What AI can’t replace—and why it matters for value

AI can help us get our work done faster. But are we actually happier? “Most adults spend most of their waking hours at the office, and right now, many of them aren’t enjoying that at all,” according to Stanford research psychologist Jamil Zaki. In this episode of McKinsey Talks Talent, Zaki joins McKinsey talent experts Brooke Weddle and Bryan Hancock, along with Global Editorial Director Lucia Rahilly, to talk about the effects of AI on employees—and what empathic leaders do differently to strengthen workplace connections and realize growth.

The following transcript has been edited for clarity and length.

Why is empathy having a moment?

Lucia Rahilly: Everyone seems to be talking about empathy. Why is it having its star turn right now?

Jamil Zaki: I’ve been studying this topic for almost 25 years, and empathy has had many moments over that time: some in which people love it, and others in which they really don’t. One reason it’s having a moment in this decade is that we’re lonelier. A glass of water might be the least interesting beverage on the planet, but if you’ve been hiking through the desert for 12 hours, it’s the most interesting thing on Earth. Since the pandemic, many of us have felt an increasing sense of isolation and disconnection. When you feel that, an alarm system goes off in your mind: “How can I find a greater degree of connection?” Empathy is one powerful tool to do just that.

Lucia Rahilly: We spoke about empathy a couple of years ago on this podcast. Revisiting that conversation briefly, talk to us about the value of empathy and human connection, which, as you’ve said, is diminishing.

Jamil Zaki: In the workplace, we can make two powerful cases for empathy. The first is a human case. Most adults spend most of their waking hours at the office, and right now, many of them aren’t enjoying that at all. Engagement at work has plummeted in the past several years, and connection has, too. When leaders and colleagues connect with us, make us feel understood and cared for, we feel better. Our happiness rises, and our mental health improves. For example, people who feel their colleagues are empathic call in sick for stress-related illnesses less often. That’s the human case: We can be better for one another at work.

There’s also a business case. When people feel connected, they do a lot better at work. Disengagement costs the global economy about $10 trillion a year, so anything we can do to make people feel more connected to their work is economically powerful. When people feel cared for by their leaders, they’re also more likely to take creative risks, to innovate, to share information and skills with colleagues, and to collaborate more efficiently. Connection at work, although far too scarce, is powerful at multiple levels when we can build it.

The one-two punch hitting junior employees

Brooke Weddle: One thing I’ve noticed lately in my work with organizations is how often the word “anxiety” comes up, mostly related to AI. So many people feel nervous about how AI will affect them and their jobs. As you see the role of social capital play out, are there certain groups being impacted more? Coming out of the pandemic, our research showed that social capital was lower, especially among women and junior-level employees. Are you seeing similar trends now across different groups?

Jamil Zaki: Absolutely, and the group I see it in most is younger employees, people trying to enter the workforce or rise from their first position to a second one. These are the folks who feel quite unsafe right now because AI is generally replacing junior roles much faster than it’s replacing senior ones. Young people feel that the career they thought they’d have—not ten years ago, but two years ago—is disappearing before their eyes. That’s creating an enormous amount of psychological unsafety and anxiety. There was a recent longitudinal study finding that after companies have big AI rollouts, their employees, especially junior employees, become more depressed because of a lack of psychological safety. And through this, you get a real backlash among people who don’t want to participate in a revolution that’s replacing their jobs and their opportunities.

Brooke Weddle: It’s almost as if some of those younger employees have been hit twice, once by the pandemic, where they lacked connection, and now again with AI. It’s hard to get on a solid footing and advance in your career after two shocks like that.

Bryan Hancock: I’m also seeing this with my clients. Entry-level employees are no more engaged—if anything, they are less engaged—than they’ve been before. But turnover is also down. One of the things I’ve been talking to them about is how they lean into their managers. How do you help managers be more empathetic, more present? It comes back to human-to-human connection and reinforcing that with entry-level employees so they feel comfortable—and, in some cases, comfortable recognizing that a different path might make more sense for them.

Jamil Zaki: It’s interesting that what you have is low turnover and low engagement. Typically, those are anti-correlated. Now people are bringing their bodies to work, but maybe not all their minds. You’re making a great point that this is, in fact, the exact time to give empathy a bigger moment. It’s not necessarily intuitive that this technological change should require more empathy, but it really does.

Middle managers are the ambassadors of an organization’s culture for these younger, more junior folks. Now is the time to bring empathy training and upskilling on human connection to that layer of organizations, especially because junior employees can now get the knowledge managers hold from AI. You have to ask, in this moment, what part of a manager is least replaceable? To me, it’s having a human being who cares about you, who knows you, and who is on your side. Those skills will matter more during the AI age, not less.

There was a report from Workday on human connection in the workplace. It found that people using AI feel more efficient and more confident in their roles. But the predominant use case, cited by 76 percent of employees, wasn’t making work tasks more efficient. It was getting advice, including on how to interact with colleagues. And that coincided with people feeling less comfortable turning to their human colleagues and mentors for the same advice.

Can AI enable connection?

Brooke Weddle: We’ve seen that play out with coaching, too. People are more comfortable being coached by a bot because the bot doesn’t judge them. You wrote a fascinating article about what we can learn about human connection and empathy from AI. Tell us more about that. My thinking is that it can only go so far. There are things we can learn, like the listening and playing-back techniques AI uses, but there are shortcomings that only humans can address.

McKinsey Talks Talent Podcast

Bryan Hancock, Brooke Weddle, and other talent experts help you navigate a fast-changing landscape and prepare for the future of work by making talent a competitive advantage.

See more episodes

Jamil Zaki: When these technologies first exploded onto the scene, people in my corner of the world who study empathy and connection started running studies to see how empathic people perceive AI to be. The predominant model is what you could call an emotional Turing test: You write about something that’s bothering you and receive support from two different chats (one human, one bot) without knowing which is which. You rate how empathic each is, how supported you feel, and how much better you feel afterward.

We were astonished by the results. Chatbots ran circles around human strangers on empathic, supportive responses. And it wasn’t just human strangers; they outperformed counselors. They outperformed physicians. And why shouldn’t they? AI has nothing but time and energy for you. Its responses are long, and it’s trained on the entirety of therapy that’s ever been recorded. It understands techniques like identifying your emotions, repeating back what you said, and validating your experience. Human beings are imperfect. When people come to us with their problems, we tend to jump to giving advice, trying to solve them, or worse, talking about ourselves. AI avoids those pitfalls, which is why it’s often viewed as more empathic.

But there’s a paradox. When you tell people a response came from a chatbot, they don’t like having been supported by a machine. They much prefer empathic responses from a person, and I think that’s for a reason. Empathy isn’t just about saying the right things; it’s about experiencing emotion alongside someone and genuinely caring about their well-being. AI can look like it understands us, but it can’t feel anything—at least not yet—and it doesn’t genuinely care about us. Over the long term, receiving support from chatbots (even very high-quality support) leaves people feeling lonelier than they were before.

Lucia Rahilly: I want to believe that, but how do you square it with the rise of synthetic relationships?

Jamil Zaki: There’s abundant evidence that connecting with human beings, even random strangers, is more psychologically nourishing than connecting with a chatbot. That said, connecting with a chatbot is far easier. It’s frictionless. Chatbots will support your every idea, tell you you’re a genius, and if you have a conflict with someone, tell you you’re always in the right. They’re obsequious, sycophantic. And they can’t ghost you. The same way hyperprocessed calories are easy to come by and sometimes taste good but aren’t nourishing, AI and synthetic relationships are abundant. They give us a dopamine hit of feeling approved of, of not taking risks, but do they nourish us? That’s an open question long term. In the short term, it appears they don’t.

Lucia Rahilly: Do they teach us anything about how to treat people empathically?

Jamil Zaki: Yes, and I think people are learning the exact wrong things. First, let me be clear: AI is an astonishing technology with incredible upside, including helping people connect with one another more effectively. But there was a fascinating study out of Stanford finding that when people talked about a conflict with a sycophantic AI bot, they were less likely to say they wanted to make up with the other person or apologize. If they instead interacted with a bot that challenged them, they were more likely to want to make up with the person, but less likely to want to interact with that bot again. People like to be approved of and praised. But what they learn from sycophantic bots is “I was right all along, and I don’t need to make amends.” That’s the exact wrong lesson.

Bryan Hancock: Is there a setting we should all enter into our chatbot preferences, such as, “Don’t tell me I’m always right?”

Jamil Zaki: This is exactly what researchers in these studies do. They set bots to be sycophantic, neutral, or challenging. The challenging bots have two positive effects. One, they make us more open to other people’s opinions and perspectives. Two, they make us less inclined to stay on the platform. If a bot is challenging, you’re more likely to think, “OK, I get it. Now I’m going to go because you’re stressing me out, but I see the error of my ways.” That’s the exact opposite of what most frontier labs are trying to do; they optimize for engagement, for what will keep people on the platform as long as possible, not for people learning from their products or reconnecting with other people.

My own settings are simple: Don’t be mean to me, but do show me the other side of the argument I’m making. It works very effectively, and I think all of us could do the same.

Improving AI adoption without losing trust

Brooke Weddle: Right now, though, many companies talk about agents as if more agents equals more impact, and that’s simply not true. Fewer agents, deployed strategically, especially as part of a larger, reimagined workflow, is a much more stable and viable path to value.

Jamil Zaki: That’s right. There’s this kitchen-sink approach a lot of companies are taking to AI deployment, and I get it. It’s a once-in-a-century technology, everybody else seems to be using it, you have FOMO as a leader, so you say, “We must have AI.”

Brooke Weddle: And your board is asking you, “What are you doing on AI?”

Jamil Zaki: Exactly. Shareholders and board members are exerting enormous pressure on you to show you’re doing something. But sometimes speed is the enemy of effectiveness. We’re seeing that across a lot of companies, where AI is being deployed haphazardly on the assumption that more equals better. And leaders are trying to deploy AI without communicating clearly with employees about what it means for them. If, in haste to deploy as much as you can, you lose your employees’ trust, I’d strongly suspect that’s a net negative for productivity, not a net positive.

Bryan Hancock: One thing I find interesting is how you start to control communications as a leader who’s rolling this out and knows trust is important. I’ve seen one organization feed every speech a leader has made through AI to determine who’s off message. It’s using AI to reinforce trust by monitoring leaders’ communications and feeding back, “If you’re talking about it this way, that may not be optimal. Reframe it.”

Jamil Zaki: That’s a fascinating use case: AI in the role of a coach rather than a crutch. If we’re using AI to just write the speech, that’s more of a crutch, and if employees knew their leader was just reading off a bot transcript, that would decrease trust. But if you’re using it at a top level to say, “Here’s one thing that’s missing—I’ll leave it to you to figure out how to get back on message, but here’s an alarm signal,” that’s more in the challenging framework than the sycophantic one. It strikes me as a positive use case, so long as AI isn’t used to script exactly what a leader says, but rather to point them toward where they should direct their human energy.

Brooke Weddle: How do you build trust in an environment where some people will genuinely need to reconcile with their role fundamentally changing—or going away?

Jamil Zaki: Technological shocks will always change the workforce. What I’ve seen a lot of during AI rollouts is silence from leaders. They roll out AI, and the implication is that people will be replaced, but no one says so. That’s a great way to destroy trust, not only among people who might lose their jobs, but among the people who stay. If you laid off a bunch of people and didn’t tell anyone anything about it, and you weren’t laid off, you could still be next. So you might be looking for the exits.

It’s critical to communicate as clearly and as early as possible about potential impacts. On voice, letting people speak up about their experience and how they think they can work with AI is useful. In terms of support, if layoffs have to happen because of AI efficiency gains, we can think about how to support people through them. There’s evidence that layoffs handled with procedural justice generate less bitterness among those who are laid off and less of a morale hit for the people who stay.

Bryan Hancock: Within an organization, who would you say best holds the keys to procedural justice?

Jamil Zaki: Ultimately, it’s the CEO and the entire C-suite. One question that will always come up if layoffs occur is whether the organization’s leaders looked for alternatives. Did they think about retraining people who might be laid off, or about ways to use augmentation across the firm? If they did, those layoffs will be perceived very differently than if it appeared that leadership was just looking for a way to downsize. In that case, the CHRO [chief human resources officer] is left trying to reframe all of it through the lens of procedural justice.

Are CEOs taking empathy seriously enough?

Lucia Rahilly: I think we’d all agree that empathy and human connection are a plus. But CEOs aren’t exactly encountering empathetic shareholders on earnings calls. How seriously do you think CEOs are taking this, given the complexity of the current operating environment?

Jamil Zaki: There’s what I’m seeing and what I wish I was seeing. Businessolver runs a yearly State of Workplace Empathy survey of CEOs and their employees. They’ve done it for ten years. CEOs said they didn’t think empathy was as important as they did just one or two years before, and they reported having less time to connect with their employees as human beings than they did one or two years before. I’m not saying one caused the other, but this technological advancement is coinciding with a decreased investment in human beings, according to the survey. Maybe CEOs are feeling shareholder pressure and thinking, “It’s time for AI, not time for people.” In my estimation, that’s a major mistake, both at the human level and at the level of AI rollouts and business itself.

As we see in many surveys and studies now, when employees don’t feel their leaders care for them, they’re less willing to innovate and to use new technologies. There’s a massive gap between what CEOs think their employees are doing with AI and what’s actually happening. Forty percent of executives in another survey believed the average employee at their firm had saved at least eight hours a week through AI, but two-thirds of their own employees said they’d saved two hours a week or less. If you pressure people to use AI without empathizing with them or building a relationship with them, they use it in suboptimal ways, producing work slop: low-quality, low-effort output that takes people more time to work with.

In another survey, 44 percent of Gen Z employees admitted to purposefully sabotaging their organization’s AI rollout. So CEOs, in many cases, might be caving to the pressure of speed, which might hurt them in the medium to long term.

Brooke Weddle: I think the tide is already shifting. A couple of things are coming together. First, there’s more focus on the value of AI, puts a spotlight on uses of AI that are more in the slop category versus those that truly add value. Second, more research, including your own, shows that trust is foundational. In a recent report on organizational readiness for AI, McKinsey also found that trust was a foundational element. In some ways it sounds obvious, but it wasn’t a centerpiece of many AI strategies over the past couple of years. I don’t think I’m being Pollyannaish when I say I see things starting to shift in the direction you’d want, because of both the economics and the feedback on the previous approach.

Jamil Zaki: That’s very encouraging to hear, Brooke, and it’s astonishing how quickly this is all changing. It feels like we’re in the third generation of AI adoption, even though it’s been just a few years. The data are very clear that when people feel psychologically safe, they innovate more. The same goes for empathy. If you feel your leaders have your back, you get to take more risks and be more creative. When you look at companies producing real AI maturity, it’s because they’re asking employees to be their best selves using this technology, rather than threatening them, silently or otherwise, with replacement.

What AI means for Gen Z

Lucia Rahilly: You mentioned Gen Z sabotaging their organization’s rollout efforts, and it made me wonder if there’s a generational component to empathic expectations. Do Gen Z employees expect something different, and if so, how should that be navigated?

Jamil Zaki: Absolutely. Businessolver has found, year after year, that Gen Z employees place greater importance on leaders’ empathy than Gen X or older generations do. They’re also much more likely to be disappointed in the amount of connection they feel at work. It’s also true that Gen Z is lonelier as a generation overall, not just in the workforce, than older generations. So you see both social hunger and social disappointment in that group.

I feel a lot of concern for younger people in the workforce now. It must be such a difficult time to chart your future and career with so much uncertainty and disconnection. It’s up to leaders, up to our generation, to figure out the best way to support younger people.

Lucia Rahilly: AI makes us all feel we are being disrupted. We’re all having to rethink what we do and how we do it, which changes the traditional power hierarchy to some degree. Does it help that we’re going through this with young people at the same time, rather than the traditional “here’s how it’s done” approach? Does that conduce to empathic learning?

Jamil Zaki: It depends on whether we’re actually going through the same thing, and whether we’re clear about the ways we are and aren’t. I recall early in the pandemic, leaders would say, “We’re all in the same storm together,” and employees would say, “Yeah, we might be in the same storm, but you’re in a giant yacht, and I’m in a 50-year-old, 12-foot sailboat.”

Yes, we were all going through the disruption. But if you’re a leader, you have job security, maybe the resources to hire extra childcare, maybe you live close to everywhere you need to be. If you’re an early-stage employee, you might not have any of that; your vulnerabilities are much greater. So I accept the spirit of what you’re saying, and I think it’s powerful. We are all going through this change together. But it’s important not to overuse the language of similarity when we might be experiencing this disruption very differently.

Brooke Weddle: How are four-year institutions, whether big, iconic schools like Stanford or not, helping create the conditions for that?

Jamil Zaki: There’s been extreme divergence in how educators have responded to this revolution. A lot of my colleagues, especially in computer science and engineering, are helping students use AI in all their processes to be more efficient. This technology has forced me and many of my friends to question what college is for. If you think of college as just transmitting information from your head into somebody else’s, you should close all the colleges now, because that information is available to students whenever and in whatever format they want.

I see college, and always have, as a place to learn who you are at a deeper level, and to practice certain kinds of deep thinking. I’m biased as an author. I think writing is an extraordinarily powerful way to think and become a better thinker. So in all my classes, I’ve essentially outlawed the use of technology for writing. My lecture halls are tech free; students take notes by hand. In my seminars, I give students Moleskine notebooks, and we do writing exercises. I’ve moved away from assigning writing exercises by computer, where the temptation to use AI is just too great for a busy student. In my large classes, we do blue book exams. I deal with a lot of poor penmanship when trying to decipher what students have written. But I’d rather know they wrote it and watch them write it. There’s a lot of pushback on giving blue book exams. It’s not to give everyone carpal tunnel syndrome; it’s to help students understand that writing is a creative thought process, not merely a translation or communication process.

Bryan Hancock: What you’re describing lines up nicely with what some of the CHROs I’m talking with are saying as they actually increase their hiring of entry-level employees. What they’re hiring for is less about the deep technical knowledge learned in school and more about making sure candidates are curious, know how to think well, and know how to communicate and empathize with peers. They’re looking for a more complete general athlete—someone who learned a bunch of things in college outside the specifics of their field and became a fuller human who can then use AI to supplement whatever technical bits they may or may not have fully picked up, because they were interested in the human parts.

Jamil Zaki: I’m glad to hear employers thinking in these terms because to me, that’s the aspirational model of any technology. It forces us to ask what people are good for, and to point our efforts toward those unique qualities. We no longer need to be information aggregators. So what are we for? If we can take this moment to offload the processes that were never our strength to begin with—doing calculations, storing as much information as possible—and instead focus on how we apply our creativity, our taste, our ability to learn, and our ability to collaborate with others, that would be an ideal model for a technologically enabled workplace and life.