AI Work Watch
Monitoring AI’s impact on Canada’s economy
AI Work Watch tracks how AI is actually being used at work, how jobs are changing, and what the real numbers show to distinguish credible data from guesswork.
DATA LAST UPDATED: 01/31/26 AT 3PM
Sorting signal from noise
Artificial intelligence is advancing faster than our understanding of its economic consequences. Without trusted ways to measure change, governments and industry are making decisions in the dark.
By testing public claims against hard economic evidence, we deliver clear, timely analysis that gives policymakers a reliable basis to act.
Applying a unique measurement framework developed by economists at the Dais, and inspired by the Budget Lab at Yale, AI Work Watch explores a series of questions to analyze the impact of AI on people at work, Canada’s labour market overall, and the economy as a whole.
In Partnership With
“Will robots replace us? Industry giants say yes.”
The New York Times “‘That field is dead’: Bracing for AI disruption, white-collar workers are pre-emptively changing careers”
Globe and Mail “AI is stealing entry-level jobs from university graduates”
The Logic “AI is forcing this Ontario teen to abandon his career dreams: ‘I want to bet on reality’”
Toronto Star How many Canadians are using Gen AI?
Source: Use of generative artificial intelligence tools among Canadian workers, March 2026
Is AI changing the job market faster than other technologies?
The Canadian economy is made up of 20 million people working across many different occupations.
Using an approach developed by Martha Gimble at Yale’s Budget Lab, we can measure changes in the Canadian job market through employment trends – for example, if people are working less in one sector and more in another over time.
When we benchmark today’s labour market shifts against the shifts brought by other disruptive technologies, like the introduction of computers and the internet, the composition of Canada’s workforce is not changing at an unusual pace.
When does change become disruption?
The labour market is always changing, and policymakers manage regular fluctuations using existing tools.
Often, those changes are relatively slow (like the gradual impact of the internet); isolated, impacting specific sectors of the economy (like Canada’s decline in manufacturing); or is a one-time shock (like the 2008 financial crisis).
AI-driven economic disruption would stem from changes that are fast, widespread, and persistent.
Examples of Change
The gradual spread of the internet.
How we responded: infrastructure investment and literacy programs
The decline of Canada’s auto manufacturing sector.
How we responded: Sectoral job transition support
The 2008 financial crisis
How we responded: Expansion of Employment Insurance.
Is AI reducing the demand for jobs where workers encounter it the most?
Where workers encounter gen AI the most (group 1), demand for these jobs has not fallen. Where they’re less likely to encounter AI (group 5), demand has also not fallen.
Researchers have developed two fairly sophisticated measures to predict whether an occupation is more or less likely to encounter AI in their work. For example, highly AI exposed jobs include customer service representatives and payroll administrators; while service technicians, plumbers, and dental hygienists are low-exposure jobs. So far, these predictive measures have lined up with how workers are actually using AI in their jobs.
We have adapted these measures for Canada’s labour market. Using employment levels from November 2022—the public launch of ChatGPT—as a baseline, this graph illustrates the changes in employment across five groups (where group 1 is the most exposed to AI, and group 5 is the least exposed to AI).
If we were experiencing an AI-driven job-pocalypse, we would expect that the jobs that are most exposed to AI show rapid declines in employment.
- Where workers encounter gen AI the most, (group 1), demand for these jobs has not fallen.
- Where they’re less likely to encounter AI, (group 5), demand has also not fallen.
- Q1 (Least exposed): Trades: Carpenters, Electricians, Plumbers
- Q2 (less exposed): Therapists, Machine Operators
- Q3 (median exposure): Sales Representatives
- Q4 (more exposed): Computer Programmers
- Q5 (most exposed): Professional Occupations in Finance (Accountants)
Are younger workers disproportionately affected?
Where age comes into play, are young workers getting shortchanged? Use this tool to see how hiring is changing when workers’ age is factored in.
Has AI changed the nature of work, overall, in Canada?
We don’t yet have a clear answer. Many factors contribute to the nature of work, including shifts in technology, culture, legislation, and new standards.
Measuring this is tough, because many long term effects of AI just haven’t come to fruition yet, and it takes well designed experiments and studies to conclude AI causes a specific change beyond correlation.
WHAT WE DO KNOW
Those who are most exposed to AI tend to have the least precarious work
Some early results show it may close the initial skill gap between junior employees and senior employees
Consistent reliance on AI may make it more difficult for workers to attain depth in knowledge
Evaluating the evidence
There’s a lot of chatter on AI and the economy – and some voices carry more weight than others. How do we sort out good evidence from the hype?
WE PRIORITIZE:
High-Quality Data
Prioritize research built on well-established, transparently sampled, and scrutinized data sources, such as Statistics Canada, official national statistics, or credible private sources.
Modest and Precise Claims
Rigorous economic research rarely makes sweeping, sensationalist predictions. High-quality findings offer specific, measured insights, recognizing that even a 5–10% shift in a labor market represents significant structural change.
Separation of Diagnosis from Prescription
Be cautious of studies that simultaneously diagnose complex labor trends and promote a specific policy or commercial product. Rigorous empirical research focuses on accurately measuring what is happening before pushing specific solutions.
Examples of High Quality Evidence
| High-Quality Data | What questions it answers and what it says | What does it not say | Why we like it |
|---|---|---|---|
|
Canadian employment trends in the era of generative artificial intelligence: Early evidence |
Whether the labour market has changed since the introduction of ChatGPT. It concludes that broadly, the labour market has not changed. If anything, demand for highly AI-exposed jobs, like software developers, has increased |
Showing that the changes (or lack thereof) is caused by AI. How extensive AI use in Canada is. |
It uses well validated data that covers the whole Canadian economy, well established AI exposure measures, the researchers are well known, and the method they don’t overclaim. |
|
Use of generative artificial intelligence tools among Canadian workers, March 2026 |
How many Canadian workers used AI in their work in the past 12 months? It concludes that between 2024 and 2025, 41.6% of Canadian workers used an AI tool to perform their work in the preceding 12 months |
AI use in 2026, the extent of the AI use, and if it had any impact on work. |
It is a representative survey conducted by Statistics Canada (highest quality data) that allows for benchmarking AI usage against other sources. |
|
What happens to productivity when customer service agents are given assistive chatbots to help draft responses to customer inquiries? It increased the number of issues resolved per hour by 34% for novices, while it didn’t impact senior agents’ productivity. |
Whether an improvement leads to employee upskilling (so when an AI tool is taken away from novice workers, their productivity doesn’t decrease). |
A fairly well-designed randomized experiment that tracks both raw output (issues resolved) and quality metrics (customer satisfaction) that can identify causality, not just correlation. |
|
|
Writing Code vs Shipping Code: Productivity Effects Across Generations of AI Coding Tools |
What happens to coding productivity across a range of measures when new AI tools are introduced? It increases the amount of code (by 958%) but of those lines, only 20% make it into final releases. It also found that apps created with this code often see low engagement. |
It is not an experiment (but still uses causal methods to determine causality.) It also doesn’t answer some specific quality measures: for example, if the AI-generated code was buggy, or had vulnerabilities compared to human-written code. |
Just like with the Generative AI at Work study, it tracks both raw output (lines of code), and other quality/ success metrics (actual app releases and engagement.) It’s not an experiment, but it makes a convincing “event study” argument which allows you to identify causality. |
