Where a Job's Boundary Lies: 16.8% of Work Requests to AI Concern Someone Else's Job

OpenAI analyzed more than 800,000 work-related ChatGPT messages. One in six concerns a task that used to belong to a different occupation, and in small companies the share is higher.
Every job comes with a familiar set of tasks. A marketer writes copy, an engineer fixes software, a lawyer reviews a contract. With AI that set has started to blur, and now the blur can be measured.
What was measured
In July 2026, OpenAI published its Work at the Frontier report. The researchers took more than 800,000 work-related messages from ChatGPT users in the US and matched each one to the sender's occupation. The messages fall into three groups.
| Message group | Share |
|---|---|
| Generic tasks: emails, scheduling, meetings | 61.5% |
| Tasks of the sender's own occupation | 21.8% |
| Tasks historically tied to another occupation | 16.8% |
Remove the generic tasks and the picture sharpens: among occupation-specific messages, 43.5% concern another occupation's work. Almost half of substantive work with AI crosses the boundary of the sender's own job.
Who takes on other people's tasks, and whose tasks get taken
Occupations behave differently. Some borrow heavily, others become the source.
- •Designers spend 35.2% of their messages on tasks from other occupations. Design tasks, meanwhile, make up only 1.7% of messages from people in other fields.
- •Engineers are the reverse: 18.5% of their messages reach outside their field, while engineering tasks account for 7.4% of messages from other workers.
- •Marketing works in both directions. Marketers spend 24.3% of their messages on other occupations' tasks, and marketing tasks account for 8.9% of messages from people in other occupations, the highest share in the sample.
The authors' conclusion: with AI, people most often take on marketing and technical work, the tasks that used to require a separate specialist.
In small companies the boundaries are thinner
Company size affects the result. In workspaces with 2-5 seats, the share of cross-occupation tasks is 18.9%; in workspaces with 101 or more seats it is 16.3%. The difference shows up among typical users with a moderate volume of messages. Among the most active users, size makes no difference.
The authors explain it this way. A large company has specialized teams and established workflows. In a small one, the person closest to the problem takes it on: with AI, an employee drafts marketing copy, troubleshoots software, reviews a contract or runs a basic analysis, because there is nobody to hand it to.
Borrowed tasks stick
In September 2026, a follow-up covered 1.5 million messages from April through July. The main result: crossing occupational boundaries recurs. People return to the outside tasks they have learned, and over time those tasks take up a growing share of their AI use. The authors put it carefully: the mix of tasks within a job may broaden while its title stays the same.
What to keep in mind about the data
The report describes requests; it does not measure the quality of the result. We see that a designer asked AI to write code, and we do not know whether that code worked. The sample consists of users of one service in one country.
What this means for a small company
Job descriptions lag behind what people already do. It is useful to compare role descriptions with actual tasks once a quarter and decide which of the new tasks the company is ready to assign to an employee formally.
The second takeaway concerns review. When a marketer prepares a contract with AI, someone with experience has to accept that contract. We wrote about why deep expertise becomes more valuable in this setting in our overview "AI and Narrow Specialization".
Related service
Express Financial Diagnostics
An independent look at your finances in 5 days: 3-5 findings and a plan for what to fix first
