Most conversations about AI at work focus on the obvious stuff — writing emails faster, summarizing documents, generating first drafts. But OpenAI’s new Economic Research, published September 16, 2026, is asking a more interesting question: what are workers doing with AI that they were never really supposed to be doing? And more importantly, which of those behaviors are sticking? The findings paint a picture of a workforce quietly reshaping its own job descriptions, one AI session at a time.
The Research Nobody Else Was Running
For years, productivity research focused on efficiency — how much faster can someone do the same task? OpenAI’s team took a different angle. Their economic research into new ways of working looked specifically at workers who adopted AI and then started doing things that simply weren’t part of their original role. Think: a marketing coordinator who starts running competitive data analysis. A support agent who begins drafting product specs. A paralegal who starts doing rudimentary financial modeling.
This isn’t productivity improvement. This is scope expansion. And it’s happening organically, without anyone asking for it.
The timing makes sense. By mid-2026, ChatGPT has crossed one billion users, and the interface is mature enough that non-technical workers aren’t intimidated by it anymore. The friction is gone. When the tool is easy to reach for, people reach for it in unexpected directions.
What’s notable is that OpenAI is investing in economic research at all. This isn’t a product announcement — it’s a company trying to understand and document the actual labor market effects of its own technology. That’s either genuinely responsible, or very smart positioning ahead of regulatory scrutiny. Probably both.
What the Data Actually Shows
The research identifies a clear pattern: workers don’t just use AI to go faster inside their lane. They use it to cross lanes. And certain crossover activities become recurring — meaning they stop being experiments and start being part of the job.
Here’s how the behavior breaks down across the key findings:
- Task boundary expansion: Workers regularly use AI to take on tasks that would previously have required a specialist — writing SQL queries, producing financial projections, drafting legal language, or building presentation decks from scratch.
- Skill bridging: AI is acting as a real-time tutor, letting workers develop just enough competence in adjacent disciplines to execute tasks without formal training or hiring.
- Recurring novel activities: The research specifically tracks which new behaviors become habitual. It’s not one-off curiosity — a meaningful share of workers integrate these expanded activities into their regular workflows within weeks.
- Role identity shifts: Some workers begin redefining what their job actually is based on what AI makes possible, not what their original job description said.
- Managerial blind spots: Many of these new activities happen without explicit manager awareness or approval — workers are quietly self-promoting their own scope.
The granularity here matters. This isn’t survey data asking “do you use AI at work?” It’s behavioral analysis of what people actually do, and which patterns repeat. That’s a much harder dataset to collect and a much more valuable one.
Which Roles Are Expanding the Most?
The research points to knowledge workers in mid-level roles as the primary group driving this expansion. They have enough domain expertise to know what’s useful, enough autonomy to experiment, and enough frustration with organizational bottlenecks to want to bypass them. Junior employees lean on AI more for core task execution. Senior employees tend to use it for synthesis and strategy. But the middle layer? They’re the ones jumping tracks.
Customer-facing roles are particularly interesting. Support staff, account managers, and sales coordinators are using AI to do work that previously sat with analysts, ops teams, or product managers. When a support agent can pull structured insights from 200 customer conversations and turn it into a product feedback memo — something that used to require a dedicated researcher — that changes the economics of that role significantly.
The Recurring Behavior Problem (Or Opportunity)
Here’s the thing: when a new behavior becomes recurring, it stops being optional. If a marketing manager now regularly produces their own data dashboards because AI makes it possible, that expectation gets baked in. Future performance reviews assume it. Future job postings include it. The role has changed, and often the compensation structure hasn’t caught up.
This is the tension buried in OpenAI’s research. Framing it as “unlocking new ways of working” is optimistic. You could just as easily frame it as workers absorbing additional scope without additional pay. Both readings are valid, and OpenAI — to its credit — doesn’t pretend the optimistic framing is the only one.
What This Means for Businesses, Workers, and the Broader Market
For businesses, this research is essentially free organizational intelligence. If your workers are already expanding into adjacent roles using AI, you have a few options. You can formalize it — restructure roles, update job descriptions, rethink team compositions. You can ignore it and let it continue organically. Or you can restrict it, which at this point seems both futile and counterproductive.
The smarter companies are already paying attention to this. We’ve seen how tools like AI executive assistants are reshaping what administrative roles look like — our earlier look at how Fyxer built an AI executive assistant people actually trust showed how much scope can shift when the right tool is in place. OpenAI’s research provides the macro data to back up what individual product stories have been hinting at.
For workers, the message is complicated. AI is genuinely expanding what individuals can accomplish — that’s real and valuable. But expanded capability doesn’t automatically translate to expanded leverage in salary negotiations or career advancement. Workers who are quietly doing two jobs’ worth of work because AI makes it possible should probably be a lot more intentional about making that visible.
For the broader AI market, this research matters because it shifts the conversation from displacement to transformation. The debate has been “will AI take jobs” for years now. OpenAI’s data suggests the more accurate frame is “AI is changing what jobs are” — which is both less scary and more complex to navigate. It also has direct implications for AI tool pricing strategy. If workers are using ChatGPT to expand their professional scope, the value being captured goes well beyond what any per-seat subscription price reflects.
It’s worth comparing this to what’s happening at the enterprise level too. ChatGPT’s Data Agent is already positioning itself to replace BI tools — and if individual workers are simultaneously teaching themselves data analysis through AI assistance, those two trends are going to collide in interesting ways inside organizations.
How Should Companies Respond?
The practical implication for HR and operations teams is that job architecture is now a live problem, not a once-every-few-years project. If worker behavior is already outpacing job descriptions, the org chart is already out of date. Companies that run regular audits of what employees are actually doing — not just what their title says — will have a real advantage in understanding where value is being created and where role gaps are emerging.
Training and upskilling programs also need to catch up. Most corporate L&D is still built around role-specific skills. OpenAI’s research suggests the more valuable investment might be in teaching workers how to use AI to cross-skill responsibly — understanding the limits of what AI-assisted work in an unfamiliar domain actually produces.
Key Takeaways
- OpenAI’s economic research tracks behavioral change in real work settings, not just self-reported survey data — making the findings more reliable than most workforce AI studies.
- Workers are expanding their job scope using AI, often without manager awareness, and certain new activities are becoming permanent fixtures in their workflows.
- Mid-level knowledge workers are driving the most significant role boundary expansion.
- The optimistic read is productivity and career growth; the realistic concern is scope creep without compensation adjustment.
- Businesses that don’t actively track these behavioral shifts are already working from an outdated picture of their workforce.
- This research strengthens the case for treating AI adoption as an organizational design challenge, not just a technology procurement decision.
Frequently Asked Questions
What is OpenAI’s Economic Research on new ways of working?
It’s a study published in September 2026 examining how workers use AI tools — particularly ChatGPT — to expand beyond their traditional job functions. The research focuses specifically on which new behaviors become recurring parts of workers’ routines rather than one-off experiments.
Who does this research apply to?
The findings are most relevant to knowledge workers in mid-level roles, though the patterns appear across industries and job types. HR leaders, operations managers, and executives thinking about workforce planning will find it particularly useful for understanding where roles are already quietly evolving.
Does this mean AI is replacing workers?
Not according to this research — the pattern it identifies is expansion, not replacement. Workers are taking on more varied tasks, not being pushed out. That said, if one worker can now cover scope that previously required two, the longer-term headcount implications are worth watching.
How does this compare to what other AI companies are tracking?
Most AI companies report usage metrics and customer satisfaction — not labor market behavioral data. OpenAI’s decision to publish economic research at this level of specificity is relatively unusual and positions them as a more active participant in the policy and workforce conversation, something we’ve seen them pursue more aggressively through moves like their recent policy push in Washington.
The real question going forward is whether organizations start redesigning around this reality, or whether they keep pretending job descriptions written in 2022 still reflect what their teams actually do. OpenAI’s research makes that pretense harder to maintain. And as AI capabilities continue maturing — with tools like GPT-6 Astra pushing into genuinely complex reasoning tasks — the scope of what workers can teach themselves to do with AI assistance is only going to widen. The organizational structures will have to follow eventually.
