment almost invert

Clerical work is 92% exposed and accounts for 7% of job losses. Professionals are 35% exposed and account for 26%. Exposure is technical; retrenchment is financial. The clerical adjustment runs through not hiring, not firing — Stanford finds separation rates in exposed occupations actually fell.

Malaysia, verified figures
FigureValueSource & period
GDP growth+6.0% y/yDOSM / BNM, Q2 2026 final, 14 Aug (supersedes 5.8% advance)
Unemployment3.0%DOSM, June 2026
Loss-of-employment claims52,607PERKESO EIS, 1 Jan – 16 Jul 2026
— H1 vs H147,053 vs 34,005+38%, PERKESO EIS
— PMET share51%PERKESO EIS
Formal-sector median wageRM3,027DOSM, reference March 2026
Skilled jobs created−4.2% y/yDOSM Employment Statistics, Q1 2026
Skilled share of new vacancies24.6%World Bank MEM, Apr 2026 (vs 30.2% of employment)
Tertiary working below qualification36.1%World Bank MEM, Apr 2026 (from 30.2% in 2015)
— wage penalty49.3%same
Job ads+33% y/yJobstreet by SEEK, H1 2026
Gini0.404 → 0.390DOSM, 2022 → 2024; poverty 6.2% → 5.1%

Industries, most exposed first

Insurance & takafulhighest in the country
91.8%
Real estate
87.4%
Financial servicesmedian RM7,700
87.1%
Employment activitiesrecruitment & staffing
78.6%
Legal & accounting
75.6%

Insurance and takaful is the most exposed industry in the country — and sits inside the second-highest-paying sector.

The global evidence

What is measured

The counter-case — strong, and we do not omit it
  • NY Fed, May 2026: declines in AI-exposed postings began before November 2022, and there is “no clear divergence” between junior and senior inside high-exposure occupations.
  • SIEPR, Jul 2026: unemployment since 2022 rose 0.77pp for the most exposed against 0.85pp for the least — the least exposed did slightly worse. About 20% of US firms use AI; about 5% report an employment impact.
  • Challenger Gray: US announced cuts Jan–Jul 2026 were down 41% year on year. AI is the leading stated reason — a rising share of a shrinking total, self-reported.
  • Confounders: the rate-hike cycle, post-pandemic normalisation, remote work weakening on-the-job training, and a graduate glut. Stanford's own February 2026 revision found the effect significant only after 2024 under firm-time controls.

What actually protects a worker

  1. Tacit over codified. The part of the job that could not be written into a manual.
  2. Complementarity, not low exposure. How the work is organised decides whether AI helps or replaces.
  3. Data literacy, not coding. Interrogating outputs and judging wrongness — not building models.
  4. Judgement, creativity, leadership, presence. The residual PwC finds growing.
  5. Physical or personally-licensed work. Trades and plant operation sit in the 0.2% bucket; in some professions accountability is legally non-delegable.

Six moves, in order

  1. Audit your week by task, not title — codified versus tacit.
  2. Move toward the decision, not the deliverable.
  3. Buy tacit knowledge with proximity.
  4. Become the person who judges AI output in your domain.
  5. Claim the HRD Corp levy. It is under-used, and only about 3.7% of places go to digital and ICT.
  6. If moving sideways, move toward the physical or the regulated.

What not to do: add a second codified qualification. Certified transferable knowledge is precisely what is being automated — and 36.1% of Malaysia's tertiary-educated already work below their qualification at a 49.3% penalty.

Sources, corrections and known gaps

Principal sources. DOSM (Labour Force Survey, Employment Statistics, Salaries & Wages, HIS 2024) · PERKESO EIS · World Bank Malaysia Economic Monitor April 2026 · World Bank & ISIS Malaysia, Novel AI technologies and the future of work in Malaysia (July 2025, LFS 2021) · Jobstreet by SEEK · TalentCorp / MyMahir · HRD Corp 2025 disbursement data · Stanford Digital Economy Lab · PwC AI Jobs Barometer 2026 · OECD · IMF SDN/2026/001 · BIS Bulletin 130 · SignalFire · Indeed Hiring Lab · ILO · NY Fed · SIEPR · Challenger, Gray & Christmas.

Corrections applied after fact-check. “37% work below qualification, up from ~25%” became 36.1% up from 30.2% (the 25% was misread off a chart axis starting at 25). “Mid-skilled 56% exposed vs high-skilled 44%” is the composition of the most-exposed quartile, not an exposure rate — the inverted-U argument here rests on the education, age and wage cuts, which do check out. The OECD creativity figure is 41.9% “more important”, not 41.5%. The OECD clerical figure was inverted in our first draft: 34.0% said more important, 16.1% said less. The GitHub Copilot 55.8% is an overall treatment effect, not a novice subgroup result. The IMF 3.6% is a regional estimate conditional on rising AI-skill demand, five years after the shock. A claim that “B40 income grew fastest” was dropped — DOSM publishes no B40/M40/T20 growth disaggregation for 2024.

Known gaps. No PERKESO figure after 16 July 2026 · DOSM Q2 2026 Employment Statistics unpublished · no HRD Corp 2026 levy or disbursement data · no Malaysian declining-occupations list from job-posting data · the Malaysian exposure data rests on the 2021 LFS · no Malaysian equivalent of the ADP payroll series · the PERKESO occupational breakdown appears only in secondary coverage, not a primary publication.

Redesign the role, then take the skill course

The six moves above are the prerequisite. The next step is not another diploma — it is applied AI training matched to your tasks, at ai-courses.edisontkp.com.

Employers: request a workforce review · occupation placements.