Who Buys the Slop When the Worker Is Fired

In 1914, Henry Ford raised the daily wage at his River Rouge plant from $2.34 to $5. The immediate reason was worker turnover — Ford was losing his entire workforce every year and the cost of retraining was eating his margin. The longer-run consequence, however, was the one the Ford Motor Company would not stop talking about for the next forty years. A $5-a-day wage made a Model T affordable to the man who built it. Ford had, by accident and then by design, produced his own consumer.

The Silicon Valley AI boom is running the Ford experiment in reverse. The labs are firing the knowledge workers who are the only people in the rich world who can currently afford a $200-a-month API subscription. The firing is the product. The customer and the worker are the same person, and the plan is to eliminate the worker and keep the customer.

This does not work. It has never worked. An economy cannot survive the abolition of its own consumer.

The Ford identity

There is a specific, unglamorous truth about mass-production capitalism that was worked out over the twentieth century by economists of every political persuasion. Keynesian, Marxist, institutionalist, even the more honest of the neoclassicals eventually got to it. The truth is this: an economy that produces mass-market goods needs a mass market to buy them. The mass market is the working class, employed at a wage high enough to absorb the output.

Call this the Ford identity. Wages are a cost to the individual employer and a revenue source to the economy as a whole. The individual employer wants to cut wages. The economy as a whole cannot afford for every employer to succeed at this.

The Ford identity is what made the New Deal, the Marshall Plan, and the postwar Keynesian consensus make sense to capital. The Ford identity is what produced the labour-capital bargain of the 1945–1975 Golden Age in the rich world: high wages, high consumption, high profits, stable growth. It worked. Everyone who pretends it did not work is lying about what their grandparents remember.

The Ford identity is what the AI industry is currently betting against.

The pitch deck and its missing page

Every AI investor pitch deck of the last three years contains a slide titled something like “Market Size.” The slide estimates the “total addressable market” for AI-replacing-a-job as a function of the total wage bill of the targeted occupation. The formula is straightforward. The US spends roughly a trillion dollars a year on software engineers. If AI captures 20 per cent of that, the AI market is $200 billion a year. Project out ten years, apply a discount rate, calculate a valuation.

The slide never contains the follow-up question. If AI captures 20 per cent of the software-engineer wage bill, what happens to the aggregate spending power of software engineers? Specifically: what happens to their ability to buy the downstream products of the companies that use their labour? What happens to the $200-a-month AI subscription those engineers were buying? What happens to the SaaS products the engineers evaluated and approved?

The answer is in the Ford identity. The wage bill is somebody’s revenue. The somebody is the rest of the economy. The rest of the economy includes the firms the investor is also holding stock in.

Every pitch deck is implicitly betting that the labour-to-capital transfer happens faster than the demand collapse. This is a bet on timing, not on economics. It can work at the individual firm level, for a while. It cannot work at the whole-economy level, ever.

The rich world is firing its own customer

Consider the user base of the current generation of AI products. Who actually pays the $20, $200, $2,000 a month? Not the global poor. Not the median wage earner in the rich world. The paying customers are concentrated in a specific segment: knowledge workers in developed economies, small businesses that employ those workers, and enterprises that pay per-seat for the same workers’ access.

The firing, when the firing comes, will be concentrated in exactly the same segment. The mid-career copywriter, the junior analyst, the entry-level software engineer, the paralegal, the translator, the customer-success specialist, the instructional designer. These are the paying customers. These are the people the firms using AI are trying to not hire.

The industry’s response is a muttered “productivity gains will be reinvested.” In what. Where. By whom. The productivity gain, in the current structure, accrues to capital as a share of profit. Capital’s marginal propensity to consume a $200 subscription is roughly zero, because capital owns the product. Capital’s marginal propensity to reinvest in more capital goods is high, which increases productivity further, which removes more wage income, which reduces consumer demand. The feedback loop is not stable.

This is the problem that the phrase “secular stagnation” was invented to describe, in the American economy of the late 1930s. It applied then. It applies now. Nothing in the AI investment thesis has grappled with it.

The Canadian version is worse

Canada has a specific vulnerability to this dynamic that does not appear in the American version of the analysis. Roughly 40 per cent of Canadian consumer spending is cross-border, denominated in goods and services produced in the United States, imported at a loonie-to-dollar exchange rate that is currently hovering around 74 cents.

When AI hollows out the Canadian knowledge-worker wage bill, the first-order effect is a reduction in Canadian consumer spending. The second-order effect is a disproportionate reduction in imports from the US, because consumer discretionary spending is import-heavy and staples are not. The third-order effect is a loonie depreciation, because the current account deteriorates, which makes every remaining AI subscription — priced in US dollars — more expensive for the remaining Canadian customers.

The Canadian knowledge worker is being fired by a process denominated in a currency the Canadian knowledge worker does not earn, for the benefit of firms that do not pay Canadian tax, providing a service the Canadian worker was paying for until they were fired. The loop closes. It closes against us.

”But new jobs will emerge”

The standard industry response is that every previous wave of automation produced new jobs that nobody could predict in advance. The blacksmith became an auto mechanic. The clerk became a software engineer. The radio operator became a network administrator. The pattern is robust. The pattern will hold.

Maybe. The pattern did hold in the 20th century, over fifty-year horizons, for economies that had strong unions, full-employment macro policy, industrial strategy, public higher education, and labour-friendly governments.

Canada in 2026 has none of these. The unions cover 28 per cent of the workforce, down from 38 per cent a generation ago. Macro policy is an inflation-targeting regime that does not treat full employment as a co-equal mandate. There is no industrial strategy in the 20th-century sense — there is a rotating list of procurement announcements. Public higher education is being hollowed by provincial cuts. The federal government has been, charitably, inconsistent on labour rights for two decades.

Even if the pattern of “new jobs will emerge” holds over fifty years, the transition is a generation long. A generation of Canadian knowledge workers, currently aged 25 to 45, will absorb the transition costs without any of the 20th-century cushions. “New jobs will emerge” is true on a graph. It is not a policy.

The argument

The generative AI economy, as currently structured, is betting that capital can extract the wages of the consuming class faster than the demand collapse. It is the reverse-Ford experiment, and it will produce a reverse-Ford result. Falling aggregate demand. Rising concentration. Falling profits, eventually, because the profits depend on the aggregate demand.

A rational industrial strategy would treat this as the single largest macroeconomic risk of the next decade. The Canadian strategy would include wage-protection mechanisms in affected sectors, a serious expansion of EI, retraining that is actually funded, and an explicit aggregate-demand target for the post-transition economy.

Canada has none of this. The industry does not want any of this.

The investor who is long on AI is implicitly short on the consuming class. Watch what happens to the consuming class. Watch what happens to the portfolio. The Ford identity will collect.