On July 27, 2026, OpenAI published the first study in its Work at the Frontier series, examining how AI helps people take on work traditionally associated with other occupations. The analysis covers more than 800,000 messages from U.S. ChatGPT users. OpenAI calls the pattern task crossover: work associated with one occupation appearing in another occupation's AI use.
The page highlights two figures. It says 16.8% of work-related messages involve cross-occupation tasks, while 43.5% of messages classified as occupation-specific concern a task associated with another occupation. These are findings from OpenAI's analysis of its own usage data, not universal estimates for every worker, region, or labor market.
OpenAI uses practical examples to describe the shift. A small-business owner may draft copy, review a contract, or perform basic financial analysis. A salesperson may explore customer data that once would have gone to an analyst. A marketer may troubleshoot a website before waiting for a developer. The examples show lower access barriers; they do not establish that the resulting work has passed professional review.
The report also points to more task crossover in small businesses. Smaller teams may have fewer specialized roles and require individuals to cover more functions, while AI can lower the cost of getting help with analysis, writing, or technical work. Turning that observation into a productivity claim would require separate measurement of output quality, rework, error costs, and accountability.
The workflow impact is more precise than a simple question about which jobs will disappear. A process may move from fixed occupational handoffs to one person using AI to explore, organize, or draft, followed by review from the relevant professional. That design still needs permissions, source handling, review thresholds, and an audit trail; access to a tool is not a control system.
The methodological limits matter. The sample comes from U.S. ChatGPT users, and the classification of occupations and cross-occupation tasks depends on the study method. It does not directly measure production, wages, employment changes, or company profit. OpenAI presents the results as directional evidence about work reorganizing in real time, so organizations should test the claims against their own workloads and risks.
The main signal from Work at the Frontier is that AI is changing who can begin a task, not only how quickly an existing role can complete it. For managers, the next question is not just which model to buy. It is who may execute each step, which data may be used, and when a qualified person must review the result.
