Artificial Intelligence in Team Dynamics: Who Gets Replaced and Why? (www.nber.org)

A new theoretical study models how a principal should allocate limited AI agents inside sequential team workflows where workers discipline each other via peer monitoring (each worker observes the predecessor’s effort). AI agents are immune to moral hazard, so replacing humans with AI changes both incentives and information flows. Using a sequential team production model, the authors ask which positions are most vulnerable, how replacement should be allocated, and what happens to wages of replaced and remaining workers. The analysis yields four counterintuitive results with practical implications for AI deployment and labor policy. Optimal replacement is stochastic—AI is allocated probabilistically across positions rather than fixed on one role—and targets the first and last stages of the workflow while deliberately preserving the middle worker because that role sustains the information chain needed for peer monitoring. Principals may leave some AI capacity unused, since over-replacement weakens monitoring and reduces total output. Surprisingly, optimal AI adoption raises average wages and compresses intra-team wage inequality by improving incentives and reducing moral-hazard rents. The study highlights that strategic, partial automation can preserve complementarities in team processes and reshape both who is at risk and how gains from AI are distributed.