Start with tasks, not job titles
Most headlines about AI and work ask whether a job will “disappear”. Most of the research asks a smaller question: which tasks inside a job could software do, and what happens to the rest? A job is a bundle of tasks. Changing one part of the bundle can reshape the job without removing it.
This article gathers dated studies from 2025 and 2026 and sorts what they say from what they cannot say. None of them is a forecast of your own career, and none of them settles the debate.
Four studies, four different questions
World Economic Forum, January 2025
Surveyed over 1,000 employers in 55 economies about 2030. Respondents expected about 170 million jobs to be created and about 92 million displaced, a net gain of about 78 million. These are employer expectations, not measured outcomes.
International Labour Organization, May 2025
Estimated that about one in four workers worldwide is in an occupation with some exposure to generative AI, and 3.3 percent are in the highest-exposure group. The authors expect transformation more often than elimination.
Stanford Digital Economy Lab, revised August 2026
Used payroll data through June 2026 and found that employment of 22- to 25-year-olds in AI-exposed occupations was about 19 percent below where it would have been relative to less-exposed peers, with no comparable gap for experienced workers.
A fourth study asked a different question again. MIT’s Project Iceberg, reported in late 2025, modeled about 151 million U.S. workers across 923 occupations. It estimated that about 11.7 percent of U.S. wage value, roughly $1.2 trillion, is tied to tasks that current AI could technically perform, compared with about 2.2 percent of wage value affected by visible AI adoption. The researchers describe it as a measure of technical capability, not a forecast of layoffs.
Reading these numbers carefully
The numbers cannot be added together, because they measure different things: expectations, exposure, observed hiring, and technical overlap. The ILO’s “exposure” means a job contains tasks that AI might do, not that the job is lost. The WEF’s figures come from a survey, and the same report finds that roughly 41 percent of employers planned to reduce staff as AI automates tasks, while 77 percent planned to upskill workers. The Stanford authors call their results early, descriptive indicators rather than proof of cause. They report that the gap weakens when education is taken into account, and that some divergent trends predate generative AI.
Which skills show up in the research?
The WEF report names technology skills in AI, big data, networks and cybersecurity among the fastest growing. It also lists analytical thinking, resilience, flexibility, leadership and collaboration as core skills, and says nearly 40 percent of job skills are expected to change by 2030. The OECD describes AI as a substitute for human labour in some tasks and a complement in others. The Stanford study points the same way: the declines it finds are concentrated where AI mainly substitutes for human work, while employment is flat or rising where AI mainly complements it.
From that evidence, here is an editorial reading, which is our interpretation rather than a research finding. Skills that tend to complement software share some features.
- List your tasks. Write down what you actually do in a week, not your title.
- Mark the routine ones. Which follow clear rules and produce standard output?
- Mark the judgment calls. Where do you weigh trade-offs, context or people’s feelings?
- Mark the physical and relational parts. Care, repair, negotiation and teaching often rely on being present.
- Plan one stretch. Pick a skill that lets you direct, check or combine tools rather than compete with them.
Human skills that deserve a closer look
- Judgment in context. Deciding what matters in a messy situation, and noticing when an output looks wrong. The AI Index notes that models show uneven performance and that benchmark success does not always carry over to real-world tasks.
- Working with people. Collaboration, teaching, caregiving and negotiation. WEF respondents rank collaboration among core skills. The how and why of teamwork is explored in the hidden teams behind famous breakthroughs.
- Learning how to learn. If tasks shift, the ability to pick up new methods matters. See how humans learn to do what once seemed impossible.
- Handling failure. New tools produce new mistakes. What failure really contributes to invention looks at why trying and revising matters.
- Physical skill. The 2026 AI Index, as summarized by MIT Technology Review, reports robots succeeding in only about 12 percent of household tasks, a reminder that dexterity in the real world remains hard for machines. That is a snapshot and may change.
Underneath these lies a broader question about how human thinking differs from machine output, which we take up in what makes human intelligence different from AI.
A sensible stance
Neither alarm nor dismissal fits the evidence well. The more defensible position is to watch your own tasks, stay curious about the tools, and build abilities that are hard to reduce to a script. Policy choices about training, hiring and safety nets will matter too, and they are decisions for societies, not only for individuals.
