OpenAI just published the most important workforce study of the year — and enterprise leaders are still sleeping on it. The headline number: 44% of occupation-specific AI use already crosses job boundaries. That's not a prediction about the future. It's happening right now inside your organization.
The research, titled Work at the Frontier: How AI is Expanding What People Do at Work, analyzed more than 800,000 work-related messages from U.S. ChatGPT Business users. It's the first study of its kind to use actual AI usage data — not surveys, not simulations — to measure how workers are already redefining their roles in real time.
The finding should change how every C-suite leader thinks about org design, hiring, and workforce strategy for the next three years.
The Number That Changes Everything
Here's what 44% actually means.
OpenAI's researchers separated work-related ChatGPT messages into two categories: generic tasks (writing, summarizing, scheduling — things everyone does regardless of job title) and occupation-specific tasks (work that's distinctly associated with a particular role). For that second category, nearly half of all messages were about tasks that fall outside the user's primary occupation.
In plain English: when a salesperson uses ChatGPT for something specific to their job, nearly half the time they're doing work that traditionally belonged to someone else entirely.
Zoom out to all work-related messages — including the generic ones — and the number is 16.8%. That means one in six work-related conversations with ChatGPT involves a worker picking up a task that's outside their job description.
This is task crossover at scale. And it's happening without waiting for HR to rewrite a single job description.
Who's Doing Whose Job?
The crossover isn't evenly distributed. Some roles are doing far more boundary-crossing than others, and the breakdown reveals something important about where AI is most disruptive.
Customer experience workers lead the crossover rankings: 77% of their occupation-specific ChatGPT use involves tasks associated with other jobs. Your support team is becoming part analyst, part marketer, part operations coordinator — inside a single AI conversation.
Designers aren't far behind at 75%. Creative teams are using AI to expand into project management, copy, financial modeling, and technical troubleshooting. A designer who once worked in isolation is now doing the work of three functions.
HR professionals show 69% task crossover. The implications here are significant: people teams are using AI to pick up legal analysis, communications work, data analysis, and strategic planning tasks that previously required escalation or outside expertise.
Legal workers (56%) and marketers (53%) round out the top five. Legal teams are crossing into financial analysis and compliance work. Marketers are borrowing from engineering, sales, and analytics to become self-sufficient operations within the business.
The pattern is consistent: roles that historically required handoffs to other functions are now handling more of that work themselves, using AI as the capability bridge.
The Tasks That Travel Farthest
Not every type of work crosses job lines equally. OpenAI's data reveals two specific tasks that show up across every single occupation in the study: financial calculation and technology troubleshooting.
Think about that. Whether someone is in customer experience, design, legal, HR, or sales — they're all using ChatGPT to do some form of financial math and some form of tech problem-solving. These used to be the exclusive domains of finance teams and IT departments.
Beyond those universal tasks, marketing work travels the widest. Marketing tasks account for 8.9% of messages among workers in other fields — the highest outward share in the entire dataset. At the same time, marketers themselves spend 24.3% of their AI conversations doing work from other functions. Marketing both exports and imports more work than any other role.
Engineering tells a different story. Engineering tasks account for 7.4% of messages from workers in other fields — meaning a lot of people are doing engineering-adjacent work with AI. But only 18.5% of engineers' own AI conversations involve outside tasks. Engineers are more focused. They're less likely to stray into other domains, but everyone else is straying into theirs.
For enterprise leaders: if you haven't thought about what happens when every department can troubleshoot your systems and every team can analyze their own financial data, that conversation is overdue.
Small Business vs. Enterprise: A Revealing Gap
One of the more counterintuitive findings in the research is the relationship between company size and task crossover. You might expect larger enterprises — with more sophisticated AI tooling and dedicated AI teams — to show higher crossover rates. The data says the opposite.
Among moderate ChatGPT users, outside-occupation task rates fall as company size increases: 18.9% in organizations with 2–5 seats drops to 16.3% in organizations with more than 100 seats.
OpenAI's explanation is straightforward: in smaller organizations, the worker closest to a problem is more likely to solve it rather than hand it off. AI amplifies that tendency. When you don't have a dedicated analyst, a specialist lawyer, or a development team to delegate to, you pick up the tool and figure it out yourself.
This has an important implication for enterprise strategy. Large companies have built their workflows around functional specialization and escalation paths. AI is eroding that architecture from the inside. The question isn't whether your people are crossing functional lines — they are. The question is whether you've built systems and policies to handle it.
Why This Matters More Than Your Last Org Chart Review
Talking to peers across industries, I keep hearing the same assumption: AI is making people better at their existing jobs. Faster, more productive, higher quality within their current role. The OpenAI data challenges that assumption directly.
AI isn't just making people more efficient within their lanes. It's moving them across lanes entirely.
A customer experience manager handling analytical work that used to go to an ops team. A sales rep building a customer segmentation model that used to wait for a data science request. A legal professional running their own financial modeling rather than waiting two weeks for a finance review. These aren't edge cases. They're the majority of occupation-specific AI use.
This creates three problems that enterprise leaders need to think through:
Accountability gaps. When a non-expert uses AI to do expert work, who owns the output? If a salesperson generates a financial model using ChatGPT and uses it to set a customer's pricing, and that model has an error — who is accountable? The existing governance frameworks in most companies weren't designed for this.
Hidden skill obsolescence. When AI enables task crossover at scale, entire specialized roles can quietly become redundant — not because those skills disappeared, but because the need for dedicated headcount to perform them did. This creates real workforce planning problems that won't show up in productivity metrics until it's too late to manage gracefully.
Governance of expertise. There's a meaningful difference between an HR professional using AI to draft a legal summary (fine) and that summary being treated as authoritative legal opinion (not fine). The line between "assisted work" and "accountable work" is blurry. Most enterprise AI policies don't address it.
3 Moves Enterprise Leaders Need to Make
The research is clear on what's happening. Less clear on what to do about it. Here are the three moves that matter most.
Move 1: Map the crossover happening in your organization right now. You can't manage what you can't see. Start by auditing ChatGPT Business usage logs for your top functional teams. Where are people spending AI time outside their role? That map tells you where accountability gaps already exist and where informal role expansion is happening below the radar. In conversations with operations leaders who have done this exercise, the results are almost always surprising.
Move 2: Redesign roles around AI-enabled scope, not historical job descriptions. Job descriptions written three years ago describe a world where access to expertise was constrained by headcount. AI removed that constraint. A support manager in your organization may now realistically own work that previously required three specialized hires. If you're still hiring and evaluating performance against the old job architecture, you're optimizing for a workflow that no longer exists.
Move 3: Build explicit governance for cross-functional AI output. This is the most urgent and most neglected piece. You need clear policies that define: which types of AI-assisted cross-functional work require review by a domain expert, which types of outputs are advisory vs. authoritative, and what the escalation path looks like when AI-assisted work has material business consequences. Without this, you're managing by accident.
The Signal in the Data
What makes the OpenAI study genuinely important is what it measures: actual behavior, at scale, in real enterprise environments. Not what workers say they do. Not what models can theoretically do. What 800,000 people actually did when given access to AI at work.
The signal is unambiguous. AI is not staying neatly inside job descriptions. Workers are using it to expand their capability into adjacent functions — sometimes deliberately, sometimes because the need is in front of them and the tool makes it possible.
The shift is already underway. The only question left for enterprise leaders is whether they're going to manage it proactively — redesigning roles, building governance, mapping accountability — or discover what happened to their org structure after the fact.
One CFO I spoke with recently put it simply: "I expected AI to make my team faster at finance. I didn't expect them to stop sending things to legal and IT. Now I have to figure out if that's good or bad."
That question has an answer. But finding it requires looking at the actual usage data — and then making deliberate choices about the organization you want to build on top of it.
OpenAI's full "Work at the Frontier" report is available at openai.com. The methodology analyzes over 800,000 work-related messages from U.S. ChatGPT Business users matched to self-reported occupations.
Follow THE DAILY BRIEF for twice-weekly enterprise AI insights:
