OpenAI just published data on what workers actually do with AI when you're not watching, and it's not what the productivity consultants predicted.
The Summary
- OpenAI's economic research team tracked real worker behavior with AI, finding adoption patterns that don't match traditional job descriptions or productivity assumptions
- Workers use AI tools to expand into adjacent tasks and roles, not just to speed up their core work
- The recurring activities that stick are collaborative and creative, not just efficiency-focused
The Signal
OpenAI's research cuts through the "AI will replace jobs" versus "AI will augment jobs" binary with actual behavioral data. The study tracked how workers integrate AI into their workflows and found something more interesting than either camp expected: people use AI to do work they weren't hired to do.
The pattern shows up across job types. A marketing coordinator starts doing competitive analysis that previously required an analyst. A customer service rep begins writing internal process documentation that used to come from managers. An accountant starts drafting client-facing reports that would have gone to a partner for review.
"Workers use AI to expand into adjacent tasks and roles, not just to speed up their core work."
This isn't productivity theater. The research shows these expanded activities become recurring parts of people's jobs. Once workers discover they can handle something previously outside their scope, they keep doing it. The AI doesn't just make the old job faster. It redefines what the job includes.
Three patterns emerged from the data:
- Workers blur the boundaries between roles, taking on tasks from adjacent functions
- Creative and collaborative work sticks, while purely mechanical tasks get automated away quickly
- The most successful adopters use AI to increase their scope, not their speed
The implication for organizations: job descriptions are becoming historical documents. The research suggests companies that try to lock workers into narrow AI-assisted efficiency gains will lose to companies that let workers expand their scope. The competitive advantage isn't faster execution of the same tasks. It's workers who can do more kinds of work.
The Implication
If you manage people, stop measuring AI adoption by time saved on existing tasks. Start tracking how many new types of work your team members take on. The workers who thrive aren't the ones who do their old job 30% faster. They're the ones who do their old job plus three things that used to require different people.
For workers, this is permission to experiment beyond your job description. The AI tools you already have access to can make you competent at tasks you've never been trained for. The question isn't whether you can use AI to do your current job better. It's what job you want to grow into.