The case study
KPMG and the University of Texas at Austin studied 523 US-based early-career professionals working with an AI agent on tasks that mirrored real client work. Researchers first established an AI-only baseline, then compared how people collaborated with the same kind of system.
The result was not explained by traditional measures of knowledge or capability. The difference came from how people directed the AI, evaluated its work, and refined the result.
The amplifier loop
The useful response is not to ask people to “use AI more.” It is to teach a repeatable loop that makes human value visible:
Copy this coaching brief
A small team application
A product team uses the loop for a weekly customer-signal report. The assistant clusters feedback and drafts themes. The product manager frames the decision the report must support, directs the assistant toward the approved sources, critiques unsupported claims and missing segments, then asks for a refined version with confidence notes and open questions.
That workflow creates a better report and a better learning signal. The team can see which parts AI handles well, where domain expertise changes the result, and what coaching future users need.
The takeaway
The strongest AI users are not merely prompt writers or final approvers. They are workflow designers. They help AI aim at the right problem, test the output against reality, and improve it until it is ready for a human decision.