Is AI Making Teams Less Collaborative?

By Victor Hoang · Updated Oct 11, 2026

Think about the last time a new hire asked a colleague a "quick question." The answer took thirty seconds. But the exchange did more than deliver the answer. The new hire learned who knows what, how a veteran reasons through an ambiguous problem, and that it's acceptable to ask. The veteran learned what the newcomer doesn't yet understand. Nobody scheduled any of it.

Now picture the same new hire asking an AI assistant instead. The answer arrives in seconds, it's polite, and it's often good. Nothing was lost on the surface. But the conversation never happened, and the conversation was carrying a second payload.

Our position: AI doesn't make people less willing to collaborate, but it removes the small, unplanned moments where collaboration used to happen as a side effect. Teams that don't replace those moments on purpose will slowly lose their informal knowledge flow and their mentoring, and leaders won't see it in any dashboard until it hurts.

This is a different problem from the one we cover in peer-led AI programs, which is about teaching people to use AI. Here the question is what AI use does to everyday knowledge flow, even among people who use it well.

Key Facts

• In a 2026 MyIQ survey of 22,481 workers, 74% of regular AI users said they now ask a chatbot questions they would previously have asked a colleague, and 59% said they ask colleagues for second opinions less often. The release itself states the findings do not establish that AI use directly weakens workplace relationships.
• In the same survey, 46% said they know less about how colleagues think or approach problems, and 38% said newer workers have fewer natural opportunities to build relationships.
• BetterUp Labs and the Stanford Social Media Lab, in a 1,000-person study, report that 83% of workers say AI takes the place of a teammate when they need quick information, and 37% say the same for a conversation or check-in.
• In a randomized trial of 52 mostly junior engineers, Anthropic found that the group using AI assistance scored 17% lower on a comprehension quiz than the group coding by hand, while finishing only about two minutes faster, a difference that was not statistically significant.
• After ChatGPT launched, Stack Overflow lost roughly 12% of its daily traffic and newer users were the most likely to leave, according to a study in Scientific Reports.

What the Evidence Does and Doesn't Say

Be careful here, because the topic invites overreach. Most of the numbers above are self-reported surveys, and several come from vendors with products to sell. They show an association between heavy AI use and fewer colleague interactions. They don't prove that AI caused the change, and the MyIQ authors say so themselves.

But the direction is consistent across independent sources. The BetterUp and Stanford figures describe the same substitution from a different angle: AI is becoming the first place people go, not the last. The Stack Overflow study shows what happens in a public community when the first stop changes. Traffic fell, and the people who left first were the newest ones, the very people who most need to learn from others.

There's also a real counterweight. In an NBER study of 5,179 customer support agents, access to an AI assistant raised issues resolved per hour by about 14%, and novice workers improved by 34% while experienced workers saw minimal impact. The authors also report suggestive evidence that the AI model spreads the best practices of more able workers. So AI can carry knowledge to a novice who would otherwise wait for a mentor.

Both things can be true. AI is an excellent source of answers. It's a weak substitute for the other things a colleague conversation delivers. That gap is the leadership problem.

The Three Flows Framework

When you ask whether a team is staying collaborative, don't ask "are they talking?" Ask which of three flows still runs through people, and which AI has quietly taken over.

1. The answer flow. Facts, procedures, syntax, "where is that template?" This is where AI is strongest, and where moving the traffic off colleagues is mostly a gain. Don't fight it. Your experts' time is better spent elsewhere.

2. The judgment flow. How someone weighs a trade-off, when they'd escalate, what they'd never say to a customer, which numbers they distrust. This knowledge is rarely written down, so a model can only approximate it. It travels by observation and by asking "how would you handle this?" Each question that goes to a chatbot is one that goes unobserved by anyone who could pass on judgment.

3. The apprenticeship flow. Mentoring, feedback, being noticed, being corrected, learning who to trust. This one isn't information transfer at all. It's relationship and standards, and it depends on people spending unscheduled time near each other's thinking. The MyIQ respondents point at it directly when 38% say newer workers have fewer natural chances to build relationships.

The framework gives you a clean diagnosis. Answer flow moving to AI is fine. Judgment flow and apprenticeship flow moving to AI is the thing to watch, and unlike the first, nobody will complain when it happens, because nobody notices an absence.

Why the Damage Arrives Late

The cost of this shift is deferred, which is why it's easy to dismiss.

Skills first. The Anthropic trial is small and measured learning right after a task, so it shouldn't be stretched into a claim about careers. Still, its pattern is useful. The participants who scored best used AI to build understanding, by asking follow-up questions or requesting explanations. The weakest results came from delegating the work entirely. That mirrors what a good mentor does and a lazy shortcut doesn't: it forces the learner to engage with why.

Then trust. In the BetterUp study, workers who feel displaced by AI are 3.3 times less likely to want to work with their team again. Think about what that means for the veteran who used to be the person everyone asked. If teammates stop coming to her, she loses the signal that she matters to the team, and BetterUp also links that displaced feeling to a greater likelihood of looking for a new job. It connects to what we've written about how leaders respond to AI job anxiety.

Finally, the pipeline. Organizations promote people who learned judgment from others. If early-career staff learn mostly from a model that never tells them "that's not how we do it here," your next generation of managers arrives with plenty of output and thin context. This is a close cousin of the question in whether experience is still worth paying for: the experienced person's value was always partly in what they transmit, not just what they produce.

Where Remote and Hybrid Teams Feel It First

Distributed teams have fewer accidental conversations to begin with, so the substitution lands harder. If your people already coordinate through scheduled calls and messages, an assistant that answers instantly is very attractive, and the corridor conversations you lost years ago don't come back. We take that up in whether AI makes distributed teams easier or harder to lead. The short version: the same fix applies, but you'll have to schedule more of it.

A Decision Table: Which Ritual for Which Flow

You can't, and shouldn't, ban people from asking AI. Rituals work when they target a specific flow and cost little. Match the ritual to the flow you're protecting.

Two principles sit under the table. First, protect the time on the calendar, because rituals that depend on goodwill die in the first busy month. Second, make the mentor's contribution visible. Reward it in performance conversations, which is where the new performance review logic helps: if you only measure output, you've told people that teaching is a hobby.

What Good Looks Like in Practice

Illustrative scenario, not a real company. A 90-person professional services firm notices that its junior analysts produce clean first drafts but struggle when a client pushes back. The managing partner doesn't restrict the AI tools. She adds three things. Every Thursday, one senior person spends half an hour walking through a live engagement decision, including what they asked the model and where they ignored it. Each junior analyst gets a named mentor with a recurring slot. And the quarterly review for senior staff now includes one line: what did you teach this quarter, and to whom?

Within two quarters the juniors aren't using AI less. They're asking sharper questions of both the model and their seniors, because they've seen how judgment gets applied. That's the goal: AI as a tool inside a team that still learns from itself, not a replacement for the team.

A Note on Psychological Safety

There's an uncomfortable question underneath all this. Is your team asking the model because it's faster, or because asking a person feels risky? If it's the second, the ritual won't fix it. A culture that punishes visible ignorance will push questions to the one listener who never judges, and our guide to psychological safety is the better starting point. A feedback culture closes the loop afterward.

What to Do Next

1. Sort your last month of team questions into the three flows. Ask five managers what questions their people used to bring them and no longer do. The gap is your exposure.
2. Decide what you're happy to hand to AI. Be explicit that answer-flow questions can go to the model. This removes guilt and frees your experts.
3. Pick two rituals from the table and calendar them this week. Start with "show your reasoning" reviews and named mentors for new hires. Don't launch six.
4. Make teaching part of senior roles. Add it to role descriptions and reviews so mentoring competes fairly with delivery. Our page on mentoring skills is a useful starting competency list.
5. Watch one lead indicator. Count cross-team questions asked in open channels, or ask a quarterly pulse question: "Who did you learn something from this month?" A falling number is your early warning.
6. Revisit in two quarters. If juniors can explain their reasoning and veterans still feel needed, the rituals are working. If not, tighten the time protection before adding anything new.

The test is simple. Ask a newer team member who they'd go to if the AI tools were switched off for a week. If they can name two people and describe how each one thinks, your team is still learning from itself.

Learn More

• Why Peer-Led AI Programs Outperform Top-Down Training Rollouts
• Does AI Make Distributed Teams Easier or Harder to Lead?
• How Leaders Should Talk to Their Team About AI and Job Security
• Is Experience Still Worth Paying For When AI Narrows the Skill Gap?
• The New Performance Review: How AI Changes How You Measure People
• Psychological Safety: Why It Matters and How to Build It
• Mentoring: How to Be an Effective Mentor