Chances are good that your organization has already installed at least one AI system that can make smart decisions. Maybe it does demand forecasting, maybe application processing. Whatever the case, chances are just as good that, even if it's a reliable system, humans are rejecting its recommendations.
This may look like a technology problem or a change management problem. Harvard Business School's Das Narayandas argues that it is usually something more basic: a problem of authority and accountability. In this issue's cover story, Why Even Your Best Employees Override AI Systems That Work, he describes organizations in which the AI performs well but humans continually overrule it.
Why? Because the person asked to trust the system is often the person held responsible when its decision goes awry. Under those conditions, choosing to override AI is not necessarily resistance to change. It can be an act of rational self-protection.
That distinction matters. If companies want AI to make binding decisions, Narayandas contends, they must reassign both the authority and the accountability for those decisions. That means redesigning roles, decision rights, incentives, and the rules governing when people should intervene.
AI adoption, then, depends not only on what the technology can do and whether employees trust it. It also depends on whether a business is willing to reorganize around what AI can do — and whether employees trust the business to follow through.
Copyright 2026 Harvard Business School Publishing Corporation. Distributed by The New York Times Syndicate.