Two Harvard professors explain why compliance and adoption aren't the same thing
Employees who nod along in AI rollout meetings and then avoid the tools aren't being difficult. They're protecting who they are at work.
That's the argument from Das Narayandas and Shunyuan Zhang, professors at Harvard Business School, whose research on why employees resist AI reframes the biggest obstacle to adoption as something other than a skills gap or weak training. The real obstacle, they say, is that AI can threaten how capable people see themselves at work, and that threat is what drives them to appear on board without ever really being on board.
Narayandas has watched the pattern repeat for decades, back to the first sales automation software, when salespeople refused to enter their data for fear it would expose how they actually won deals.
"When a technology disrupts your identity, the way you react is not rational. It's not an economic decision," Narayandas said. "You feel threatened, your existence is threatened, and you decide, you know what, I will smile, but I will not adopt."
The researchers call this symbolic adoption. A recent report found that nearly a third of employees are sabotaging their company's AI strategy in at least one way, even as most leaders report far less resistance than actually exists.
In practice, it comes down to this: people say yes and mean no.
Three ways AI shrinks a job
The research identifies three ways AI threatens professional identity, all of which surface in workflows people teams will recognize.
The first is role compression. An employee keeps the title and the salary, but the parts of the work that made them valuable get absorbed by the machine.
"You can still have the job, your title can still be the same, but what you do and what makes you valuable is different," Zhang said. "I feel that my role becomes smaller, because if AI can do all of this, then what is the expertise and all the professional training that makes my role, my job, me, valuable?"
Narayandas points to a scenario close to home for HR. A recruiter still holds the job, but an AI system builds the shortlist, runs the first round of evaluation, and recommends who to advance.
"At some point soon, the recruiter begins to wonder, what is my expertise here, because the AI seems to be doing everything," he said. "It's like being the sovereign, the king or the queen, the royalty, but having a kingdom without subjects."
The second threat, the researchers say, is control shift. Here the work itself doesn't disappear, but the authority to decide does, moving from the person to the algorithm.
"Without AI, you as a recruiter, you would go through all the candidates, you know their profiles, and you pick the top ones and you recommend, these might be the top match," Zhang said. "And now AI is doing that. The control of making a decision of who is the top one shifts from you to AI."
The third is span erosion. Here a person keeps their decisions and their title, but their reach across the team, the budget, or the process they used to oversee starts to shrink. Zhang describes a manager pulled into a decision only when the system flags a problem.
"You're still involved in that decision, but your influence now is really shrinking to specific points," she said.
Why mandates backfire
Executives who force the issue often make things worse, Narayandas says, especially with the top-down declarations that pair sweeping AI plans and a headcount target in the same breath. The pressure to move fast has only grown as AI adoption accelerates across business functions.
"A lot of the CEOs are making claims and saying, we're going to do AI, or we're going to eliminate so many jobs," he said. "That actually creates a deer in the headlight mentality in people."
Narayandas illustrates the mentality with an old joke about two people running from a lion. One asks the other how fast he can run.
"He says, no, I don't need to. As long as you're the one that the lion is going after, I'm safe," Narayandas said. "People have that mentality, that it looks like that person is more likely to lose. So it becomes more a question of attrition, and hoping that you're not the one who goes."
Turning AI into an ally instead of a threat
The fix, the researchers argue, is a set of what they call identity-compatible advantages. The starting point is reframing the technology itself, as intelligence that augments the person using it.
"You can only help them if you say, look, 'you are going to work smarter, not work harder,'" Narayandas said. "You've got to create meaning in their job. The HR function has a very important role here in redefining the purpose for individuals, not just of the company."
Zhang says the work has to start with how a role is defined, whether that's a new hire coming in or a current employee being reassigned, and that a fancier job title on its own changes nothing. The shift from automation toward genuine role redesign is where she sees real transformation take hold.
"You can say someone is not a salesperson but a strategic consultant. Changing a title, it's nothing," she said. "You need to redefine the role, redefine the job. What remains human, what remains AI, and what training the company will provide."
Underpinning all of it is a change in how leaders regard their people. Narayandas frames it as moving employees off the profit-and-loss statement and onto the balance sheet.
"When you think of employees as cost, costs show up in your profit and loss, and you manage by minimizing them," he said. "But change the whole thing and think about employees as talent, and they are basically your assets. The way you manage assets is you nurture them."
That means naming the hard truth about job losses rather than dodging it, while making a credible commitment to the people who stay. A recent Kyndryl report found that the organizations investing in workforce readiness are pulling further ahead on AI, while those leaving employees to figure it out on their own see adoption stall.
"There is no doubt that there will be jobs that are lost. But that happens every time technology advances," Narayandas said. "Those that we keep, we are expecting more from. We're going to help them do more."
For all the talk of algorithms and automation, Narayandas says the real challenge is still a human one.
"The machine doesn't need to be impressed or convinced or motivated," Narayandas said. "But the human needs to understand how they can benefit."
