Here's a question I've started asking owners: when was the last time someone on your team asked a colleague a question instead of asking AI?
Most can't answer. That's the problem.
AI is making your people faster. New research says it may also be making your team slower. Not at the task level. At the team level, where the real work of a growing company gets done.
In October, Harvard Business Review published a study from researchers at BetterUp and Stanford. They surveyed 1,033 U.S. full-time desk workers on project teams across more than 20 industries.
The good news: as AI use went up, people felt their teammates had more specialized knowledge.
The bad news: the more people went to AI instead of a teammate, the more trust and coordination fell. On teams where that was the norm, coordination dropped about 22% and trust dropped about 22%.
The researchers have a name for it: "relationslipping." It's the slow erosion of working relationships when people substitute AI for each other.
Nobody decides to stop collaborating. It happens one skipped question at a time.
The authors use a Formula 1 pit crew as the model of a great team. Everyone knows their role. Nobody re-checks anyone's work. The whole thing is done in under two seconds.
That only works because everyone knows who knows what. Psychologists call it a transactive memory system. Owners just call it "the team works."
AI quietly breaks that map in two ways:
The result? People double-check each other's work. Effort gets duplicated. Projects hit dead ends. Then you schedule a meeting to figure out what everyone already did. Sound familiar?
This is the part owners need to hear. Relationslipping isn't laziness. The research points to three drivers, and each one is a reasonable choice in the moment:
And it isn't generational. The pattern showed up across age, seniority, industry, and team size.
So let me get this straight. We tell people to use AI more. We measure who uses it. We reward individual output. Then we're surprised when the team stops talking? That's an incentive problem, not a technology problem.
The human cost is real. Nearly half (49%) of workers said teammates displace them with AI at least some of the time. People who feel displaced are much less likely to feel they matter to their team, and more likely to look for another job.
The researchers modeled a 10,000-person company and estimated roughly 243 extra departures and more than $25M in turnover cost.
Most of our clients aren't 10,000 people. So here's our rough extrapolation, using the study's assumptions, for a 150-person company:
Estimated Annual Impact, 150-Person Company
Rounded, labeled, and not precise. But it's enough to wipe out a lot of AI time savings.
The authors are clear on this, and so are we. In their words:
"We're the problem, it's us."
Some displacement is healthy. One sales manager in the study was relieved a coworker used AI when they were too busy to help. The goal isn't fewer AI conversations. It's more of the right human ones.
In our experience, the fix isn't a policy. It's a habit. Here's what we put in place with clients.
If AI saves someone 1 to 2 hours, that's a nice personal win. If they share it with 3 to 5 teammates, as a scheduled prompt or a Claude skill, it becomes a team win worth several times that. And it gives people a reason to talk. In our AI Value Drivers Program, a win doesn't count until it's taught to at least one person outside the team.
Our rule is simple: it only counts if you share it.
Even among people who go to AI first, the study found about 28% higher coordination and 19% higher trust when they were in daily contact with their team, especially through Slack, Teams, or email. We add a 30-minute demo every two weeks where people show what they built and how.
If asking a teammate feels riskier than asking a chatbot, the chatbot wins. Leaders go first: ask questions out loud and admit what you don't know.
The authors found a 5-minute AI role-play before a tough conversation reduced anxiety and built confidence. Use it to get ready. Then go talk.
We run shadow AI surveys and quarterly team surveys to see who's using what, where tools are creating confusion, and whether people understand the plan. Usage dashboards tell you who's busy. Team surveys tell you if you're winning.
Every sprint, leadership sees three numbers: skills shipped, adoption rate, and documented ROI. Adoption rate is the one that tells you if the team is learning together.
One of our core beliefs at Coach.Win: people know what winning looks like. AI doesn't change that. It just makes it easier to win alone and lose together.
Ask your leadership team one question this week:
"What's one AI win you've had that nobody else on the team knows about?"
Then make them share it.
Where are you seeing this? Has AI made your team more connected or more siloed? Tell me in the comments.
And if you know an owner who needs to read this, share it with them.
If your team is using AI but nobody's sharing the wins, we can help you build the habit. Book a discovery call with Coach.Win, or join Stay in the Loop for more like this.
Book a Discovery Call Stay in the Loop