AI is changing how work gets done. So far, that change is showing up as adjustment rather than a jobs apocalypse: labour markets remain broadly resilient, total employment has continued to grow, and unemployment is close to where it has been for years. What the aggregate data cannot yet capture is displacement that arrives quietly, through positions not filled rather than positions eliminated. That is where we believe the real story is unfolding, and it is why the resilience of the headline numbers should not be mistaken for an all-clear.

Highlights

• The headline labour market data remains stable. Since ChatGPT’s release in November 2022, employment rates in Canada and the U.S. have moved within a narrow range and total employment has continued to grow, offering little evidence so far that AI is weakening overall employment.

• AI adoption has been slower than expected. Implementation across sectors, particularly at large firms, takes time, which has tempered the near-term labour market impact.

• Displacement is arriving through hiring, not firing. The early effects of AI are concentrated in entry-level and support roles and are showing up as positions that are quietly not replaced, which is largely invisible in layoff and employment statistics.

• Younger workers are where the signal is clearest. Recent employment softness has been more pronounced among workers aged 15 to 24, and particularly among new graduates, an area we are watching closely.

Slow AI rollout

AI has arrived in a labour market already undergoing structural change from an aging population. But the pace at which sectors, especially large enterprises, can adopt AI has proven slower than anticipated, and this lag between innovation and implementation has dampened its impact on employment so far.

Manufacturing constraints

In manufacturing, AI-driven automation has progressed slowly due to high costs and implementation challenges. For many companies, replacing labour with automated systems requires significant capital investment and, in some cases, the redesign of entire assembly lines, limiting the pace of adoption.

Non-manufacturing integration challenges

AI adoption across non-manufacturing sectors has also been slower than expected at large organizations. Companies need time to approve, implement, and integrate these tools, and the process is far more complex than simply replacing a worker with technology. As a result, there is little evidence so far of widespread layoffs or job destruction. Most occupations still require human judgement and accountability, limiting the extent to which AI can fully substitute for workers.

A closer look at the North American labour market

Nor is there evidence that AI is depressing employment rates. Since the release of ChatGPT in November 2022, the employment rate has declined by 1.3 percentage points in Canada and 0.8 percentage points in the United States. In November 2022, employment rates were 62.1%[1] (CAN) and 59.9%[2] (U.S.). As of August 2026, they stand at 60.8%[3] in Canada and 59.1%[4] in the U.S. Much of Canada’s decline reflects rapid population growth and an aging workforce rather than job losses, and total employment in both countries is higher today than it was in late 2022. With recent headlines painting an ominous picture, it’s important to focus on the facts. On a month-over-month basis, the employment rate edged down by 0.1 percentage points in Canada and increased in the U.S., while unemployment rates were unchanged.

Job displacement

Monthly sectoral employment changes, on their own, are insufficient to determine whether AI is driving significant job displacement. However, recent employment numbers continue to challenge claims of large-scale AI job replacement. The broader evidence since 2022 suggests that employment has generally continued to rise across sectors, even among occupations exposed to AI.[5] A recent study found that workers in occupations highly exposed to AI tend to have a stronger ability to adapt during a job transition.[6]

What the data cannot see yet

Here is where we think the debate has been framed too narrowly. The absence of layoffs is not the same as the absence of displacement. In practice, AI is entering the workplace not by eliminating existing roles but by changing what gets hired for. A firm that once needed two junior analysts finds it can manage with one. Design, research, and drafting work that used to be sent to a support department is increasingly done in-house with AI tools. A position that opens through natural attrition is simply not backfilled. None of this registers as a job loss, and none of it appears in the layoff statistics. It shows up, slowly, as a labour market that stops creating certain kinds of jobs.

This kind of substitution is happening fastest in smaller and mid-sized organizations, where adopting a new tool does not require a procurement process or an enterprise integration, and it is precisely the kind of change that national surveys are slowest to capture. Aggregate employment data will be the last place this becomes visible, not the first. That is why we read the stability of the headline numbers as evidence about the pace of change, not about its ultimate scale.

Youth employment

This also explains why the clearest signal so far is among young workers and recent graduates. If displacement arrives through positions not created rather than positions cut, the people who feel it first are those trying to get in the door. Youth unemployment always runs above the core-aged rate, but August data show the gap has widened, and entry-level roles are where AI-related disruption is most likely to appear first. More telling is a reversal in the historical relationship between college-educated and non-college-educated youth unemployment, which is consistent with AI disproportionately affecting the white-collar occupations typically sought by new graduates rather than occupations such as the trades.[7] We cannot attribute all of this to AI, but it fits the mechanism we describe above, and it is the area we are watching most closely.

What it means for portfolios

For investors, the slower-than-expected pace of enterprise adoption matters. It suggests the AI capital expenditure cycle is likely to be longer and less front-loaded than the most optimistic forecasts assume, and that the productivity payoff will accrue over years rather than quarters - a profile that suits patient, long-term investors. It also broadens the set of potential beneficiaries beyond the infrastructure providers to the companies that ultimately put these tools to work, particularly those with the labour-intensive cost structures that stand to gain most from doing more with fewer people. Both are relevant to how we think about the theme within our long-term allocation.

Bottom line

We often overlook the resilience of the labour market and its historical ability to evolve and adjust. Simply put, the current data does not support a near-term “jobs apocalypse,” and we would caution against reading every soft employment print as proof that one has begun. But we want to be equally clear about the longer term. We expect AI to displace a meaningful number of roles over the coming decade, concentrated in entry-level, support, and routine analytical work, and we expect much of that displacement to happen quietly, through hiring decisions rather than layoffs. Roles built on trust, judgement, accountability, and relationships will be far more durable. Many others will not be.

Consider the Jevons paradox: when technology makes something more efficient, demand for it often increases rather than falls. We have seen this before: spreadsheets did not eliminate accountants, they multiplied the analysis accountants could do, and in the 19th century more efficient steam engines increased rather than reduced coal demand. AI may do something similar by making certain tasks abundant, while shifting human value toward what remains scarce. That is a reason to expect total employment to hold up better than the pessimists fear. It is not a reason to expect every job, or every career path, to survive the transition intact.

We are still in the early stages of this evolution. Labour dynamics and the risk of AI-related disruption will require continued monitoring, and we will keep sharing what we see.

Dean Colling 

Rachel Campbell-Johnson