Paul Glavin is an associate professor of sociology at McMaster University.
Scott Schieman is professor of sociology and Canada Research Chair at the University of Toronto.
Alexander Wilson is a graduate student in sociology at the University of Toronto.
If one of the leading voices in artificial intelligence is right, a storm could be brewing for white-collar workers. Dario Amodei, chief executive officer of AI startup Anthropic, recently predicted that within the next five years AI could wipe out up to 50 per cent of entry‑level white‑collar jobs – and push unemployment to 10‑20 per cent.
Skeptics dismiss this as Silicon Valley hype or the predictable boosterism of a CEO selling his technology.
History supports the skeptics. Previous job-killing innovations rarely met dire predictions. Economists often reassure us that technology complements workers rather than replacing them, boosting productivity.
When ATMs arrived in the 1970s, for example, banks redeployed tellers into customer service and financial advisory roles requiring more creativity and human judgment.
But AI’s advance may be different: it cuts across countless occupations and today’s AI models already draft emails, summarize documents, write code and generate reports – striking at the heart of white-collar work.
Our ongoing Canadian Quality of Work and Economic Life Survey (C-QWELS) reveal how these changes are registering with Canadian workers. From September 2023 to April 2025, the share who said machines or computers were “very” or “somewhat” likely to take over much of their current job rose from 17 to 23 per cent. Strikingly, this increase occurred almost entirely in the last six months – and was driven by university graduates, whose concern jumped nearly 10 percentage points over that period.
Few workers anticipate losing their position outright to automation (9 per cent). Workers instead increasingly expect adaptation rather than replacement. The share who believed their role would be adapted to work alongside new technology rose to 23 per cent in April from 16 per cent.
Yet the word adapted deserves scrutiny. While it may signal continuity, it doesn’t guarantee that work remains meaningful, skill-intensive or fairly compensated.
As sociologists of work, we see several reasons for concern, even if fears of immediate and widespread AI displacement are potentially overblown. Claims of a “white-collar bloodbath” and “job apocalypse” – that is, “something alarming happening to the job market” – certainly make for attention-grabbing headlines (and, at this stage of the purported advancements, they probably should).
Erosion before displacement
If predictions about future AI capabilities are even partly correct, we may be seeing only the early contours of what lies ahead. Already, signs are emerging that the conditions and perceived value of some white-collar work is shifting. At Amazon, software engineers report that AI tools are accelerating production timelines while reducing time for thoughtful coding and increasing output expectations. According to New York Times reporting, many now spend more time reviewing and debugging machine-generated code than writing their own. The work remains, but its character is changing – less autonomous, more pressure, and arguably less fulfilling.
This shift in work quality may be creating broader economic ripples. Barclays economists have found that workers in AI-exposed roles are experiencing measurably slower wage growth – nearly three-quarters of a percentage point less per year for every 10-point increase in AI exposure. Employers may already be recalibrating the value of these positions, even as hiring continues.
Uneven impacts
Different forms of white-collar work may face vastly different futures under AI, depending on professional autonomy and control over the technology. Consider radiologists, initially seen as vulnerable given AI’s strength in image analysis. Yet, the profession has grown, with AI enabling faster analysis rather than replacement. Crucially, radiologists retain control. They make final diagnoses, communicate with patients and carry legal responsibility. Here, AI complements human expertise in what economists refer to as Jevons Paradox – where technological efficiency increases demand by making services cheaper and more accessible.
Medical transcription offers a more cautionary tale. As AI speech-to-text tools improve, transcriptionists have shifted from producing reports to editing and error detection. In theory, this sounds like higher-skilled oversight work. In practice, it often means scanning AI output under time pressure and reduced job discretion. While jobs such as this one still exist, their perceived value is diminishing. The U.S. Bureau of Labor Statistics projects a 5 per cent employment decline between 2023 and 2033 – and given the rapid improvement in transcription models, that estimate may prove conservative.
Adaptation isn’t necessarily promotion
AI will undoubtedly create new roles and opportunities, particularly where human judgment remains essential. But we shouldn’t assume this future will preserve job quality. The story of retail banking offers a sobering lesson: automation first increased the number of teller jobs – but didn’t raise pay. Ultimately, tellers weren’t replaced by machines but by digital banking, shifting many to call centre jobs with less autonomy and lower wages. Even in the absence of widespread job displacement, AI may follow a similar pattern –reshaping many jobs in ways that reduce discretion, increase surveillance and erode its overall value.
There remains considerable debate about how disruptive AI will be. But amid that uncertainty lies a risk of public complacency – or even disengagement from the issue. As Canadians, we need a sustained and open conversation about how these workplace changes are unfolding and where they might lead.
This column is part of Globe Careers’ Leadership Lab series, where executives and experts share their views and advice about the world of work. Find all Leadership Lab stories at tgam.ca/leadershiplab and guidelines for how to contribute to the column here.