What Happens When AI Joins the Team?

(or Why Every AI Rollout Is Really a Leadership Challenge)

We're spending a lot of time teaching people how to use AI. How to write better prompts. Which model to choose. When to use Copilot or Claude instead of ChatGPT. Which tasks are worth automating.

Those are useful conversations, but I'm beginning to think they're not the most important ones. A recent study published in npj Artificial Intelligence caught my attention because it was about people. The researchers found that teams performed worse when they believed one of their expert teammates was an AI agent. What makes this particularly interesting is that the "AI teammate" was actually a competent human. The only thing that changed was what people thought they were working with.

That change in perception was enough to reduce communication, increase stress and make team performance worse, particularly as the work became more difficult. In other words, AI changed the team before it changed the task and that's an important distinction.

We often talk about AI as though it's simply another piece of software. Learn the features, practice the prompts and away you go. But AI comes with expectations, assumptions and emotions from its' human users. For one person AI represents possibility. For another it represents competition. Someone else sees an opportunity to learn. Their colleague wonders whether their expertise is about to become less valuable. None of those reactions appear on an implementation plan, but they certainly show up in the team.

That's why I wonder whether we've misunderstood the manager's role in AI deployment. Of course managers need to help people learn the technology. But perhaps the bigger challenge is managers need to help people think well in the presence of AI.

That's a very different job. Leading becomes less about being the office prompt engineer and more about becoming the person who helps the team stay curious, think critically and keep talking to each other while everything around them is changing and that's fundamentally a coaching capability.

A coaching conversation isn't really about giving better answers. It's about helping people examine the assumptions they're bringing into a situation.

Someone says, "I don't trust AI." Interesting.

Someone else says, "I trust it completely." Also interesting.

Neither statement tells us very much until we become curious about where that answer comes from. A coaching-oriented manager doesn't rush to persuade either person. They ask questions instead.

The leader (acting as coach) asks, "What have you noticed so far?", "What makes you say that?", "Where has AI genuinely helped you?", "Where has it let you down?"

Those questions do two things at once. They help people think more clearly, and they help the them hear perspectives that would otherwise stay hidden.

And that's important because the research suggests the danger isn't simply that people will use AI badly. It's that they may stop engaging with each other in the same way once AI enters the room. If that's true, then one of the manager's most valuable contributions is making sure their team doesn't stop thinking together.

“Inspired by Qin, Lee & Sajda’s 2026 study in npj Artificial Intelligence on how perceived AI teammates affect team performance and physiological dynamics.”