Work at the Frontier: How AI Is Expanding What People Do at Work
Citation. Chin, C., & Richmond, A. M. (2026, July). Work at the Frontier: How AI Is Expanding What People Do at Work. OpenAI Economic Research.
Summary. Chin and Richmond analyze more than 800,000 work-related ChatGPT messages from U.S. users whose self-reported roles were linked from ChatGPT Business accounts, mapping message tasks to O*NET detailed work activities. They find that 16.8% of all sampled work-related messages, and 43.5% of occupation-specific non-generic messages, concern tasks historically associated with another occupation. The report calls this pattern task crossover and treats it as early descriptive evidence that AI may broaden the task bundles workers attempt before job descriptions or occupational classifications catch up.
Key claims
- In the sampled messages, 16.8% were cross-occupation, 21.8% were within occupation, and 61.5% were generic; among non-generic occupation-specific messages, 43.5% were cross-occupation.
- Cross-occupation work was a majority of occupation-specific messages in five of the eight analyzed groups: customer experience, design, human resources, legal, and marketing.
- Marketing and engineering tasks appeared broadly in other occupations’ AI use, while workers in design, sales, and human resources more often used AI for tasks outside their typical roles.
- Among typical-volume users, task crossover was more common in smaller workspaces: 18.9% in 2–5-seat workspaces versus 16.3% in workspaces with 101 or more seats; the report does not establish why that pattern occurs.
- The evidence describes AI-aided task activity rather than delegation, completed work, productivity, employment, or durable role change, but it identifies occupational-boundary recomposition as a distinct pathway beyond fixed-task automation or augmentation.
Evidence & limitations
- The analysis is a descriptive random sample of more than 800,000 work-related messages from U.S. users with role information linked from ChatGPT Business accounts, classified hierarchically to O*NET activities; it covers eight occupation groups.
- The unit is a message, not an hour, project, job, or verified output. The study does not observe whether AI output was used, its quality, time saved, whether the user could complete the task unaided, or whether a specialist reviewed it.
- Role information is self-reported, occupational boundaries are O*NET-based historical baselines, and individual ChatGPT-account activity should not be generalized to ChatGPT Enterprise users or the U.S. workforce.
- This is an OpenAI Economic Research report, not peer-reviewed scholarship; the source organization has a direct strategic interest in interpretations of AI use and work change.
- Public provenance was verified against OpenAI’s public report page and CDN PDF; the public PDF exactly matches the ingested Drive binary SHA-256.