How might AI's rapid adoption uniquely impact white-collar workers compared to past automation waves, and what political consequences could arise?
The article mentions the importance of 'sequencing.' What does this mean in the context of AI adoption, and why is it crucial for managing potential negative impacts?
In the Weekly Wrap, editor-at-large Sean Speer analyses the big stories shaping politics, policy, and the economy in the week that was, exclusively for Hub subscribers.
Two weeks ago, the Weekly Wrapasked the provocative yet serious question: Can we survive the journey to AI abundance?
The basic premise of my essay was that while AI holds out enormous potential for scientific discovery, path-breaking innovation, and other socially beneficial outcomes, the risk is that the speed and scale of its short-run disruption may have such significant socio-political consequences that it undermines its long-run progress.
Another way to frame the question is: What if the magnitude of destruction overwhelms the creation in Joseph Schumpeter’s famous axiom about technological progress?
A new working paper by Eduardo Levy Yeyati, a senior fellow at Brookings and professor of economics at Universidad Torcuato Di Tella, seeks to broadly address the same question. In it, he argues that the debate over artificial intelligence and jobs has been focused on the wrong variable. As the paper’s working title, “Too fast to adjust,” implies, his central claim is that we’re too concentrated on how much gets automated rather than how fast the process unfolds.
Schumpeter’s concept of creative destruction remains the right framework for thinking about these issues. As Charles Lammam recently wrote for The Hub, markets are dynamic rather than static. Innovation renders some products, firms, and jobs obsolete while generating new ones. This process is inherently disruptive and even often painful, but ultimately enormously beneficial.
The history of capitalism is largely a history of creative destruction working as advertised. Yet it’s possible that this time is different. The sequencing may outpace our capacity to manage the transition. AI’s destruction arrives so fast and at such a large scale that it outstrips the creative phase before institutions have time to adapt.
The potential for AI’s rapid advancement to outpace society’s ability to adapt, leading to socio-political instability, needs to be noted. There are real concerns about the speed of AI adoption and whether the destructive aspects of AI’s implementation will overwhelm its creative potential, potentially overloading retraining systems and causing permanent worker displacement. Given that AI could fundamentally restructure the knowledge economy, there must be proactive institutional adaptation to mitigate potential political and economic disruptions.
What if the magnitude of destruction overwhelms the creation in Joseph Schumpeter’s famous axiom about technological progress?
If, as expected, the next wave of disruption falls more heavily on white-collar and professional workers—the lawyers, analysts, consultants, and programmers who populate the modern knowledge economy—the political consequences could be far less predictable.
The answer isn’t to slow innovation or expand the state’s direct role in the economy. It’s ensuring that the labour market’s reallocation infrastructure (as well as broader public policies) is prepared for adjustment at an unprecedented speed and scale.
Comments (3)
Ken Kopke
14 Mar 2026 @ 8:15 pm
I think Rob has it correct. Retrain white collar to what new (similar level of income producing occupation) is the big question with few answers. Some are saying go “Skilled Blue Collar”. This may be good in the short term for some. But given that Tesla is stopping car production in Fremont, CA to build human labour replacing robots at scale points to blue collar displacement at scale as well.
I think we need to escalate talking about that transition period between our current labour (blue and white) economy and the new one that is being built now. Yes there will be retraining to something at some point. But retraining as the major solution is insufficient given the potential speed of adoption and its employment ending impact.
The current reality in Canada is far too many people are living off every pay cheque with little or no financial buffer. And far too many are unable to get a foot hold to start their careers and families. A long transition will not go well for too many. There are only so many rent free basements to move into.
Are we now talking about Universal Basic Income? Universal Basic Services? Job Sharing? Some kind of new technology or productivity dividend? Some mix of these and other models? What will dampen the impact of so many families and those seeking to start careers and form families as we move into the new technological economy?
I am listening for a believable set of ideas that explain how this transition can be managed well.
How might AI's rapid adoption uniquely impact white-collar workers compared to past automation waves, and what political consequences could arise?
The article mentions the importance of 'sequencing.' What does this mean in the context of AI adoption, and why is it crucial for managing potential negative impacts?
What is the author's proposed solution to mitigate the risks of rapid AI adoption, and why does it emphasize a pro-market approach?
There are early indications that the pace may indeed be unusual. Surveys suggest the share of firms using AI rose from roughly half in 2023 to close to 90 percent by 2025. Corporate investment patterns are moving in a similar direction, with many large companies now reporting that a meaningful share of their capital budgets is dedicated to AI deployment. And new research points to nascent signs of disruption for entry-level white-collar work. That last point may prove particularly significant. Much of the political reaction to globalization and automation over the past two decades was concentrated among industrial and regional working-class communities that experienced manufacturing decline. If, as expected, the next wave of disruption falls more heavily on white-collar and professional workers—the lawyers, analysts, consultants, and programmers who populate the modern knowledge economy—the political consequences could be far less predictable. Levy Yeyati’s model, which focuses on worker displacement, retraining, and re-employment, provides a framework for understanding what happens when technological change outpaces the labour market’s ability to adjust. When automation displaces workers, they enter a retraining pipeline with limited capacity. Fast AI adoption compresses this displacement into a shorter window. The same total number of displaced workers arrives in a fraction of the usual timeline and overloads the retraining system. Workers who rationally assess their prospects—including long queues, depressed wages, and uncertain re-employment—exit the labour force permanently rather than wait. And once they leave, the risk is that they don’t come back. Levy Yeyati’s key insight can be simplified as follows: two countries could automate exactly the same share of jobs and end up in the same place, but the one that gets there faster will have left far more workers permanently behind along the way. The historical parallel the paper reaches for is instructive. When Henry Ford reorganised Highland Park around the moving assembly line in 1913-14, output per worker tripled within 18 months. The technology was unambiguously productivity-enhancing. But the transition was brutal. The pace of reorganization was so relentless that annual worker turnover hit 370 percent in large part because workers simply walked off the line rather than endure the monotony and speed of the new system. Ford eventually responded by more than doubling wages to $5 per day as a market correction, and turnover collapsed almost immediately. The episode illustrates a broader point: technological progress often arrives through sudden organizational restructuring. When it does, the speed of adjustment can matter as much as the technology itself. That sequencing problem is precisely what’s worth watching now. Technological revolutions often begin with incremental improvements before suddenly reorganizing production around a new capability. Recent developments suggest AI may be approaching that moment. The release of Claude Code and comparable agentic AI tools has materially reduced the friction that has historically slowed the diffusion of new technologies into the workplace. Previous AI tools were powerful but still required significant technical intermediation to deploy at scale. These new tools don’t. The barrier to widespread adoption in the knowledge economy has, for all intents and purposes, been eliminated. The optimists will reach for the bank teller story. ATMs were supposed to automate tellers out of existence. Instead, branches got cheaper to operate, and so banks opened more of them, and teller employment actually rose. It’s a genuine parable. But it’s only half the story. As a new essay by technology writer David Oks reminds us, what eventually reduced bank teller employment was the iPhone. Not because the iPhone automated teller tasks, but because it made the entire experience of visiting a bank branch largely irrelevant. The ATM substituted tasks. The iPhone made them obsolete. Agentic AI tools look more like the iPhone than the ATM. They don’t just automate specific knowledge work tasks within existing firm structures. They collapse the intermediary layer altogether. The last three decades of the knowledge economy were built substantially on outsourcing: firms hiring specialist companies and workers to perform legal, analytical, financial, and administrative functions too expensive and complex to bring in-house. Think of examples like Bloomberg, Deloitte, and Salesforce. That layer existed because of friction. Agentic AI tools are rapidly removing it. Firms can now bring those functions in-house at a fraction of the cost and with a fraction of the headcount. The junior lawyers, analysts, consultants, researchers, and project managers who populated that layer aren’t being automated within an existing paradigm. The paradigm itself is being restructured around them. Recent volatility in the share prices of several software and business-services firms suggests that equity markets may already be beginning to price in that possibility. The result is mostly predictable. Productivity will rise. Prices will fall. This is Schumpeterian capitalism working as intended. But the economic logic of creative destruction doesn’t guarantee a smooth political transition. Periods of rapid economic restructuring have historically been politically turbulent. The mechanization of early 20th-century industry—including the spread of Ford’s assembly line—coincided with intense labour conflict across North America and Europe, culminating in episodes such as the Winnipeg General Strike of 1919. The causes were complex and extended well beyond factory technology itself. The key point, though, is that when economic change outpaces society’s ability to absorb it, the political response can be volatile. Recent experience offers a reminder. Comparatively modest disruptions associated with contemporary automation, globalization, and regional industrial decline have already fueled a wave of populist politics across much of the advanced world. If the scale and speed of AI-driven disruption prove significantly larger, it’s reasonable to ask whether the socio-political reaction could be even more destabilizing. The question, as Yeyati puts it, is whether institutions can absorb AI-induced displacement at a speed and scale that exceeds the historic norm. His policy conclusion cuts against the instinct to reach for the state. The retraining pipeline needs to be built up before the crisis rather than during it. Stronger institutions and faster adoption, in other words, are complements rather than trade-offs. That’s a fundamentally pro-market and pro-growth framing. The answer isn’t to slow innovation or expand the state’s direct role in the economy. It’s ensuring that the labour market’s reallocation infrastructure (as well as broader public policies) is prepared for adjustment at an unprecedented speed and scale. Yeyati’s paper closes with a line that should command our collective attention. As he writes, “the cost of getting the sequencing wrong is not temporary dislocation but permanent exit—workers who lose the train and never find another.” Creative destruction has long been capitalism’s engine of progress. But its success has always depended on a delicate balance between disruption and adaptation. If the pace of technological change begins to outrun society’s ability to adjust, the risk isn’t simply economic dislocation—it’s that the political system itself becomes increasingly volatile and unpredictable. That’s why the challenge posed by AI isn’t whether creative destruction will occur. It almost certainly will. The question is whether our institutions can keep pace with it.
Sean Speer is The Hub’s Editor-at-Large. He is also a university lecturer at the University of Toronto and Carleton University, as well as a think-tank scholar and columnist. He previously served as a senior economic adviser to Prime Minister Stephen Harper.
Comments (3)
I think Rob has it correct. Retrain white collar to what new (similar level of income producing occupation) is the big question with few answers. Some are saying go “Skilled Blue Collar”. This may be good in the short term for some. But given that Tesla is stopping car production in Fremont, CA to build human labour replacing robots at scale points to blue collar displacement at scale as well.
I think we need to escalate talking about that transition period between our current labour (blue and white) economy and the new one that is being built now. Yes there will be retraining to something at some point. But retraining as the major solution is insufficient given the potential speed of adoption and its employment ending impact.
The current reality in Canada is far too many people are living off every pay cheque with little or no financial buffer. And far too many are unable to get a foot hold to start their careers and families. A long transition will not go well for too many. There are only so many rent free basements to move into.
Are we now talking about Universal Basic Income? Universal Basic Services? Job Sharing? Some kind of new technology or productivity dividend? Some mix of these and other models? What will dampen the impact of so many families and those seeking to start careers and form families as we move into the new technological economy?
I am listening for a believable set of ideas that explain how this transition can be managed well.