Managing AI Upheaval
An Answer to Yglesias
When the near future is murky, thinking too far out is sketchy. AI tail risk has been over reported. I’m interested in the middle of the catastrophe to paradise distribution. Tell me that AI will increase GDP ten points over baseline during the next five years and I would respect you. Anything more optimistic than that unlikely. The extraordinary burst of AI progress in 2024 was not sustainable, yet progress is still brisk.
Artificial intelligence is poised to disrupt the American workforce. Jobs once considered stable—software development and data analysis—are increasingly vulnerable to automation. Clerical and support work is also on the verge of redundancy. If tens of millions are displaced in short order, the resulting shock could destabilize households and entire communities, resulting in hundreds of thousands of deaths of despair.
Yet the upside is staggering. AI promises productivity gains that could sharply raise GDP and living standards. For the first time in history, we may have the tools to liberate large segments of humanity from mechanical, repetitive labor. The central question is political: how do we divide the windfall?
Matt Yglesias has recently argued for expanding the safety net and deploying make-work programs. I agree. He also acknowledges that the traditional center-left playbook may fall short—and he deserves credit for that honesty. In an era when many commentators evade the implications of transformation, Yglesias’s humility is refreshing. I take his uncertainty as an invitation for new voices. As a student of economic history and a small businessman, I offer the following.
1. The Scale of the Challenge
About 13% of total U.S. payroll is at risk of redundancy from AI technologies over the next 2 to 8 years (author’s estimate based on BLS Occupational Employment and Wage Statistics, Brookings AI exposure data, and McKinsey Global Institute reports). Worst realistic case, AI will increase unemployment by 10% by 2030. These roles won’t disappear overnight and many will evolve, but significant restructuring is very likely. Below is a breakdown of the most exposed occupational categories:
Occupational Category Key Automatable Tasks Est. Share of U.S. Payroll
Software Developers & Coders Code generation, debugging, documentation ~2.5%
Clerical & Administrative Staff Data entry, scheduling, record management ~3.0%
Data Analysts & Researchers Report generation, forecasting, data interpretation ~2.0%
Customer Service Representatives Call handling, chat support, basic issue resolution ~1.5%
Bookkeepers & Accounting Clerks Transaction recording, reconciliations, financial reporting ~1.5%
Paralegals & Legal Assistants Legal research, document review, drafting ~0.5%
Medical Billing & Coding Staff Insurance claims, medical coding, billing follow-up ~0.5%
Drivers (Truckers, Delivery, etc.) Route navigation, delivery scheduling, long-haul driving ~2.0%
Total Estimated Impact ~13.5%
These figures reflect payroll, not headcount, and focus on jobs where core tasks are highly automatable with modestly improved versions of existing tools. The payroll at risk is large, but the risk is not unmanageable. Crucially, AI will only replace human labor when it offers better performance. A lawyer who fires a paralegal to use AI that drafts poor documents will hurt his reputation. A trucking company won’t replace drivers if crashes spike. The logic is simple: if AI is used, the job is getting done. When a firm adopts AI, its output will likely hold steady—or rise—and its costs will fall. Unless displacement creates a large fiscal drag through depressed demand, aggregate GDP will increase. With smaty, targeted intervention GDP will increase robustly.
Clerical workers and bank tellers illustrate what technological displacement really looks like. The number of typists in the U.S. fell by more than 60% between 1980 and 2010 (BLS), yet documents became faster to produce and easier to revise. Similarly, bank teller employment declined by over 20% since 2000, while inflation-adjusted wages have stagnated or fallen (BLS, 2023). Yet banking has become far more convenient—apps, ATMs, and online tools let customers handle routine tasks instantly. The pie has grown even though less work is needed to bake it. This leaves the question of how the slices are allocated.
2. AI is Inherently Deflationary
AI replicates the dynamics already visible in electronics manufacturing: steep productivity growth and collapsing real prices. Between 2000 and 2020, quality-adjusted prices for televisions dropped by 98%, and computers by nearly 90% (BLS Consumer Price Index). As AI moves into sectors like healthcare, education, finance, and law, it will slash the administrative costs of services that have long resisted automation. Tasks like scheduling, document review, claims processing, and basic research are now well within AI’s reach. The net effect will be downward pressure on prices and a freeing of skilled human labor for more complex work. This dynamic—efficiency gains matched by price declines—creates fiscal capacity.
3. Policy Opportunity: Fiscal Stimulus Without Inflation
In ordinary settings, creating money or expanding public spending risks inflation. However, if productivity is soaring and prices are falling, that risk evaporates. Policymakers can distribute income to dislocated workers through more generous unemployment and retraining subsidies, without overheating the economy. AI doesn’t just reduce prices. It raises real GDP. That expands the tax base, allowing the state to extract more revenue without raising rates.
This is fortunate, because traditional tax-the-rich redistribution remains politically fraught. America is better at buying treasury bonds with newly created money than passing tax hikes. That’s not ideal, but it’s not catastrophic. If we can maintain demand and avoid a sharp transitional downturn, the long-term fiscal outlook looks better, not worse. With higher output and broader capacity, the U.S. economy will be in a position to afford more—as long as we avoid a disemployment-driven slump in aggregate demand.
This strategy is a riff on the old Keynesian playbook. Rather than using stimulus to combat slumps in aggregate demand, we can run the economy hot for several years because strong deflationary pressures are right around the corner. The best part is we are already doing this. Our deficit is 6.3% of GDP, compared to a post-World War II average of 2.5%. Inflation is at acceptable levels and should remain there if AI is deployed quickly. In political terms, the most important thing is avoiding mass unemployment, which would increase political pressure to restrict AI.
4. Restricting AI Won’t Work
Workers and professionals whose jobs are lost or threatened will push for restrictions. Some may succeed temporarily. However, such bans will collapse under competitive and geopolitical pressure. No business will locate major operations in states that outlaw cost-saving tools. National bans would send capital abroad. In military terms, surrendering AI capacity is unthinkable. AI may soon determine tactics, logistics, and targeting better than any general. China knows this. If the United States dithers, it risks strategic irrelevance. Like nuclear fission or the steam engine, AI cannot be uninvented. The challenge is not to stop it, but to steer it.
Humanity is about to achieve its greatest victory yet in the war on drudgery. Fear of AI is both rational and a tragic policy failure. If workers cannot count on dignified lives during and after the transition, politics will grow more toxic. Politicians must offer something real: shorter workweeks, longer vacations, and more time to do as we please. Those who default to thinking we should work as hard as our parents are dinosaurs. With the right social guarantees and agile policy, the next generation can not only be richer and happier than us—but can enjoy a rate of progress unmatched in our species’ 200,000-year history.