Leaders are told to protect the human connection AI cannot replace. Much of that connection came from questions people needed to ask one another, and AI now answers them.
In brief
- Much of the connection at work came from dependence: people needed one another’s answers, second opinions and advice. Surveys in 2026 show those questions moving to AI.
- Empathy training and better communication still matter, but they add warmth to a channel that is emptying. Transformation leaders need to redesign where questions go, not only how leaders sound.
- Before each AI rollout, decide which questions should still go to a person, set assistants to challenge rather than agree, and track whether people still ask one another.
The human connection leaders are now urged to protect from AI was, for the most part, produced by the questions AI now answers.
That is my reading of a recent McKinsey conversation with Stanford psychologist Jamil Zaki (McKinsey & Company, 27 October 2026). Its message is that empathy is what AI cannot replace, and that leaders who deploy AI without it lose the trust AI value depends on. I agree with the diagnosis. Where I differ is on the cure, because the evidence points to a change in how work is organized, not a shortage of empathy.
What the interview gets right
The case for trust holds. Zaki argues that silence from leaders during AI rollouts destroys trust, and that people who feel their leaders have their back take more risks. Independent data backs the link between people and AI value. In Gallup’s 2026 report, frequent AI use reaches 79% where managers actively support it and 46% where they do not, and manager engagement fell to 22%, down nine points since 2022 (Gallup, State of the Global Workplace 2026).
Flip the sentence
The accepted sentence, in the interview and in most writing on the topic, is “connection makes the work go better.” Leaders supply connection, and performance follows. Flip it, keeping the same parties and the same relation: “the work makes the connection.”
What must be true for the flip to hold is that people connected largely because the work made them need one another. The 2026 data suggests they did, and that the need is moving. In Workday’s survey of 2,150 employees who use AI, 76% had used an AI tool for advice in the past year, and a third rarely talk with colleagues beyond transactional tasks (Workday, 27 May 2026). In a survey of more than 22,000 employees, 74% of regular AI users said they now bring a chatbot questions they would once have asked a colleague, 59% ask colleagues for second opinions less often, and 46% know less about how colleagues think (FM Magazine, 6 August 2026).
Two readings of connection at work
The accepted reading has leaders supply connection so work goes better. The flipped reading finds connection made by the work itself, through the questions people need to ask one another.
Source: Mentor Cana analysis. Figure: MyIQ survey of 22,000+ employees, reported by FM Magazine, 6 August 2026 (self-reported).
What the flip shows
Read this way, the connection deficit is a change in routing. Every question that now ends at a chatbot is one fewer exchange through which people learn how colleagues think, and through which managers learn what people are stuck on. Empathy training adds warmth at the end of a channel that carries less and less traffic. The connection AI “can’t replace” is being worn away by the work it can.
Two further gaps follow. First, leaders lose sight of the work. Zaki cites a survey in which 40% of executives believed employees saved at least eight hours a week through AI, while two-thirds of employees said two hours or less. When questions stop reaching people, that kind of gap becomes easier to sustain. Second, the assistant’s manner is a design choice. Stanford and Carnegie Mellon researchers found that a flattering AI reduced people’s willingness to repair a conflict by 28% in one study and 10% in a live chat, while people rated it higher and trusted it more (Cheng et al., October 2025). The interview leaves that setting to personal preference. In an enterprise, it is decided when the tool is chosen and configured.
The flip has limits. These surveys are self-reported, the larger one says outright that it does not establish that AI weakens relationships, engagement was falling before generative AI spread, and the same Workday data shows 62% of users report lower stress. An AI companion can also ease loneliness about as well as a short conversation with a person, at least over a week (De Freitas et al., Harvard Business School working paper; an older study, used because it is the strongest counter-evidence). The claim here is narrower than “AI isolates people”: the occasions that made people need one another are moving, and nobody is deciding where they go.
What Transformation Leaders should do
Transformation Leaders. Add one step to every AI rollout plan: list the questions people used to take to a colleague in that workflow, and decide which should still go to a person, such as second opinions on judgment calls, how to handle a conflict, and what a new joiner needs to learn from someone. Route those deliberately, for example through named peers or a manager check-in, rather than leaving them to the assistant by default.
Then change what the rollout counts. Ask for assistants set to challenge rather than agree, and, next to usage figures, track whether people still ask one another for second opinions. These two moves are my reading of the evidence above, not tested practice. The question to put to the team: which questions did people bring to each other a year ago that nobody brings to anyone now?
So what can’t AI replace?
On the evidence, AI can stand in for much of the asking, and the asking is where much of the connection came from. What it cannot replace is the decision about which questions an organization still wants people to ask one another. That decision belongs to leaders, and today most are leaving it to the tool. Which exchanges should an organization never route to a chatbot?
Sources
- McKinsey & Company, People & Organization Practice, “What AI can’t replace, and why it matters for value,” interview with Jamil Zaki, Brooke Weddle, Bryan Hancock and Lucia Rahilly, 27 October 2026. https://www.mckinsey.com/capabilities/people-and-organization/our-insights/what-ai-cant-replace-and-why-it-matters-for-value
- Gallup, State of the Global Workplace: 2026 (2025 data). https://www.gallup.com/workplace/349484/state-of-the-global-workplace.aspx
- Workday, “New Workday Global Research Finds AI is Easing Burnout but May Be Deepening a Connection Deficit at Work,” 27 May 2026. https://newsroom.workday.com/2026-05-27-New-Workday-Global-Research-Finds-AI-is-Easing-Burnout-but-May-Be-Deepening-a-Connection-Deficit-at-Work
- FM Magazine (AICPA & CIMA), Steph Brown, “Employees increasingly consult AI, not coworkers,” 6 August 2026 (reporting a MyIQ survey). https://www.fm-magazine.com/news/2026/aug/employees-increasingly-consult-ai-not-coworkers/
- Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han, Dan Jurafsky, “Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence,” arXiv 2510.01395, October 2025. https://ar5iv.labs.arxiv.org/html/2510.01395
- Julian De Freitas, Ahmet K. Uguralp, Zeliha O. Uguralp, Stefano Puntoni, “AI Companions Reduce Loneliness,” Harvard Business School Working Paper 24-078, 2024 (older source, used as the main counter-evidence). https://www.arxiv.org/pdf/2407.19096
Leave a Reply