AI will replace most white-collar jobs within a decade

What's this about?

People disagree about whether AI will take most office jobs within the next ten years. AI can do some work tasks, but jobs include many different tasks.

What supporters say

• AI can write, sum up notes, study data, and answer routine messages.
• Many office workers may see AI change at least some of their daily tasks.
• Tests show AI can help workers finish writing and help jobs faster.
• Firms may hire fewer new workers when AI handles simple starter tasks.

What critics say

• Changing tasks does not mean AI will erase whole jobs.
• Workers still need judgment, trust, teamwork, and knowledge of real-world problems.
• Job losses may hit some young software workers, but we cannot prove AI caused them.
• AI effects differ between jobs, workers, and types of work.

The bottom line

AI will likely reshape office work by about 2036. But the evidence does not show that AI will replace most office jobs by then.

AI is likely to reshape office work over the next decade, but the evidence does not show that it will replace most white-collar jobs by about 2036. The distinction matters: AI can change or speed up many tasks without making entire occupations unnecessary.

The case for

The argument for major disruption starts with the broad reach of generative AI. Professional, administrative and clerical jobs contain many tasks involving writing, summarising, analysis and routine communication—work that current AI systems can assist with or, in some cases, automate (see Figure 2). One study estimated that about 80% of U.S. workers could see at least 10% of their tasks affected by large language models, while roughly 19% could have half or more of their tasks affected. International assessments by the IMF and Goldman Sachs also find especially high exposure in advanced economies and in office-based occupations (see Figure 3). 1

Experiments show why employers may have an incentive to adopt the technology. In one controlled study of professional writing, generative AI helped people finish work faster while improving average quality. In customer support, workers using an AI assistant became about 14% more productive on average, with the biggest gains going to less experienced or lower-performing staff. These results suggest that companies could reorganise work and need fewer people for some tasks. 1

Any disruption may first appear not as mass layoffs, but as slower hiring, fewer entry-level roles and changing skill requirements. Analyses of online job postings have linked greater AI exposure to weaker demand in some occupations, though the effects are uneven. One observational study also found employment declines among some younger software developers after generative AI became widely available. That may point to particular risks for entry-level workers, but it does not prove AI was the cause or show a trend across white-collar work. 2

The stronger version of the claim depends on what comes next. If future AI systems become reliable enough to manage multistep workflows, work across business software and require little supervision, they could go beyond today’s limited tasks. Rapid improvements in capability could therefore make present-day forecasts too cautious. But existing productivity studies do not establish that these advances will happen—or that they will translate into the disappearance of whole occupations. 3

The case against

The central problem is that task exposure is not job replacement. A job is usually a bundle of tasks, and AI may handle some while leaving others to people. The major exposure studies themselves draw this distinction: an affected task might be assisted, redesigned or combined with human work rather than eliminated. 4

Many barriers also stand between a technically capable system and a fully automated job. International assessments from the ILO and OECD point to the need for human interaction, accountability, contextual judgment, regulation, organisational change and worker acceptance. Cost and the availability of skills also matter. Clerical work may be especially exposed, but the ILO’s overall conclusion is that generative AI is more likely to augment most occupations than fully automate them, because jobs often include duties machines cannot easily take over.

The direct evidence so far supports this more limited picture. Customer-support, writing and consulting experiments generally show people working better with AI, not being removed from the workforce. The consulting research also found worse results when people used AI on tasks beyond its strengths, underlining the continued need for human judgment, oversight and careful task selection. 5

Even when AI replaces particular tasks, the effect on jobs is uncertain. Higher productivity can lower costs, increase output and create demand for complementary work or new kinds of roles. Research on automation and OECD analysis suggest that firms may need more workers for tasks AI does not perform, or may expand production enough to offset some displacement. 6

The bottom line

The evidence strongly supports the view that AI will bring substantial, uneven restructuring to white-collar work, especially in routine, digital and easily measured tasks. It may alter hiring, career paths and wages, with some groups—possibly younger workers in exposed fields—more vulnerable than others.

But the evidence does not support the stronger claim that AI will eliminate or make unnecessary more than half of existing white-collar jobs within a decade. Current studies measure exposure and productivity on specific tasks, not economy-wide job elimination. Early hiring and employment signals are limited, short-term and cannot yet show cause or predict long-run outcomes. 7

The decisive uncertainty is whether future AI reliability and adoption will outpace the forces that preserve human roles: supervision, regulation, changing demand and complementary work. For now, the best-supported forecast is transformation rather than wholesale replacement.