Future of work

Key takeaways 

• Occupations with low precarity—that is, those that tend to be comparatively stable and secure—had greater exposure to LLMs than occupations with medium, high, or very high precarity. 

• LLM exposure was lower in occupations with higher levels of temporary employment, irregular hours, and involuntary part-time work. 

• The current wave of automation may be different from earlier waves, which focused more on streamlining routine, low-quality, precarious tasks. 

• The results may suggest an increase in job-security risk for workers in higher-quality jobs (those with more stable employment and better working conditions). They may also suggest that workers in lower-quality occupations could benefit from the productivity gains associated with LLMs. 

Plain language summary 

This study examined how exposure to large language models (LLMs), such as AI systems that can process and generate text, code, or images, relates to precarious work (characterized by poor employment conditions) in Canada.  

The authors used Labour Force Survey data from 2021 to 2024 to analyze results at the occupational level rather than at the level of individual workers. They created a combined index to classify occupations as having low, medium, high, or very high exposure to precarity. They measured the precarity of work using four dimensions:  

• temporary employment 

• low wages 

• irregular hours 

• involuntary part-time work 

They measured LLM exposure using a task-based approach that estimated how much of an occupation’s work could potentially be affected by LLMs. 

The findings show that occupations with the lowest levels of precarity tend to have the highest exposure to LLMs. In other words, jobs with more stable employment conditions are more likely to include tasks that LLMs could potentially perform or support. This differs from some earlier waves of technological change, where automation was often associated with more routine or lower-quality jobs. 

The study does not show whether LLMs are currently being used in these occupations, whether workers are benefiting from them, or whether jobs are being displaced. Instead, it shows that the potential exposure to LLMs is higher in occupations that have historically had better employment conditions. The authors argue that more research is needed to understand how LLMs affect job quality, worker health, and labour market inequality. 

Definitions 

Large language model (LLM): A type of AI system that can process and generate language, code, or images. These systems can be used for tasks such as summarizing information, drafting text, supporting customer service, or analyzing documents.

LLM exposure: A measure of how much of an occupation’s tasks could potentially be affected by LLMs. In this study, exposure does not mean that workers are already using LLMs in their jobs.

Precarious work: Work characterized by poor employment conditions, such as temporary employment, low wages, unpredictable hours, or involuntary part-time work.

Involuntary part-time work: Part-time work among people who would prefer to work full-time hours.

Why the results matter 

For the public sector: 

The findings suggest that LLMs may affect occupations with relatively stable employment conditions rather than only occupations associated with higher precarity. This may be relevant for public sector organizations that are tracking how AI systems interact with job quality, workforce transitions, and labour market inequality. 

For employers and employment services: 

The study suggests that occupations with higher potential exposure to LLMs may not be the same occupations that are considered most precarious. Employers and employment services may need to consider how AI-related training, job redesign, and workforce supports vary across occupational groups. 

For post-secondary institutions and students:

The findings indicate that LLM exposure is more concentrated in occupations with lower levels of precarity, many of which may require post-secondary education or knowledge-based skills. This may be relevant for institutions considering how AI literacy, task redesign, and career preparation are incorporated across programs.

For educators, career development professionals, and job seekers: 

The study provides evidence that LLMs may be more relevant to some stable or higher-quality occupations than to more precarious ones. Career development professionals can use this distinction to help job seekers understand that AI exposure varies by occupation and does not necessarily mean the same thing as job loss or workplace adoption.

For people working with underrepresented and underserved groups: 

The study notes that employment conditions, AI exposure, and worker characteristics—such as age, gender, and education—may intersect in important ways. More research is needed to understand how different groups of workers experience LLM-related changes in job tasks, employment conditions, and access to potential productivity benefits.

For people in the labour market information ecosystem: 

The findings highlight the value of measuring AI exposure alongside job quality indicators. For LMI users, the study also shows the importance of distinguishing between potential task exposure, actual technology adoption, and observed labour market outcomes.

For most Canadians: 

The study suggests that LLMs may affect work in occupations that have traditionally enjoyed more stable employment conditions. However, the research does not show whether these effects will be positive or negative for workers. It points to the need for better evidence of how LLMs are changing job tasks, working conditions, and access to good-quality employment. 

Notes and important disclaimers 

• The study measures potential occupational exposure to LLMs, not actual LLM use in workplaces. 

• The analysis was conducted at the occupation level, so the findings should not be interpreted as describing the experience of every worker in a given occupation. 

• The study did not measure job displacement, wage changes, productivity gains, worker health outcomes, or long-term impacts. 

• Labour Force Survey data exclude some groups, including people living in the territories, people living on reserves and in other Indigenous settlements, full-time members of the Canadian Armed Forces, people living in institutions, self-employed workers, and people in extremely remote areas. 

• The LLM exposure measure is based on whether job tasks could potentially be performed more quickly through LLM use. It does not account for all forms of AI, robotics, or future changes in AI capabilities.