Generative AI can become a very polished yes-machine. Knowledge workers interviewed for a study found a more useful role for it: challenge their assumptions, surface unfamiliar perspectives and help them think through a problem, rather than simply produce the finished answer.
Watch Desk analysis
What happened
Monideepa Tarafdar, a professor of information systems at the University of Massachusetts Amherst, describes a study in which she and colleagues interviewed 45 knowledge workers at a US public university and analysed their iterative AI conversations. The workers used generative AI in tasks including idea generation, complex problem-solving and designing new solutions.
Tarafdar says the interviews and chat analysis pointed to a useful shift: deliberately asking AI to create friction can help people spot possibilities they might otherwise miss. One marketing professional asked it to invent customer personas likely to dislike a proposed accounting certificate programme. Their objections raised a specific idea the designer had not considered: using case studies from different industries rather than repeatedly drawing from the same ones. Tarafdar’s article was republished by Greater Good from The Conversation.
Why it matters
The common pitch for workplace AI is that it saves time by doing the work. This account suggests a less flashy, potentially more valuable use: make the system generate counterarguments, awkward scenarios or perspectives outside your usual lane, then use your own expertise to judge what holds up.
Tarafdar also describes an attorney using AI to look for obscure legal loopholes and an operations research scientist adapting ideas from other technical fields. Those are examples from the study, not proof that an AI suggestion is correct or that every worker will get better results. The point is to treat the model as a source of prompts for further thought, not as the final word. An agreeable answer is pleasant; it is not necessarily useful.
Our read
“Ask it to disagree” is a better workplace habit than “ask it to do everything”. Try a prompt that requests the strongest case against your plan, a customer who would reject it, or a plausible failure mode. Then check the response against your knowledge and the evidence. The friction is the feature, not a personality defect in the chatbot.
What to watch
- Whether workers use AI to test their ideas, not just draft their deliverables.
- Whether organisations make room for colleagues to share effective prompts and examples.
- Whether future studies measure performance gains beyond the cases and interviews described here.
Discussion spark: Should employers teach staff to use AI as a critic of their work, or does that risk giving a fluent model too much influence over decisions?
Sources and evidence
- How Artificial Intelligence Can Help With Critical Thinking – Greater Good: The Science of a Meaningful Life (12 October 2026, 19:16 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.