AI at work: OpenAI study finds 43% of skilled use crosses job boundaries — what it means for your SMB

On July 27, 2026, OpenAI published Work at the Frontier: How AI is Expanding What People Do at Work, the first in a new research series on how AI reshapes daily tasks. The study analyzed more than 800,000 messages sent by US business users of ChatGPT: an AI model read each message, summarized the underlying task, and mapped it to an occupation in O*NET, the US Department of Labor’s occupational database. When the mapped task belonged to a different occupation than the one inferred for the user, OpenAI counted it as “task crossover.” The headline number: 16.8% of all work-related messages, and 43.5% of messages tied to a specific occupation, involve a task that formally belongs to someone else’s job.

Who crosses over the most

The rate varies sharply by profession. Customer experience workers lead, with 77% of their occupation-specific messages involving tasks from another field — followed by designers (75%), HR professionals (69%), legal staff (56%) and marketers (53%). Engineers sit at the other end of the scale, staying closest to their core job. OpenAI frames this as evidence that AI is not just making people faster at their own job — it is letting them competently take on tasks that used to require a different specialist entirely.

What the research already showed before this study

This crossover pattern lines up with an earlier, widely cited result. A 2023 study by Shakked Noy and Whitney Zhang published in Science, Experimental evidence on the productivity effects of generative artificial intelligence, had college-educated professionals complete realistic writing tasks with and without ChatGPT and found that the tool cut completion time by about 40% and raised output quality by about 18% on average — with the largest gains going to workers who scored lowest without AI assistance. The gap between a strong performer and a weaker one on a given task shrank substantially. Read together with OpenAI’s new data, the pattern is the same one at a larger scale: generative AI does not just speed up specialists, it narrows the skill distance between a specialist and someone doing that task for the first time.

What it means for your SMB

A small business already runs on informal task crossover — the ops manager who drafts customer replies, the salesperson who cobbles together a report, the assistant who edits a visual for a social post. What OpenAI’s data confirms is that this is happening with AI now on a very large scale, whether or not a company has decided to organize it. The real choice for an SMB is not whether employees will use AI to step outside their core role — the study suggests most already do, in customer service and HR more than anywhere else. The choice is whether that crossover runs through an unsupervised personal ChatGPT tab, with no company context and no review step, or through a structured agent built on your own processes and data — a customer support AI agent that a salesperson can lean on without becoming a support specialist, an email triage agent that routes requests correctly the first time, or an automated reporting agent that turns scattered figures into a document without whoever compiled it needing a finance background.

• →Map where crossover is already happening informally in your team — who is quietly doing marketing, reporting or support tasks outside their job title — before deciding where to invest in a structured agent.
• →Prioritize the functions the study shows cross over most: customer service and HR, not necessarily the technical team, which OpenAI’s data shows stays closest to its core tasks.
• →Keep a human review step on any AI-assisted output that leaves an employee’s normal domain — the productivity gain Noy and Zhang measured came with AI assistance, not with an unreviewed AI output standing in for expertise.

The question this data raises for an SMB owner is not whether job boundaries are blurring — OpenAI’s numbers say they already are, in nearly half of specialized AI use. It is whether that blending happens inside a system you can see and supervise, or invisibly, one browser tab at a time.