AI isn't replacing workers. It is quietly eliminating the jobs they would have had

Everyone is asking the wrong question about artificial intelligence.

The debate has fixated on whether AI will replace jobs. That makes for good headlines but misses what is actually happening in real time — and what should be far more worrying to employers, policymakers and workers alike.

The real question is this: what happens when the jobs are never even created?

A recent study by Anthropic offers an answer that should make everyone in labour economics pause. Despite the rhetoric about mass displacement, there is no corresponding spike in unemployment in the occupations most exposed to AI. Workers are not, at least not yet, being shown the door in large numbers.

But that is not the story.

The story is what is missing: hiring.

There is no wave of layoffs. There is a hiring freeze, quietly embedded in technological change.

The data shows that since generative AI tools such as ChatGPT became widely adopted, hiring in certain entry-level, AI-exposed roles has fallen by roughly 14 per cent. Other categories of employment have remained largely stable.

That divergence is not accidental but structural.

The roles being affected are not fringe occupations destined for obsolescence. They are the traditional on-ramps into professional life: junior programming, customer support, basic data processing and routine analytical work.

These are the jobs that once formed the foundation of career progression. And in some of them, AI can already perform a substantial portion of the task mix, particularly in coding-related functions, where it can now assist with a majority of routine outputs.

This does not eliminate the function. It reduces the number of bodies required to perform it.

And employers, being rational actors, respond accordingly.

The logic is both simple and entirely predictable. If one employee supported by AI can now do the work that previously required two, then two hires become one. In some cases, one becomes zero.

There is no redundancy process. No termination dispute. No severance negotiation. No headline-grabbing layoff announcement.

Instead, there is simply a requisition that is never opened.

From a legal perspective, nothing occurs. But from an economic perspective, everything changes.

For years, conventional wisdom held that automation would first displace lower wage, lower skilled labour. This time, the pattern is inverted.

The roles most exposed to AI are disproportionately held by better-educated, higher-paid workers in white-collar occupations. On average, they earn significantly more than those in less affected roles.

These are not manufacturing jobs. They are office jobs. Professional services roles. The very positions that have long been considered stable entry points into the labour market.

Much of the policy debate remains anchored to unemployment statistics, but that is a lagging indicator. By the time unemployment rises materially, the adjustment has already occurred.

The more important metric is hiring velocity. That is where the early signals are appearing.

If entry-level hiring contracts, the pipeline of future professionals contracts with it. That is not a short-term labour market fluctuation. It is a structural change in workforce development.

Careers do not begin with senior roles. They begin with access.

This transition is unfolding against a weak growth backdrop. Employers are already cautious. Capital allocation is tighter. Headcount expansion is under scrutiny.

AI provides an additional incentive to delay or avoid hiring altogether.

Why hire two juniors when one senior employee, armed with AI tools, can deliver comparable output?

The result is predictable: slower hiring today becomes fewer experienced workers tomorrow. And unlike cyclical layoffs, these foregone entry points are not easily reversed.

Employment law is built around termination, not non-creation. Wrongful dismissal regimes, reasonable notice frameworks and severance entitlements all assume an employment relationship that has already commenced.

But what legal remedy exists for a role that was never created? For a job that was never posted? For an opportunity that was eliminated upstream by technology?

The answer is none.

From a legal standpoint, there is no dismissal. From a human capital standpoint, there is a silent erosion of access.

Policymakers and commentators remain focused on unemployment rates. That is comforting, but misleading. AI is not yet producing mass unemployment. But it is already doing something more subtle — and in many ways more consequential.

It is reducing the creation of entry-level roles that allow people to begin careers at all.

For young workers, that is the real disruption. Not displacement, but absence.

And it is occurring quietly, without legal recourse, and with remarkably little public scrutiny.

Howard Levitt is senior partner of Levitt LLP, employment and labour lawyers with offices in Ontario and Alberta, and British Columbia. He practices employment law in eight provinces and is the author of six books, including the Law of Dismissal in Canada. Arash Omidvar is an articling student at Levitt LLP.