The debate about AI and jobs has been stuck in a loop for years — will it replace workers, or won’t it? OpenAI’s latest research flips that question entirely. According to new findings published by OpenAI, ChatGPT users aren’t just doing their existing jobs faster. They’re doing entirely different jobs — ones they were never hired to do. And that shift has enormous implications for how organizations think about hiring, training, and the value of expertise itself.
The Old Automation Story Got It Wrong
For the better part of a decade, the dominant narrative around AI at work was essentially a threat model: machines come in, tasks get automated, workers get displaced. That story made sense in the era of narrow automation — robotic arms on assembly lines, optical character recognition replacing data entry clerks, that kind of thing.
But large language models don’t work that way. They’re not purpose-built to do one thing. They’re generalists, and it turns out that’s exactly what workers needed.
OpenAI’s research, drawing on usage patterns and survey data from ChatGPT users across industries, found that people are routinely using AI to step outside the formal boundaries of their roles. A marketing manager writes a Python script to automate a reporting task. A nurse drafts a grant proposal for a community health initiative. A small business owner builds out a financial model they’d normally pay a consultant to handle. None of this was in their job description. All of it happened because AI lowered the skill barrier enough that they could actually attempt it.
This is sometimes called role expansion — and it’s the central finding of OpenAI’s new report. The question isn’t just “is AI replacing jobs?” It’s “what happens when AI lets people do jobs they were never trained for?”
What the Research Actually Shows
The OpenAI findings paint a specific picture of how this role expansion plays out in practice. A few things stand out.
Workers Are Crossing Traditional Skill Boundaries
The most striking pattern in the data is how frequently workers are using ChatGPT to perform tasks that would traditionally require specialized training. This isn’t about drafting emails faster. It’s about people doing meaningful work in domains adjacent — or sometimes completely unrelated — to their primary expertise.
- Technical tasks by non-technical workers: Writing code, building automations, and analyzing datasets were reported by workers in non-technical roles at a surprising rate.
- Communication and writing tasks by technical workers: Engineers and developers using AI to handle stakeholder memos, presentations, and documentation they’d previously outsourced to comms teams.
- Legal and compliance drafting: Operations and HR staff using AI to produce first-draft policy documents, contract summaries, and compliance checklists.
- Creative and design work: Non-designers generating visual briefs, brand copy, and content strategies that would previously have required agency support.
- Financial modeling and analysis: People without finance backgrounds running basic projections and scenario analyses with AI assistance.
The common thread is that AI is acting as a competency bridge — not making workers expert in these areas, but making them functional enough to get real work done.
Job Titles Are Starting to Lag Behind Reality
Here’s something that doesn’t get discussed enough: organizational structures are built around the assumption that people have defined, bounded skill sets. You hire a designer to do design. You hire an analyst to do analysis. Job descriptions, salary bands, and team structures all flow from that assumption.
OpenAI’s research suggests that assumption is eroding. When a single employee can credibly produce work across five or six traditional job functions — even imperfectly — the neat organizational chart starts to look like a relic. I wouldn’t be surprised if we start seeing HR departments seriously wrestling with how to evaluate and compensate workers who are effectively doing multi-role work, especially in smaller companies where headcount is tight and AI makes that kind of versatility genuinely achievable.
The Quality Question Remains Open
None of this means AI-assisted cross-functional work is equivalent to work done by a trained specialist. A nurse writing a grant proposal with ChatGPT isn’t the same as a professional grant writer. A developer-turned-marketer isn’t going to out-strategize a seasoned CMO. The research doesn’t claim otherwise.
But “good enough” is a real category, and in many organizational contexts, good enough — done faster and cheaper — wins. That’s the uncomfortable trade-off sitting underneath all of this. As NTT DATA demonstrated with OpenAI Codex, even partial AI assistance on complex technical tasks can produce dramatic efficiency gains without requiring expert-level human input at every step.
Why This Matters More Than the Replacement Debate
The replacement narrative, whatever its merits, focuses on a relatively clear economic question: does AI reduce headcount? The expansion narrative is murkier, and arguably more important.
If AI is expanding what individual workers can do, then the productivity gains don’t just show up as labor cost reductions — they show up as capability increases at the individual level. A team of ten people can now produce output that previously required fifteen, not because five people lost their jobs, but because each of the ten is operating with a wider functional range.
This has a few second-order effects worth thinking through:
Smaller Teams, Broader Scope
Startups and small businesses are likely the first to feel this clearly. When a three-person team can cover legal, finance, marketing, and ops with AI assistance, the case for hiring dedicated specialists weakens — at least in the early stages. This is already visible in how AI-native startups are structuring themselves, with lean founding teams taking on functions that would have demanded early hires two or three years ago. OpenAI has been pushing directly into this space, and ChatGPT’s small business offerings are clearly designed with exactly this use case in mind.
The Mid-Level Specialist Crunch
If generalists with AI can cover a lot of mid-level specialist work adequately, the market for those mid-level roles could soften. This isn’t entry-level displacement (those workers often lack the domain context to effectively direct AI) and it’s not senior specialist displacement (deep expertise still commands premium). It’s the layer in between — the competent generalist specialist — that faces the most direct pressure from this kind of capability expansion.
New Expectations on Workers
There’s a flip side to all this expanded capability: employers are going to start expecting it. Once it becomes normalized that a marketing manager can produce basic code or that an operations analyst can draft legal summaries, the bar for what a “complete” employee looks like shifts upward. That’s not necessarily bad, but it does mean that workers who don’t engage with AI tools will increasingly look limited by comparison — not because AI replaced their job, but because it expanded everyone else’s.
This dynamic connects to broader questions about AI literacy and access. OpenAI has been vocal about wanting ChatGPT to function like a knowledgeable colleague available to everyone, and the enterprise rollout of tools like OpenAI Presence suggests they’re building infrastructure for exactly that kind of always-on workplace AI integration.
What This Means for Different Audiences
The practical implications of this research cut differently depending on where you sit.
For individual workers: The workers who benefit most are those who actively use AI to stretch into adjacent functions rather than waiting to be told to. Curiosity and initiative matter more now. If you’ve been avoiding AI tools because your core job doesn’t require them, you’re probably underestimating how much they could expand your apparent value.
For managers and team leads: Think carefully about how you’re structuring work. The traditional assumption that tasks map cleanly to roles is weakening. Some of the best productivity gains from AI won’t come from automating existing tasks — they’ll come from discovering that certain roles can absorb new responsibilities that previously required separate headcount.
For HR and talent teams: Job descriptions need to catch up. Competency frameworks built around narrow role definitions will increasingly fail to capture what high-performing AI-assisted employees actually do. This is a harder organizational design problem than it looks.
For businesses making hiring decisions: The calculation on when to hire a specialist versus when to extend an existing team member’s scope with AI has genuinely changed. That doesn’t mean specialists are obsolete — deep expertise matters more than ever at senior levels — but it does mean the threshold question for a new hire deserves more scrutiny than it used to.
Frequently Asked Questions
What exactly did OpenAI’s research find about AI and work?
OpenAI found that ChatGPT users are regularly performing tasks well outside their formal job roles — things like writing code, drafting legal documents, building financial models, and producing creative work. The research frames this as “role expansion,” meaning AI is growing what workers do rather than simply replacing what they already do.
Does this mean AI is not replacing jobs?
The research doesn’t settle that debate definitively. What it shows is that replacement isn’t the only — or even primary — dynamic at play right now. Expansion of individual worker capability is a distinct and significant effect. Both things can be true simultaneously, and the long-term balance between them is still playing out.
Which types of workers benefit most from this role expansion?
The research suggests workers who are already competent in their primary domain and are willing to use AI to stretch into adjacent areas benefit most. People with strong judgment but limited technical or specialist skills — like experienced managers and operators — seem particularly well positioned to use AI as a capability bridge.
How does this compare to what other AI companies are doing?
OpenAI isn’t alone in positioning AI as a workplace capability expander. Anthropic’s Claude models, including the recently released Claude Opus 5, are similarly designed for complex multi-domain tasks. The difference is that OpenAI has now produced specific research quantifying how this plays out in actual workplace behavior — which is useful both as a product narrative and as genuine evidence for enterprise buyers evaluating AI investments.
What OpenAI has done here is give the “AI as colleague” argument some empirical grounding, and the timing feels deliberate — enterprise AI spending decisions are intensifying, and this kind of research positions ChatGPT not just as a productivity tool but as something that changes the fundamental economics of what a team can accomplish. Whether organizations move fast enough to actually restructure around that new reality is the harder question, and one the research doesn’t answer.
