KPMG and the University of Texas at Austin conducted field research with 523 new employees in the US (with tenure under 18 months) to evaluate how newcomers collaborate with AI in real-world business tasks.
Participants completed professional exercises with the support of an AI Agent, after which the results were compared against a benchmark completed by AI alone to determine the value added by humans.
The results categorized participants into 3 groups: AI Apprentices (24.1%) performed lower than AI; AI Delegators (25.8%) achieved results equivalent to AI; and AI Amplifiers (50.1%) generated results superior to AI.
Surprisingly, critical thinking, domain expertise, and AI literacy did not predict who would fall into which group. The key difference lay in how they collaborated with AI during the work process.
The AI Apprentices group had a solid foundation but did not know how to leverage AI. They often asked AI to rewrite or reorganize information instead of using AI to test assumptions, evaluate results, and elevate output quality.
The AI Delegators group largely accepted AI-generated outputs with minimal adjustments. Because AI produced reasonably good output, they still achieved acceptable performance, but added almost no extra value.
The AI Amplifiers group used AI as a thought partner: they framed problems correctly, asked AI to explain its reasoning, tested assumptions, explored alternative possibilities, and continuously refined outputs to meet real-world work standards.
The study recommends that businesses rethink how they train new employees, focusing on practical exercises embedded in AI-driven workflows rather than teaching AI skills or domain knowledge in isolation.
KPMG is implementing a personalized AI training program using real-world job simulation scenarios and continuous learning pathways to foster AI Amplifier behaviors company-wide.
The authors argue that companies should evaluate employees not only on final outcomes but also on how they work with AI, how they frame problems, verify outputs, and make decisions. In the future, human value will increasingly lie in the ability to direct, evaluate, and scale AI-generated outputs rather than simply generating answers independently.
📌 Conclusion: Research demonstrates that AI does not automatically level the playing field among employees; instead, it highlights differences in how individuals work. In a survey of 523 new employees, about half produced results superior to AI, while nearly a quarter performed worse than AI itself. The deciding factor was not knowledge or AI literacy, but the ability to collaborate with AI, frame questions, critically evaluate, and continuously improve outputs. This indicates that businesses must redesign how they train, evaluate, and develop talent in the AI era.
