Stanford HAI 研究：想让员工拥抱 AI，别把它包装成效率工具
Want Employees to Embrace AI? Stop Selling It as a Productivity Tool
Stanford HAI 报道 Karunakaran 等人对律所、广告公司和 IT 服务公司的 AI 部署研究，发现把 AI 介绍为生产力工具时员工不愿使用，介绍为工作丰富化手段时使用意愿明显提升。在一家律所的对照观察中，以丰富工作为目标引入 AI 工具 LawBot 的部门，使用量比强调提效的部门高 58%，尝试新用法多 70%。研究还强调配套培训、探索时间和调整考核指标的重要性。
研究给出一个可操作的管理方法，把 AI 介绍为工作丰富化并配套培训和时间，比强调提效更能推动员工真正使用。
Want Employees to Embrace AI? Stop Selling It as a Productivity Tool
Date
October 07, 2026
Topics
Workforce, LaborGenerative AIEducation, Skills
How managers introduce and support AI determines whether workers engage with or resist the technology.
Adoption of generative AI tools has been swift. In 2023, 55% of organizations worldwide reported using AI. Two years later, that figure hit 88%. Yet despite this rapid growth — and the sizable investment it represents — employee feedback on the technology has been mixed.
“How these tools are framed for employees when they’re rolled out is an important part of this story,” says Arvind Karunakaran, an assistant professor in the School of Engineering at Stanford and a faculty affiliate at the Stanford Institute for Human-Centered AI (HAI). “Even more important is whether managers follow through on the claims that they make. If they say AI is for job enrichment, then are they committing resources to make this true?”
Karunakaran, together with Roshni Raveendhran of the University of Virginia and Tami Kim of Dartmouth College, studied AI deployments at a law firm, an ad agency, and an IT services firm to better understand what drives employee adoption. They found a consistent pattern: When AI was framed as a productivity tool, employees tended not to engage with it. However, when it was presented as a way to enrich their jobs, workers were motivated to figure out how best to use it — assuming they were given the time and opportunity to explore.
Making Work More Meaningful, Not Just Efficient
The researchers conducted two workplace studies. In the first, they interviewed 184 employees at the three companies that had adopted generative AI tools. Across administrative, creative, and technical roles, employees who were required to use AI described feeling that their work had been stripped of meaning. That loss, in turn, sapped their motivation to learn new skills. “I’m hesitant to learn new skills related to my job,” one explained, “because it feels like whatever I learn today will be outdated by next month.”
The most revealing evidence emerged from the second study, in which the team spent nearly two years at the law firm observing two divisions that were virtually identical in practice area, size, and pay and promotion structure. Administrators granted both groups the same firm-wide license for an AI tool — dubbed “LawBot” by the researchers — designed to help draft contracts and nondisclosure agreements. The critical difference between the divisions was how the managers introduced the tool.
In the first division, the manager pitched LawBot to paralegals as a way to draft documents faster and shorten turnaround times. After using the tool, paralegals reported feeling like “just another cog in the machine” and used it only for existing tasks rather than exploring novel applications.
In the second division, the manager asked paralegals: Which tasks do you find most boring, and what have you always wanted to do but never had time for? Could LawBot help take care of the routine tasks and free up time for more interesting work? Given time to play around with LawBot, the paralegals began using it for legal research and case analysis, exchanged tips about what worked, and lobbied to attend legal strategy meetings with the attorneys. Paralegals in this division used LawBot 58% more than their peers in the other division and experimented with it 70% more.
“When AI tools are introduced, managers are often obsessed with productivity gains,” Karunakaran says. “But if you aren’t also working to answer how people’s jobs are going to improve or be enriched, the initiative will most likely backfire.”
How managers frame the reason for introducing AI plays a decisive role in its successful deployment, but it’s not the only factor. Managers in the second division offered the paralegals formal training on LawBot, weekly knowledge-sharing lunches, Q&A sessions with the vendor, and informal mentorship on more complex tasks, such as legal research. In addition, time each week was set aside for paralegals to explore and learn about the tool on their own.
“You need to put this scaffolding in place to make the commitment feel credible,” Karunakaran says. “Without that, employees recognize it as little more than cheap talk.”
Measuring Value, Not Volume
As AI reshapes job responsibilities, the way employees are evaluated must also change. Paralegals, Karunakaran notes, have traditionally been judged by their annual or quarterly case volume.
“But that was not the right metric for the new type of work they were doing with AI,” he says. Paralegals who used the technology to conduct in-depth legal research and case analysis produced high-value work, but it was also more time-consuming, which meant they took on fewer cases. “So, the metrics need to be realigned to suit the new job profile and task portfolio.”
The wrong metric can also trigger counterproductive behavior. Some companies track employee AI token consumption, or the amount of text processed or generated, and equate heavy use with value. But that strategy encourages workers to use AI for its own sake — generating cat memes to hit a quota, as Karunakaran puts it — rather than deploying it for legitimate business objectives.
The Best Jobs for AI
One more factor contributing to successful AI adoption is whether a particular job consists of strongly or weakly bundled tasks.
Sales roles, for example, are made up of strongly bundled tasks like researching customers, building relationships, negotiating, and closing deals. “You can’t randomly unbundle and rebundle these,” Karunakaran says. A company might deploy AI for customer research, but the salesperson remains involved throughout the process. “The salesperson still needs to understand the customer and be there to close the deal, so it’s a strongly bundled job.”
Conversely, jobs consisting of weakly bundled tasks — such as paralegal work or medical transcription, which includes disparate tasks related to document management — may be more vulnerable to AI automation, but their modular nature makes adoption far easier.
Regardless of how far or fast AI is deployed, its success in the workplace will hinge less on the technology itself and more on the reason for its deployment, the metrics used to reflect real value, and the time and resources afforded employees to experiment and engage with it in meaningful ways.
“Though the promise of AI is that it will take care of routine tasks and free people up for higher-level work, what often happens is that managers see the time savings and raise expectations for how much of the same work people should do,” Karunakaran says. “If managers want AI to enrich people’s jobs and create organizational value, they have to create room for employees to learn, take on new responsibilities, and be recognized and rewarded for the value of those contributions. Realizing this possibility takes a commitment to developing people alongside a commitment to adapting the technology.”
Share
Link copied to clipboard!
Contributor(s)
Dylan Walsh
Related News
The Tests That Grade AI May Be Getting It Wrong
Andrew Myers
Sep 25, 2026
News
Benchmarks — the standardized tests that rank AI models on safety, bias, and reasoning — drive markets and shape regulation. New Stanford research finds they often don't measure what they claim to.
News
The Tests That Grade AI May Be Getting It Wrong
Andrew Myers
Generative AIPrivacy, Safety, SecurityFoundation ModelsSep 25
Benchmarks — the standardized tests that rank AI models on safety, bias, and reasoning — drive markets and shape regulation. New Stanford research finds they often don't measure what they claim to.
Can AI Be Slowed Down? Stanford HAI Experts Weigh the Risks, Rules and Race Ahead
Sep 22, 2026
News
In a new series called Prompt Response, Stanford’s Surya Ganguli, Diyi Yang and Rob Reich examined emergent agent behavior, recursive self-improvement, independent evaluation and whether a kill switch can make advanced AI safer.
News
Can AI Be Slowed Down? Stanford HAI Experts Weigh the Risks, Rules and Race Ahead
Privacy, Safety, SecurityRegulation, Policy, GovernanceGenerative AISep 22
In a new series called Prompt Response, Stanford’s Surya Ganguli, Diyi Yang and Rob Reich examined emergent agent behavior, recursive self-improvement, independent evaluation and whether a kill switch can make advanced AI safer.
AI Companions May Worsen Loneliness for Vulnerable Users, Stanford Study Finds
Alexander Gelfand
Aug 04, 2026
News
Stanford research finds that users with limited social networks who seek emotional support from AI companions experience lower well-being.
News
AI Companions May Worsen Loneliness for Vulnerable Users, Stanford Study Finds
Alexander Gelfand
HealthcareGenerative AIAug 04
Stanford research finds that users with limited social networks who seek emotional support from AI companions experience lower well-being.
Navigate
Participate
Stay Up To Date
Get the latest news, advances in research, policy work, and education program updates from HAI in your inbox weekly.
© Stanford University. Stanford, California 94305.
Pause Media
Want Employees to Embrace AI? Stop Selling It as a Productivity Tool | Stanford HAI
来源：Stanford HAI News · hai.stanford.edu