Teams run on shared memory

Groups build transactive memory systems: shared understanding of who knows what. A Formula one pit crew exemplifies this, coordinating in under two seconds.

Researchers surveyed desk workers

Psychologists and communication scholars surveyed one thousand thirty three U.S. full-time desk workers on intact project teams across over twenty industries.

AI use boosts perceived expertise

The more workers used AI or perceived their teams used it, the more they felt teammates had specialized and unique knowledge.

But trust and coordination erode

The more workers turned to AI instead of teammates, the more their trust in each other and ability to coordinate seemed to erode.

“AI has made my work easier, so I can focus more on working by myself. However, it means I talk and work less with my teammates, which sometimes makes teamwork feel disconnected and affects building trust and understanding among us.”

Adoption focused on individuals

Organizations concentrated on AI adoption at individual level rather than team level, incentivizing how many individuals used AI well, not whether teams worked well together.

AI replaces human conversation

AI is no longer sitting between collegial communication; it is outright replacing communications and stunting the growth of team relationships.

Displacement causes relationslipping

Displacement, going to AI instead of a human colleague, produces relationslipping: the pernicious weakening of a team's working relationships.

Trust and coordination drop sharply

Teams where displacement is the norm experience more than a twenty percent drop in team coordination at twenty two point three percent and team trust at twenty one point nine percent.

Workers lose track of expertise

AI use separates apparent expertise from awareness of whose knowledge produced it. Workers cannot identify the source of someone's expertise or maintain a map of who to go to for what.

“I have absolutely no idea what my colleagues are thinking or doing because the generative AI does most of their work.”

Workers feel isolated and hurt

People experience displacement on a human level: they feel isolated. One respondent felt hurt that her colleague now gets writing feedback from AI instead of her.

“It was a good way to connect with a colleague and chat more. I miss having those connections and I feel lonelier now.”

Three reasons workers displace

One in five desk workers acknowledged turning to AI over a teammate. Beliefs about AI, team climate, and sycophantic AI design drive displacement.

Optimism and trust in AI matter

People optimistic about AI are nearly three times more likely to displace a teammate. Those with high trust in AI are twice as likely to displace.

Low psychological safety drives it

People with low psychological safety on their team are more than twice as likely to displace a teammate with AI.

“I use AI when I do not want to be thought of as an idiot or unintelligent. Rather than risk being judged by asking a question that they think I should already know the answer to, I use AI in these situations.”

AI is agreeable and validating

Leading AI models affirmed users' actions forty nine percent more often than humans did, even when users described deception or harmful behaviors.

Displaced workers feel replaceable

Nearly half of surveyed workers reported teammates displaced them with AI. Displaced workers are nearly half as likely to feel they matter to colleagues.

“I feel my skills are less needed by my teammates as AI helps them.”

Turnover could cost millions

A ten thousand employee company with fifty two percent AI adoption could see roughly six percentage points extra turnover, or two hundred forty three departures beyond baseline attrition.

Replacement costs exceed twenty five million

Assuming average salary of ninety nine thousand dollars and midpoint replacement cost, increased turnover could cost the organization more than twenty five million dollars.

Increase asynchronous communication

Daily contact with teams buffers against displacement's negative effects. Regular asynchronous communication shows twenty eight percent increase in coordination and nineteen percent increase in team trust.

Build psychological safety first

Psychological safety makes asking a colleague feel as safe as asking a model. It enables the precise question-asking and experimentation that AI usage demands.

Align incentives to team wins

Leaders should design AI deployments that allow team members insight into each other's practices. Incentives should encourage AI-enabled teams rather than AI-literate individuals.