A Job Is a Bundle of Tasks: Which Occupations Fade and Which Change?

In 2013 two researchers at the University of Oxford estimated that about 47% of US jobs were "at risk" from computerisation. Three years later an OECD study estimated that only about 9% of jobs were automatable on average across 21 of its member countries. Which figure is more accurate? Both are accurate in what they measured; the difference lies in how they counted.

A Job Is a Bundle of Tasks

Frey and Osborne estimated the probability of computerisation for 702 occupations, a calculated probability and not a forecast that this share of people would lose their jobs. The OECD study held that counting the whole occupation inflates the risk, because many "high-risk" occupations contain tasks that are hard to automate, so it analysed the mix of tasks within each occupation. The share ranged across countries from about 6% in Korea to 12% in Austria, and the researchers suggest that the cause is work organisation, earlier investment in automation and workers' education.

Take an illustrative example: an accountant combines entering ledger entries, auditing them, and explaining the financial position to a hesitant client. The first is likely easier to automate than the third. That is the practical meaning of the "bundle".

What Did Language Models Add?

Those studies preceded large language models, which work on speech, text and code rather than on mechanical movement. In 2023 researchers estimated that about 80% of US workers may have at least 10% of their tasks affected by them, and about 19% may have half or more of their tasks affected. Exposure spans all wage levels and may be higher in better-paid jobs, contrary to the old assumption that automation begins with manual work. The authors made clear that they do not predict when such models will be developed or adopted, and that "exposure" is an estimate of the possibility of being affected, not of what will happen.

Lists Are a Snapshot, Not Destiny

In the World Economic Forum's 2025 report on jobs, postal clerks, bank tellers and data-entry clerks are among the fastest-declining roles, and clerical and secretarial jobs lead the decline in absolute numbers. In relative growth, big-data specialists, fintech engineers and AI and machine-learning specialists lead. But the largest growth in absolute numbers was in jobs that are mostly non-digital: farmworkers, delivery drivers, construction workers and salespersons. So the scene cannot be reduced to technologists rising and everyone else falling.

The figures also point to the effect of human decision: 50% of employers said they plan to move staff from declining roles to growing ones, and 40% said they plan to reduce staff whose skills become less relevant (World Economic Forum, 2025). Retraining is a choice; it is not an automatic result of technology.

The Entry Gate

A study by researchers at Stanford University examines extensive US payroll data up to June 2026 and finds that employment of people aged 22 to 25 in the occupations most exposed to AI is about 19% lower than it would have been had it kept pace with peers in the least exposed occupations, with no comparable gap among the experienced. It is an observational study in an American market, and this is its revised version of August 2026, so it does not establish causation and does not necessarily apply to our market. But it warns that the danger may strike the entrance to a career before its middle, leaving the beginner no opportunity in which to learn.

How Should We Read This?

Better than memorising a list is for each worker to ask about the tasks of their job:

• Which routine, documented tasks might the tool get ahead of me on?
• Which tasks demand judgement, trust, relationship and accountability?
• How do I make the tool work for me and move my time to the second kind?

The employer, for its part, should announce its retraining plan before deciding on dismissals, for a person is not a line in a budget. In that lies mercy towards people and justice in taking the means, and provision rests with God alongside effort. The skills required are discussed in the article "The Skills That Stay Valuable When Machines Master the Work".