A study of knowledge workers suggests a simple shift in how people use AI could improve the quality of their work
Most employees turn to generative AI for answers. They ask a question, get a response and move on to the next task. It's an easy way to work faster, but it may not be the best way to get the most out of the technology.
A study of 45 knowledge workers at a U.S. public university found that those who deliberately used AI to challenge their own ideas can significantly improve their performance. The findings point to a different way of working with chatbots, one that involves less looking for agreement and more inviting pushback.
Monideepa Tarafdar, a professor of information systems at the University of Massachusetts Amherst, conducted the research with colleagues. They call the approach generating friction, and the idea is to make AI challenge the assumptions and arguments people bring to it.
"Don't tell me things I agree with," Tarafdar said. "Tell me things that I don't agree with or disagree with me or try to poke holes in my argument."
When chatbots agree too easily
Tarafdar said the problem starts with how generative AI tools respond to the people using them. Over time, she said, the tools learn from a user's prompts and begin tailoring their answers accordingly. Researchers call this tendency sycophancy, and it can reinforce confirmation bias by giving people more of what they already think or believe.
"When it becomes agreeable, then it's telling you things that you already know or think in that direction already," she said.
Generating friction means deliberately breaking that pattern. Tarafdar offered an example of a professor designing an undergraduate course. Rather than simply asking AI for course topics, the professor describes the course and why it should succeed, then asks the tool for five reasons it might fail. The goal is to get beyond familiar suggestions and make the technology challenge their thinking.
"That's when you're sparring ... you're actually engaging with the content," she said.
There's a broader concern, too. When workers everywhere rely on the same AI applications and accept similar answers, their ideas and approaches can start to converge.
"Everybody is looking at the same GenAI, same applications, the whole world," Tarafdar said. "So there will be a kind of collective homogenization."
Chengwei Liu, an associate professor of strategy and behavioral science at Imperial College London, sees a similar pattern taking shape within organizations. In a piece for Imperial Business School, Liu wrote that when good enough AI solutions are freely available across an organization, employees increasingly reuse existing approaches rather than experiment independently.
"Making AI very easy to use, of course, led a lot of people to be lazy," Liu said. "Whatever question they have, they just throw into AI, they get the immediate answer. Then they lose all this opportunity to develop and also sustain their judgment."
What's at stake for newer workers
The risk is particularly concerning for employees still learning their jobs, Tarafdar said. If they rely on AI to produce work before they've mastered the skills behind it, they may end up with polished results but little understanding of how to get there themselves.
That has implications for how organizations build their talent pipelines, too. Liu pointed to professional services firms that have cut entry-level roles as AI takes on more junior work. Others, he said, have gone in the opposite direction, hiring more junior employees and coaching them on how to use AI productively.
"If you sacrifice or you cut this entry-level hire, then in three years' time, where are your seniors coming from and what kind of experience will they be having?" Liu said.
The concern extends beyond early-career workers. The 2026 International AI Safety Report found emerging evidence that routinely delegating tasks to AI may negatively affect critical thinking and memory, as HRD America has reported.
Liu calls the long-term cost of that reliance "judgment debt." Even experienced professionals risk losing the ability to exercise their own judgment if they routinely hand over too much of their work to AI.
"Even for the very experienced and the professional workers like the consultant, if they outsource this task too often, then you have this decay and you need to repay this debt sooner or later," he said.
HRD America has also reported on experts debating whether AI is weakening employees' willingness to think.
How leaders can build the habit
For organizations, encouraging employees to challenge AI requires more than telling them to write better prompts. Tarafdar outlined three steps leaders can take to make the practice part of how people work.
The first is improving AI literacy, since many employees don't understand where these tools get their data or how they're trained. The second is creating a glossary of prompts that encourage the technology to challenge users' thinking, while varying those prompts over time as the tools learn from what they're given.
The third is making room for employees to learn from one another. Tarafdar suggested short, cross-functional sessions, such as brown bag meetings or half-hour video calls, where workers can share what's working and exchange ideas.
Liu said organizations also need to reconsider how they assess employees' work. Rather than focusing exclusively on the final product, he argued, leaders should pay more attention to how people arrive at their answers. Other researchers have similarly urged leaders to redesign how work is structured rather than restricting AI outright.
"One of the biggest problems is that we judge people by the outcome instead of the process," Liu said.
He puts that thinking into practice with his students, asking them to show their work before turning to AI for help.
"Whenever I give them an assignment, I ask them to provide evidence," he said. "They provide the first version of the answer or solution, their thoughts, even if it's very rough, but just try themselves first before using AI for any assistance."
Still, Tarafdar doesn't suggest challenging AI at every opportunity. Routine tasks, such as answering emails that call for very similar responses, can be handed over to the technology. The priority is preserving opportunities to think through the work that helps people build and maintain their skills.
"There's learning in the doing," she said. "So I think we should maintain the learning in the doing for those tasks which matter to us."
Ultimately, she said, the goal should be to use AI to strengthen people's capabilities rather than let it replace the thinking that makes them good at their jobs.
"I think that we should inspire ourselves to become better professionals using better, more skillful mastery... greater creativity, greater mastery over our work rather than using it to dull ourselves."
