What: A survey of 1,033 US full-time desk workers on intact project teams finds that turning to AI instead of a colleague raises perceived expertise but cuts team coordination by 22.3% and team trust by 21.9% where the habit becomes the norm.

Why it is important: The remedies are inexpensive and within leaders' control: daily asynchronous contact restores 28% of coordination and 19% of trust even among displacers, and incentives can be re-architected to reward AI-enabled teams rather than AI-literate individuals.

Great teams run on a shared map of who knows what. Surveying 1,033 US desk workers across 20+ industries, the authors find AI redrawing that map in conflicting ways: heavier AI use raises the perceived expertise of teammates, yet going to AI instead of a colleague erodes mutual trust and coordination. One in five workers admit a tendency to displace teammates; 49% say they have been displaced at least some of the time.

Two costs accumulate. Teams lose track of who knows what, since AI-generated work hides whose expertise produced it, and trust gives way to second-guessing and re-checking. Only 32% of those who prefer AI input report high coordination, against 60% of those who favour teammates. Displaced employees feel they matter less and search for jobs more actively; for a 10,000-employee company, the authors model 243 extra departures and over $25 million in replacement costs.

Displacement tracks AI optimism, low psychological safety and the models' agreeable design rather than age or seniority. The authors recommend frequent asynchronous contact, psychological safety, transparency norms on AI use, team-level incentives, augmentation over automation, and using AI to prepare for conversations rather than replace them.

IADS Notes: The displacement effect described here extends a pattern that workforce research has been documenting through 2026. The hidden costs of AI acceleration were first framed as a "fluency gap": employees become proficient with the tools while feedback loops weaken and the judgment needed to oversee outputs erodes, because adoption is pursued without redesigning how work is done (Seramount, March 2026). The sycophancy the authors identify as a driver of displacement had already been flagged in a consumer context, where chatbots' validating design was shown to reinforce users' beliefs and remove the critical feedback that human exchange provides (Stanford University, April 2026). On the organisational side, psychological safety was identified as the condition that lets employees ask questions, admit confusion and challenge AI outputs, with executives shown to overestimate staff enthusiasm for the technology (Seramount, June 2026). Middle managers were then found to be where AI initiatives most often stall, with five distinct profiles each requiring a matched intervention rather than generic evangelism (Harvard Business Review, September 2026). Most recently, data from more than 10,000 US employees showed that 88% of organisations use AI regularly while only 30% redesign processes around it, a gap producing "quiet compliance" in which tools are used without questioning outputs, with a recommendation to track psychological safety alongside adoption (Seramount, October 2026). The team-level trust and coordination losses quantified here add a relational dimension to that same adoption-versus-value gap.

Team dynamics weaken when employees turn to AI instead of colleagues