Research

AI and White Collar Jobs — What the Research Actually Says

9 min read·April 7, 2026

The Gap Between Headlines and Data

"AI Will Replace Half of All Jobs." "Your Job Is Safe From AI."

Both headlines are wrong. The research is more interesting and more useful than either.

Here is what eight major studies actually found.

What Anthropic Found

Anthropic's 2026 Labor Market Impacts report is the most granular analysis of AI's actual impact to date.

The key finding: actual AI usage is a fraction of theoretical capability.

For computer and mathematical workers, AI could theoretically handle 94% of tasks. It is currently handling 33%.

That gap — between what AI can do and what it is doing — is the most important number in the entire debate. It represents both the risk (if adoption accelerates) and the buffer (there is still time).

The report also found no systematic increase in unemployment for highly exposed workers. But there is "suggestive evidence" that hiring of younger workers has slowed in exposed occupations.

What Microsoft Found

Microsoft Research analysed 200,000 anonymised Copilot conversations to measure AI's real-world applicability to different jobs.

Their finding: knowledge work and communication-focused roles have the highest AI applicability. Manual and physical roles have the lowest.

Importantly, they note that AI-task overlap does not equal job displacement. Historical examples show automation often increases employment in affected sectors.

What McKinsey Found

McKinsey's estimate that today's technology could automate 57% of current US work hours is alarming until you understand what it means.

Not 57% of jobs. 57% of hours worked involve tasks that AI could handle. Many jobs contain both automatable and non-automatable tasks. The automatable tasks get handed to AI. The human does the remainder plus new work.

What Brookings Found

The most practically useful finding in all of the research.

Adaptive capacity — financial resilience, transferable skills, network strength — predicts outcomes better than exposure alone.

Two people in the same job with the same exposure score can have radically different outcomes depending on their capacity to adapt.

What the WEF Found

The most shared headline: 92 million jobs displaced by 2030. The less shared context: 170 million new roles emerging. Net gain of 78 million jobs.

The disruption is real. So is the opportunity. The question is who is positioned to access the new roles.

What PwC Found

Workers with AI skills command a 25% wage premium. The gap between AI-fluent and AI-avoidant professionals is growing. AI fluency is becoming the most valuable credential in the modern workforce.

What Goldman Sachs Found

Up to 300 million full-time equivalent jobs could be exposed to automation globally. But Goldman's economists also note that AI historically creates more jobs than it destroys — the transition period is the risk, not the endpoint.