OpenAI's new research shows AI is blending job roles. Has your skills strategy caught up?

What OpenAI found

OpenAI's Work at the Frontier report analysed more than 800,000 work-related messages from ChatGPT Business users across eight occupation groups: customer experience, design, engineering, finance, HR, legal, marketing, and sales.

The headline finding: 43.5% of occupation-specific AI use is cross-occupation. People are using AI to do tasks that have historically belonged to someone else's job.

The cross-occupation share varies a lot by function. Customer experience leads at 77%, followed by design at 75% and HR at 69%. Legal comes in at 56% and marketing at 53%. Engineering is much lower at 18.5%.

The pattern also varies by direction. Some occupations are borrowers: HR and design draw heavily on tasks from other fields. Others are exporters: marketing and engineering tasks travel across nearly every other function. Marketing is both. Marketers take on external work, and marketing tasks also appear widely in other roles.

‍

The HR number

For HR and L&D leaders, the number that stands out to me is 69%, the share of HR workers' occupation-specific AI use that is cross-occupation.

The function that designs, approves, and funds skills programmes is itself using AI heavily outside traditional HR work. HR professionals use it for financial analysis, legal research, marketing content, and engineering troubleshooting, all at significant rates.

I read that two ways. Either HR is becoming much more of a generalist function, or the gap between what HR professionals are trained for and what they now do with AI has widened without anyone noticing.

Either way, most skills strategies have no answer yet to the question this raises. If the people designing training programmes are themselves doing work outside their traditional expertise with AI, what does that mean for how those programmes are designed?

‍

Why job descriptions are becoming unreliable

Most leaders have sensed this for a while without the data to prove it. Job titles and role descriptions have always lagged behind the work people actually do, and this report shows AI widening that gap. Hiring plans and capability budgets both lean on those descriptions.

A salesperson now explores customer datasets that would previously have gone to an analyst. A marketer troubleshoots code that would previously have gone to a developer. A finance professional drafts internal communications that would previously have sat with corporate affairs. In each case the task changed and the job title stayed the same.

OpenAI draws its own conclusion: people may need training to assess AI-assisted work outside their own field, and organisations will need "clear processes for review and accountability."

That gives HR and business leaders a clear job to do. People need to learn how to judge output in domains where they are not specialists: how to tell when an AI-generated financial analysis is trustworthy, when a piece of AI-drafted legal language is safe to use, and when a technical solution holds up. This is the judgment layer, and most AI upskilling programmes do not build it.

What this means for skills strategy

Here are the three implications I'd draw for how organisations invest in capability:

• Job titles are no longer a reliable input to skills gap analysis. If the work people do diverges from their job description, an assessment built on that description misses a large share of the real capability gap. The analysis has to follow the actual work, not the org chart.‍
• The generalist layer is becoming the highest-leverage investment. Marketing and engineering tasks are showing up across seven of eight occupation groups. The implication is not that every employee needs to become a marketer or engineer. It is that every employee increasingly needs enough cross-functional literacy to use AI well in those adjacent domains and enough judgment to know when the output needs a specialist to review it‍
• Smaller organisations face a sharper version of the challenge. The OpenAI data shows cross-occupation AI use of 18.9% at companies with 2-5 seats against 16.3% at companies with 101+ seats, among typical-volume users. Without specialists on staff, employees reach for AI as the substitute. That is exactly where the stakes for getting it right are highest, and where formal review is thinnest.

‍

Three places to start

If you're setting next year's priorities, here's where I'd start:

• Audit how your teams are actually using AI: The OpenAI data came from real messages, not a survey. Most organisations still rely on job titles and guesswork instead. Find out which tasks your people are already handing to AI before you decide where to invest.
• Design cross-functional AI literacy into programs: Train people to evaluate AI output in adjacent domains, not just their own. The question is not "can you prompt effectively?" but "can you assess whether this output is trustworthy when you are not the expert?"‍
• Build accountability processes into AI rollouts: Define when AI-generated work needs specialist review, and write it down instead of leaving it to individual judgment. Organisations that pair training with clear review processes will outperform those that rely on training alone.