We're All in
This Together

The Question We Keep Getting Wrong

The conventional approach to AI workforce policy starts with a simple question: who will be displaced? Which occupations? Which demographics? Which regions?

This diagnostic impulse is understandable. It's how we've approached every previous technological transition—from the Luddites to the Trade Adjustment Assistance Act. Identify the losers. Target the interventions. Move on.

But here's the problem: that framework assumes the disruption is narrow enough to target. It assumes we can draw a circle around the affected workers and design programs specifically for them—while the rest of the economy hums along undisturbed.

The evidence from the first two years of generative AI tells a different story. This isn't a surgical disruption affecting one corner of the labour market. It's a systemic transformation affecting nearly everyone—just differently.

Approximately 60% of the Canadian workforce is now exposed to AI-related job transformation. But that statistic obscures more than it reveals. The professors and the administrative assistants who support them are both "exposed"—one gets a productivity boost, the other gets a layoff notice. The senior software engineer and the junior developer are both "affected"—one becomes more valuable, the other becomes redundant. The public sector (74% exposed) and the private sector (56% exposed) face the same technology—but with wildly different institutional capacities to respond.

1. The Universality of Disruption

The "High Complementarity" Illusion

Current AI labour market analysis divides workers into two categories: "high complementarity" (AI makes you more productive) and "low complementarity" (AI replaces you). Statistics Canada estimates 31% of Canadian workers fall into the replacement risk zone, while 29% are in the augmentation zone.

This framework is useful for short-term planning. It's dangerously misleading for policy design.

Why? Because the boundary between "complementarity" and "substitution" isn't fixed—it's a moving target that shifts with every model improvement. AI is already reading CT and MRI scans with radiologist-level accuracy. It's generating legal briefs, writing code, and producing strategic analyses that would have required senior professionals two years ago. The comfortable assumption that "high-skill cognitive work" is safe doesn't survive contact with GPT-5.

The handloom weavers of the 1810s thought they were in a "high complementarity" relationship with the spinning machines—until the power loom arrived. Their "Golden Age" of high wages and expanding employment (1788–1820) was actually a trap. The very success of mechanized spinning created the economic incentive to mechanize weaving. Within two decades, wages collapsed 75% and most adult weavers never successfully transitioned.

The Demographic Distraction

Some analysts emphasize that 57% of workers in AI-disrupted roles are women, compared to 43% of men. This is accurate. It's also largely irrelevant for policy design.

The gendered distribution of AI exposure reflects the gendered distribution of the current workforce—women are overrepresented in administrative support, data entry, and customer service roles that happen to be highly automatable. But men are overrepresented in professional services, finance, and technical roles that are also highly automatable.

Designing separate "AI transition programs for women" versus "AI transition programs for men" would be like designing separate climate adaptation programs based on eye colour. The exposure cuts across demographics. The policy response should too.

The Real Divide: Augmentation vs. Displacement

The meaningful distinction isn't between men and women, or public and private sector, or high-skill and low-skill. It's between workers whose organizations use AI to expand capacity versus workers whose organizations use AI to cut costs.

Consider University-Rosedale—one of Canada's most AI-exposed ridings. A professor at the University of Toronto who uses AI to analyze more research, teach more students, and produce more papers is experiencing augmentation. The administrative assistant who used to schedule meetings and format documents is experiencing displacement. Same institution. Same technology. Opposite outcomes.

The difference isn't inherent to the technology or the occupation. It's a function of organizational choices about how to deploy AI—choices that policy can influence.

2. What Six Centuries of Disruption Teach Us

The printing press. Steam power. Electrification. The automobile. The computer. The internet. Each of these technologies eventually created more jobs than it destroyed and lifted living standards dramatically. But "eventually" could mean 50 years—and the specific workers displaced often never recovered.

Five patterns recur across these transitions:

Pattern 1: Productivity gains consistently take 15–50 years to materialize. Steam power was invented in the 1710s but didn't drive significant productivity growth until the 1830s. Electric motors were commercially available in the 1880s but didn't transform manufacturing until the 1920s—when factories were completely redesigned around distributed power rather than central shafts. Robert Solow observed the same pattern with computers in 1987: "You can see the computer age everywhere but in the productivity statistics." The resolution didn't come until the mid-1990s.

Pattern 2: Displaced workers face persistent wage penalties even after economic recovery. The British Industrial Revolution eventually doubled living standards—but real wages for working-class families stagnated or fell from 1780 to 1840. The handloom weavers, telephone operators, and newspaper journalists who lost their occupations to technology rarely returned to their previous earning levels. The aggregate gains masked concentrated losses.

Pattern 3: Government retraining programs show modest effectiveness across six decades of evaluation. From JTPA through WIOA, the evidence is consistent: short-term training adds little value (roughly $1,600–1,800 in increased earnings over 30 months for adults, no significant impact for youth). What works: longer-duration education, intensive individual support, and integration with actual employer demand.

Pattern 4: Policy responses evolved from repression to the welfare state—but slowly. The British government's initial response to industrial displacement was the Frame-Breaking Act of 1812, which made machine-breaking a capital offense. Sixty people were executed. It took another century to build unemployment insurance and labour protections. We have better options now—but only if we choose to use them.

Pattern 5: New jobs emerge—but require different skills, in different places, often a generation later. 60% of U.S. workers today hold occupations that didn't exist in 1940. But the ATM that eliminated bank teller jobs in one decade created IT support jobs in another. The displaced teller in rural Ohio couldn't simply become a software engineer in San Francisco. Geographic and skill mismatches create permanent losers even when aggregate employment grows.

The core lesson: technological transitions ultimately create more prosperity than they destroy—but only when policy actively manages the transition rather than leaving workers to market forces alone.

3. A Universal Transition Strategy: Four Pillars

If AI disruption is universal, the policy response must be universal too—not in the sense of identical treatment for everyone, but in the sense of a comprehensive framework that addresses the full scope of the challenge. That framework rests on four pillars.

Pillar 1: Social Safety Nets That Match the Scale of Disruption

Canada's Employment Insurance system was designed for cyclical unemployment and temporary layoffs—not structural technological displacement requiring extended retraining. The maximum 45 weeks of regular benefits doesn't come close to covering the 1–2 years needed for meaningful reskilling.

Historical evidence is clear: income support during transition is among the strongest predictors of successful worker adjustment. The IMF found that regions with more generous unemployment benefits experienced smaller wage losses from automation. The New Deal provided relief during the Depression—not recovery, but a floor that prevented total destitution while the economy restructured.

Policy Direction: Extended EI benefits (up to 104 weeks) for workers in AI-displaced occupations. Wage insurance for workers 50+ accepting lower-wage reemployment. Automatic enrollment triggered by large employer AI deployment notices, rather than reactive claims after the fact.

This isn't about creating dependency. It's about providing the runway people need to adapt—because adaptation takes time, and poverty breaks people before they can rebuild.

Pillar 2: Education Ladders from Childhood to Late Career

The conventional response to technological displacement is retraining programs. Six decades of program evaluations tell us those programs don't work—at least not as currently designed. Short-term courses add marginal value. Generic "digital skills" training produces credentials without capabilities.

But there's a deeper problem: we've been thinking about education as a one-time front-loaded investment (K–12 plus post-secondary) followed by occasional "upskilling" patches. That model was already inadequate for the computer age. It's completely obsolete for the AI age.

The farmer's child learned alongside the farmer. The blacksmith's apprentice spent years watching problems being solved before taking on work independently. That model of embedded, experiential learning was destroyed by the separation of education from work—and we never replaced it with anything that actually prepares young people for the chaos of real economic life.

Kids entering the workforce today need portfolios and networks, not just resumes and entry-level jobs. They need exposure to real problems being solved in real time—not sanitized case studies in classrooms. The single "take your kid to work day" is laughably insufficient for an economy where entry-level job postings have declined 35% since January 2023.

Policy Direction: Restructure the education-work boundary. Expand co-op and work-integrated learning from a nice-to-have to a standard pathway. Bring young people into workplaces earlier and more systematically—not just to learn AI tools, but to learn how to identify problems and add value. For mid-career professionals, fund longer-duration credentials (minimum 6 months, preferably 1–2 years) with income replacement, not just tuition subsidies.

The goal isn't to teach people to compete with AI. It's to teach them to see what AI can't see—the problems worth solving, the contexts that matter, the judgment calls that algorithms will get wrong.

Pillar 3: Incentivizing Augmentation Over Elimination

Here's the fundamental choice every organization faces with AI: use it to do more work with the same people, or use it to do the same work with fewer people.

Current incentive structures push toward elimination. Public companies face quarterly earnings pressure. AI that replaces workers shows up as immediate cost savings. AI that enables workers to take on more clients, serve more customers, or expand into new markets shows up as uncertain future revenue.

Fifty years ago, an executive assistant was valued for shorthand note-taking and typing speed. Then Microsoft Office. Now it'll be AI skills. The role evolved because the value evolved. The assistant who can orchestrate AI tools, synthesize information, and manage complex workflows is more valuable than ever—if organizations are structured to capture that value.

Policy Direction: Tax incentives for companies that demonstrate revenue growth and workforce retention alongside AI deployment. Procurement preferences for federal contracts going to firms that show augmentation rather than elimination strategies. Required "transition impact assessments" for large employers deploying AI systems affecting significant workforce shares—not to block deployment, but to surface the organizational choices and ensure alternatives are considered.

This isn't about protecting obsolete jobs. It's about creating incentives for organizations to find the path where AI makes workers more valuable rather than redundant. That path exists—but it requires deliberate design, not default to the spreadsheet logic of cost-cutting.

Pillar 4: Enabling Entrepreneurship as an Escape Valve

If traditional employment contracts shrink, self-employment and business creation must expand to absorb displaced workers. This isn't just an economic necessity—it's an opportunity.

AI dramatically lowers the cost of starting certain businesses. Marketing, customer service, basic legal and accounting tasks that previously required staff or expensive contractors can now be handled by a solo operator with the right AI tools. The "solopreneur with AI" may become a significant employment category—but only if policy removes the barriers that currently make self-employment risky and difficult.

Current barriers: Regulatory complexity for small business formation. Access to capital (especially for non-traditional entrepreneurs). Risk aversion in social insurance design—EI clawbacks discourage gradual transition from employment to self-employment. Credential requirements that assume traditional employment paths.

Policy Direction: Streamline business registration and compliance for micro-enterprises. Expand small business financing and grants for displaced workers. Create a "business runway" provision allowing people to earn self-employment income while receiving partial EI during a startup phase. Fund "entrepreneur-in-residence" programs connecting displaced workers with startup support—not just for young tech founders, but for mid-career career-changers building service businesses.

If AI reduces the number of traditional jobs available, we need to make it easier for people to create opportunities rather than wait for employers to offer them.

4. What This Means for University-Rosedale

University-Rosedale is Canada's most educated federal riding—and one of its most economically stratified. From the personal support workers in St. James Town earning $35,000 to the executives in Rosedale earning $850,000, our constituents span the full spectrum of the Canadian economy.

That diversity means we can't hide from the universal nature of AI disruption. Every income quintile in this riding is exposed:

Maria in St. James Town works as a personal support worker. AI scheduling algorithms already control her shifts with 36 hours notice. Her husband's warehouse job has cut staff 30% in two years through algorithmic management. They have no savings buffer.

Jordan in the Annex is a communications coordinator using ChatGPT daily. They've become more productive—and watched freelance writing rates collapse from $300 to $75 per article. They're living inside the Golden Age trap, building the systems designed to automate themselves out of existence.

Priya near Dupont is a senior policy analyst at a federal agency. Her work is being transformed by AI document review tools. She's productive enough to survive—but she worries about the junior analysts who will never learn by doing, and about what her kids should study.

David on Admiral Road is a law partner whose firm has reduced junior associate hiring by 40% since 2019. He's making more money than ever. His kids are struggling to find entry points into any professional career.

Beth in Rosedale sits on three boards and is actively championing AI adoption—she pushed her telecom to accelerate its AI customer service rollout over the COO's objections, and it cut call handling costs by 35%. She's seen this pattern before at the bank where she was CFO: 30% workforce reduction while revenue doubled. She calls it "doing more with less"—which is what successful businesses have always done. But her documentary filmmaker daughter can't afford an apartment, and Beth senses that something in the social contract is fraying, even if she's not sure what to do about it.

None of these constituents needs targeted demographic intervention. They all need the same thing: a transition strategy that provides income security during disruption, education pathways that adapt to changing demands, organizational incentives that favour augmentation over elimination, and entrepreneurship opportunities for those who want to create their own futures.

5. The Path Forward

The debate about AI and work has been framed as a question of prediction: How many jobs will be lost? Which occupations are safe? When will the disruption hit?

That framing misses the point. The future of work isn't something that happens to us. It's something we build through the choices we make—as individuals, organizations, and a society.

Historical evidence shows that technological transitions ultimately create more prosperity than they destroy. But "ultimately" can mean 50 years of pain for the specific workers displaced. The question isn't whether AI will transform the economy—it will. The question is whether we manage that transformation so the gains are broadly shared and the costs don't fall on those least equipped to bear them.

The four pillars outlined here—robust safety nets, comprehensive education ladders, incentives for augmentation, and support for entrepreneurship—aren't a complete answer. They're a framework for building one. The subsequent papers in this series will develop specific policy proposals for each pillar, grounded in evidence about what's actually worked in previous transitions.

But the essential insight is simple: we're all in this together. Not because AI affects everyone identically—it doesn't. But because the disruption is broad enough that targeted interventions won't be sufficient, and because the choices we make about how to deploy AI will determine whether it augments human capability or replaces human workers.

Those choices are ours to make. Let's make them deliberately.