Future of work

A study by Anthropic titled “Which Economic Tasks Are Performed with AI? Evidence from Millions of Claude Conversations” offers one of the first empirical looks at how people are using AI across occupations.  

Drawing on more than four million anonymized interactions with Claude, the authors mapped real user prompts to O*NET’s catalogue of work tasks to quantify which kinds of work activities are being supported or automated by AI in practice. The analysis spans roughly 20,000 occupational tasks and establishes a detailed picture of AI’s current footprint across work that requires thinking. 

The results show that AI use is highly concentrated rather than evenly distributed. Nearly half of all observed use cases involve software development and writing tasks, with limited evidence of adoption in roles that require physical manipulation or direct, interpersonal labour. Roughly one-third of occupations show AI use in at least a quarter of their tasks, while only a small minority exhibit near-universal adoption. This pattern supports the view that AI is entering the economy through narrow, high-use channels rather than driving broad occupational replacement. 

The study also finds that AI exposure peaks in mid- to high-wage occupations that rely on analytical and communication skills. These are tasks where language models perform well and where the boundary between automation and human collaboration is still fluid.  

About 57% of the observed usage reflects support for tasks rather than full automation of them, suggesting that workers are incorporating AI as a complementary tool rather than being displaced by it. In this sense, AI’s economic role is additive—it is accelerating specific forms of cognitive work instead of eliminating them. 

For Canada’s labour-market analysts, these findings underline the need to track AI’s influence at the task level rather than through job counts alone. Measuring these shifts requires access to a new data infrastructure that can link digital tool use to occupational task frameworks.  

 The key challenge now is to prepare workers to collaborate effectively with AI systems—that is, to develop the adaptive, human-in-the-loop skills that will keep pace with how technology is actually being used on the ground.