The Library / Concepts
Economic Brief · April 2026 · The Library · Concepts
Why the price of human knowledge work is about to collapse. In eight chapters.
60%
of jobs in advanced economies have significant AI exposure
OECD · 2023
300M
global jobs at risk of automation by AI
Goldman Sachs · 2023
1
Canadian AI law passed in the last ten years
Stanford AI Index · 2026
0–15%
success rate of mid-career retraining programs
OECD ALMP · Brookings
30,000
Oracle layoffs in one morning, capital redirected to AI
Q1 2026
01 · The Mechanism
In any market, price is set by supply and demand. Labour is a market. The forecasts model the quantity displacement. They do not model the price effect. The price effect is what determines whether this is a manageable adjustment or a structural crisis.
Shock 1: Supply
Millions of skilled workers, suddenly available.
Shock 2: Demand
Direct substitute, available at a fraction of the cost.
Result
Price ↓
The price of human knowledge work falls. Not gradually. In a cascade.
A worked example, in five sentences.
A firm employs 100 knowledge workers. It deploys AI. One operator, with the right tools, produces the output of the full team.
The firm retains the operator. It releases 99.
The 99 do not disappear. They are professionals with mortgages. They learn the same tools, often free. Within months, all 99 apply for the one remaining role.
The salary for that role does not hold at senior-consultant levels. It craters, because the candidates cannot afford to refuse.
AI does not merely eliminate jobs. It eliminates the pricing power of the humans who remain.
Sources: OECD Employment Outlook 2023 · Goldman Sachs (Hatzius / Briggs / Kodnani), March 2023 · IMF Working Paper, Cazzaniga et al., January 2024 · Bewley, Why Wages Don't Fall During a Recession, Harvard UP 1999.
02 · The Plumber Fallacy
A common response to AI displacement is "learn a trade." Plumbers cannot be automated. Electricians require physical presence. The advice is correct about automation resistance. It is wrong about price resistance.
Today in Thornhill, Ontario, it costs $400 to hire a plumber. That price reflects constrained supply, years of certification, and a guild structure built on scarcity. Now imagine 300,000 displaced knowledge workers. Capable. Driven. Watching YouTube tutorials on pipe repair. Going to Home Depot. Offering to fix the neighbour's leak for $100. Then $75. Then $50.
The plumber's skills are superior. But in a price war against a hundred desperate former analysts, the premium erodes.
$400 → $50 Thornhill plumber. Not because AI took his job. Because 102 desperate ex-analysts already quoted $60. The premium does not survive a price war it was never built for.
Source: The dynamic generalizes from labour-economics first principles, supported by NBER WP 32487 (Bloom et al. 2024) on freelance-rate compression in AI-exposed categories.
03 · The Adoption Inversion
The most reassuring myth in the AI debate is that big institutions move slowly. The reason they were slow is the reason they will now be fast.
The reason banks ran COBOL for forty years is the reason they are about to replace knowledge workers in eighteen months.
Corporations only act when the math is unambiguous. For decades, the math wasn't there. Custom software cost millions. ERP migrations failed. Cloud felt risky. The short-term profit case was murky, so the file room kept its job and the mainframe kept humming.
The math is here now.
A senior knowledge worker costs $150,000 to $300,000 a year. The AI seat that produces equivalent output for a wide swath of cognitive tasks runs roughly $20 a month. That is not a forecast. That is a SaaS line item on a budget right now.
The one thing we know about corporations is this: they move quickly when the spreadsheet says move. The spreadsheet says move.
Annual cost of one senior role, vs. the AI seat that replaces a wide swath of it
Sources: StatCan Labour Force Survey 2024 · Robert Half Salary Guide 2025 · published consumer and enterprise pricing of frontier-model providers, 2026 · sector-adoption analyses, McKinsey Digital and BCG, 2024.
04 · The Receipts
Tech layoffs in Q1 2026 alone: ~80,000 workers, with roughly half attributed by employers to AI. By April 24, the cumulative 2026 figure had already passed 95,000. We are still in April. This is the first inning.
30,000
Oracle. One morning.
Q1 2026. Not financial distress, Oracle is among the most profitable software firms on Earth. The capital was explicitly redirected to AI infrastructure. The largest single-company event of 2026 to date.
Bloomberg, Reuters · Q1 2026
8,000
Meta. Ten percent of staff.
Announced April 23, 2026. Including 6,000 open roles eliminated outright (no backfill). Meta said the move was to free capital for AI infrastructure spend. Two days before this brief was published.
CNN Business, CNBC · April 23-24, 2026
7%
Microsoft. Buyouts.
Approximately seven percent of US workforce offered buyouts on April 24, 2026. Quieter mechanism, same direction. Capital is moving to AI capex. Headcount is moving down.
CNBC, Yahoo Finance · April 24, 2026
8 firms
The 10,000-club.
Eight companies announced AI-attributable layoffs of 10,000+ each in 2026: Accenture. Amazon. Citigroup. Dell. Intel. Microsoft. TCS. UPS. The cuts are not isolated. They are a pattern.
Layoffs.fyi aggregation · April 2026
15–40%
Freelance rate compression.
Across copywriting, translation, basic code, and graphic design on major platforms, since GPT-4 (March 2023). The leading indicator was already on screen long before headline layoffs.
NBER WP 32487 (Bloom et al., 2024) · Upwork Future Workforce Report 2024
↑ / →
Margins up. Headcount flat.
Major technology firms reporting expanding operating margins with declining or flat hiring. Productivity gains flowing to shareholders, not workers. The Great Transfer in real time.
FactSet S&P 500 commentary · 2025-2026
In Q1 2026 alone, the industry's own statements attribute roughly half of tech-sector layoffs to AI. The companies are now saying, on the record, what the thesis predicted. This is the first inning.
05 · The Cascade
Wage compression in knowledge work propagates outward. Each stage feeds the next. The cascade is the reason the policy clock matters.
Now → Q4 2026
Knowledge work compression
AI directly displaces lawyers, analysts, developers, marketers, consultants. Headcount drops. Surviving roles face price pressure from displaced peers flooding back into a shrinking pool of seats.
2027
The Plumber Fallacy
Displaced knowledge workers, unable to find work at previous rates, move into trades and physical service work. They compete on price. They learn fast. Wages compress in fields that were supposed to be automation-proof.
2027 → 2028
Demand destruction
Wages fall across sectors. Consumer spending contracts. Restaurants, retail, entertainment, travel, all face revenue declines. They cut staff. The displacement that began in AI-exposed sectors propagates through the broader economy.
2028
Fiscal crisis
Income tax revenue contracts. Consumption tax revenue contracts. Demand for the social safety net surges. The fiscal gap widens precisely when government capacity to address it shrinks.
2028 → 2029
Housing trigger
Canadian mortgages were stress-tested for interest-rate shocks. They were never stress-tested for income shocks. When professions are repriced by 30 to 50 percent over three years, defaults rise. Concentrated, in the demographics most exposed.
The cascade is not a list of risks to be managed in sequence. It is a sequence of feedback loops, each one accelerating the next. The window for proactive policy is measured in quarters, not years.
06 · The Policy Vacuum
Stanford AI Index 2026 counted AI-related laws passed in the last decade. The score, ten years in:
Stanford AI Index 2026, Ch. 8, Fig. 8.4.3. Read the full vacuum analysis on the landing page →
One number tells you the legislative count. It understates the point. The relevant question is: of the policy instruments that exist elsewhere to capture some portion of AI's productivity gains for citizens, how many has Canada built?
AI windfall / excess-profits tax
Captures monopoly rents from AI productivity gains, as the 2022 Canada Recovery Dividend did for banks and insurers.
None proposed
Compute / GPU-hour levy
Taxes the production function directly. Discussed in EU and US academic policy, not in Canada.
None proposed
Mandatory Workforce Impact Assessment
Requires firms to disclose displacement before deploying labour-replacing models in regulated sectors.
None
Federal labour-displacement reporting
Tracks and prices the transition. US OMB M-24-10 has a thin version for federal contractors. Canada has nothing comparable.
None
Stanford AI Index 2026 · OECD ALMP review, Employment Outlook 2021, Ch. 4 · Brookings (Muro / Maxim / Whiton) 2019.
The capital that paid those salaries is not gone. It moved. The instruments to share it back with citizens are the instruments we have not built. That is the vacuum, and the vacuum is the choice.
Postscript
Canada's Budget 2024 set out a $2.4 billion package "to secure Canada's AI advantage." The breakdown is on the public record. $2 billion for the AI Compute Strategy (development capacity). $200 million for AI start-ups. $100 million for the NRC AI Assist Program (industrial AI deployment). For the Canadian AI Safety Institute, the body explicitly tasked with studying frontier-AI risk: $50 million.
That is roughly two percent of the package. Forty-six dollars on development for every one dollar on safety. The arithmetic of a country that has decided which side of the trade-off it is on, and the side it picked is not the side this brief argues for.
There is no global rush to develop. The frontier labs are ahead by years. There is, however, a global window to get safety right, and we only get that right once. The campaign is not asking Canada to brake alone. It is asking Canada to convene the global table where safety becomes the world's first concern. The case is for a coordinated international body, not a unilateral pause. That is Demand 1.
An international treaty and supporting UN agency akin to the International Atomic Energy Agency are necessary to standardize access permissions, cybersecurity countermeasures, safety restrictions, and fairness requirements of AI globally.Yoshua Bengio, written testimony, U.S. Senate Judiciary Committee, July 26, 2023.
Sources: Government of Canada, Budget 2024 (Department of Finance, April 16, 2024). ISED launch announcement, Canadian AI Safety Institute, November 12, 2024. Government of Canada, Canadian Sovereign AI Compute Strategy, December 2024.
What to do, in three steps
Sign once. Read the long form when you have an hour. Send the brief to the person in your life who needs to see it.
Back to the Library · Demand 1: Emergency Global AI Governance Summit, convened by Canada · Demand 2: Real Liability for AI-Caused Harms · Demand 3: National AI Transition Body