In the early 1970s, the late Yale University research psychologist Irving Janis coined the term “groupthink.” Journalist Kathrin Lassila explains that groupthink occurs when “a group of intelligent people working together to solve a problem sometimes arrive at the worst possible answer.”[1] She reports, “[Janis’] radical new theory … changed the way we think about decision making.” Wikipedia provides a more formal definition: “Groupthink is a psychological phenomenon that occurs within a group of people in which the desire for harmony or conformity in the group results in an irrational or dysfunctional decision-making outcome. Cohesiveness, or the desire for cohesiveness, in a group may produce a tendency among its members to agree at all costs. This causes the group to minimize conflict and reach a consensus decision without critical evaluation.”

Groupthink isn’t the only way that people come together to make bad decisions. Wikipedia defines collusion as “a deceitful agreement or secret cooperation between two or more parties to limit open competition by deceiving, misleading or defrauding others of their legal right. Collusion is not always considered illegal. It can be used to attain objectives forbidden by law; for example, by defrauding or gaining an unfair market advantage. It is an agreement among firms or individuals to divide a market, set prices, limit production or limit opportunities.”

Recently, a group of academics has come up with a new decision-making model for the Artificial Intelligence Era: Agentic Convergence. They warn companies using AI that Agentic Convergence is a trap in which they could be caught.

The Agentic Convergence Trap

Researchers, Patrick van Esch and Yuanyuan Gina Cui, professors at Coastal Carolina University’s E. Craig Wall Sr. College of Business Administration, and their colleague J. Stewart Black, a professor of global leadership and strategy at INSEAD, explain, “When companies deploy AI systems trained on the same market data, optimizing similar objectives at machine speed, they risk falling into a ‘Agentic Convergence Trap’: independent systems arrive at identical decisions, eroding differentiation and sometimes triggering regulatory scrutiny.”[2] The researchers discuss three case studies of companies caught in the Agentic Convergence Trap: “Three industries. Three decisions executives thought were competitive positives. Three outcomes that resulted in strategic, self-inflicted wounds.”

The thing I find fascinating about their work is that it updates the dangers of “group” decision-making even though the group never gets together. The first decision they discuss is a federal class action lawsuit against six major hotel chains. The suit claims “that their shared AI pricing platform had produced coordinated room rates across competing properties.” The Department of Justice didn’t call this collusion, they called it “price coordination.” The second decision discussed by the researchers involved a regional grocery chain that replaced human promotion planners with an AI system trained on market signals. As in the hotel case, the grocery chain’s promotions mirrored its competitors’ promotions. The final case study involved a national landlord who adopted the same AI rent optimization platform used by thousands of other property managers. All of the firms raised rents in lockstep even though they had never had any contact with one another. In that case, the Department of Justice named the national landlord in an antitrust action.

The researchers explain, “When multiple companies deploy AI systems that learn from overlapping market data, optimize similar objectives, and operate at machine speed, a pattern consistent with what we and other researchers have documented in AI-mediated markets, and one that peer-reviewed research in the American Economic Review and Journal of Political Economy has now measured empirically, those systems tend to arrive at the same conclusions independently. We call this the Agentic Convergence Trap. Understanding it requires understanding not just how AI systems behave, but how executives have enabled the behavior.” The question they raise is this: “What happens when your AI and your competitors’ AI agents are deployed in the same market and begin learning from each other simultaneously?”

Groupthink, Collusion, or Something Else

Framing the problem using older models isn’t really very informative. When you read about AI systems learning from each other, its sounds like a case of groupthink. When optimization decisions generated by those AI systems move markets in lockstep, it sounds like collusion. However, since the parties involved never actually communicate with one another, the researchers preference the term “convergence.” They explain, “Independent AI agents, trained on similar data, optimizing similar objectives, at machine speed, develop nearly identical models of market reality and act on them in near-identical ways. Not by design. Not through communication. By learning.” They add, “This is not a theoretical risk. It is already measurable across retail, hospitality, airlines, and housing. And it is accelerating, because the agentic AI systems driving convergence are themselves getting faster.”

If, as the researchers believe, this is a trap, organizations need to learn how to avoid the trap or escape it once they are in it. “Switching vendors,” they note, “won’t help.” Why? Because, they explain, “the root of the convergence is not the common software, it is the common learning process and its speed.” Humans, they explain, can avoid groupthink by introducing natural variations in world views. They conclude, “Those frictions, often regarded as inefficiencies, are the mechanism that produces strategic diversity.” AI systems don’t insert human variations. As a result, the researchers note, “The strategic question is not whether your AI works. It is whether it is working for you or quietly working for the whole market.” In other words, it is unknowingly colluding.

If you are wondering whether the researchers are suggesting that AI systems be dumped, they are not. They explain, “The convergence trap is not primarily a technology problem. It is a leadership and governance failure that technology makes invisible. … Most AI governance frameworks focus on accuracy, bias, and legal risk. Almost none treat the preservation of strategic variation as a governance objective. The companies that will avoid the convergence trap are those with processes designed to ask one question before delegating any decision to AI: What would happen to our competitive position if every rival made exactly this same AI-driven choice today?”

They offer four suggestions to help organizations avoid the agentic convergence trap. They are: 1) Decide where humans stay in the loop; 2) Define what your AI optimizes beyond the platform default; 3) Feed your AI data your competitors cannot access; and, 4) Measure convergence, not just performance.

Concluding Thoughts

The researchers make their position very clear. “This is not an argument against AI,” they write. “It is an argument for executive accountability. … Competition is no longer about who has the strongest algorithm. The winners in the next phase of the AI transformation will be the companies that have designed their AI to reach conclusions their competitors’ AI will not.” I find their conclusions to be very compelling. Like groupthink and collusion, agentic convergence can lead to decisions that do more harm than good. The decision-making process can benefit from a little human variation, which can best be infused by asking pointed “what if” questions.

Footnotes

[1] Kathrin Lassila, “A brief history of groupthink,” Yale Alumni Magazine, January/February 2008.

[2] Patrick van Esch, Yuanyuan Gina Cui, and J. Stewart Black, “Beware the Agentic Convergence Trap,” Harvard Business Review, 13 May 2026.