Tuesday, May 27, 2025, at the Baker Library at Harvard Business School on the campus of Harvard University in Boston, Massachusetts, USA. A recent study conducted by Harvard Business School’s Institute for Digital Data Design examines where AI is most effective at improving productivity and performance, and where humans still have an advantage.
Bloomberg | Bloomberg | Getty Images
Data from Anthropic shows that the adoption of AI in the workplace is at an all-time high, but just because an organization is using AI doesn’t mean it’s effective.
“No one knows the answer, even though many people say they do,” says Jen Stave, principal operator at Harvard Business School’s Digital Data Design Institute (D^3). As much of the business world explores where AI can best be deployed, the team at D^3 is researching where technology is most effective at improving productivity and performance, and where humans still have an edge.
Collaboration in the workplace has long been the norm for innovation and productivity, but AI is changing that. Individuals equipped with AI perform at the same level as teams without access to AI, according to a recent study co-authored by D^3. procter and gamble find. According to the study, “AI can replicate certain benefits typically gained through human collaboration, potentially revolutionizing the way organizations structure teams and allocate resources.”
Think about AI-powered teams, not just AI-powered individuals.
While AI-equipped individuals show significant improvements in factors such as speed and performance, strategically selected AI-equipped teams have unique advantages. Considering the quality of outcomes, the best and most innovative solutions come from AI-enabled teams. Although this study relies on AI tools that are not optimized for collaboration, AI systems dedicated to collaboration could further enhance these benefits. In other words, simply replacing humans with AI may not be the solution companies expect.
“If companies are actually thinking about role change and not only leaning into it, but also needing to protect human jobs, and if that’s our competitive advantage, then we need to even add some roles in that area. That, to me, is a sign of a very mature way of thinking about AI,” Stave said.
The D^3 experiment at P&G also shows that AI integration can significantly reduce the gaps that exist between pockets of domain expertise in an organization. For example, having a knowledge base at hand can make any team’s work more universally useful, beyond a single team such as human resources, engineering, or R&D.
Lower-level workers benefit more, but this is a double-edged sword.
Another experiment, D^3, conducted with Boston Consulting Group showed that AI makes results more homogeneous. “Humans have more diverse ideas, and people using AI tend to generate more similar ideas,” Stave said, acknowledging that companies with the goal of standing out in the market should focus on human-driven creativity.
Performers in the lower half of the skill spectrum show the greatest performance improvement (43%) when equipped with AI compared to performers in the upper half of the skill spectrum (17% increase in performance). Both outcomes are significant, but entry-level employees reap the greatest benefits.
But for low-skilled workers, it’s a double-edged sword. For example, if AI allows juniors to do their jobs more efficiently, senior-level workplaces may be less likely to delegate tasks to juniors, leading to a lack of training that will negatively impact future performance. Companies need to carefully consider what they delegate and what they don’t, with the future of the company in mind.
Human managers are not equipped to supervise AI agents. they need to learn
Stave says it’s “definitely going to happen” that humans act as managers for a series of AI agents, but the scaffolding to do so effectively and with minimal negative impact just doesn’t exist. Stave had this experience herself, which contrasted sharply with her management and leadership education. “You learn how to manage with empathy and understanding, how to maximize human potential,” she said. “I was trying to personally build and manage all these AI agents. It was a fundamentally different experience.”
Additionally, Grammarly CEO Shishir Mehrotra said that entry-level employees could become new managers (managed by AI agents rather than people), but it hasn’t really been proven that junior employees are native to enterprise AI or have the managerial capabilities. “We want AI to give humans more opportunities to grow. The challenge I have is that younger employees step in and assume they know how to do it right away,” Stave said.
She added that the companies that are getting real value from AI implementation are those that are working on process redesign. Rather than relying on AI note-taking to save time, focus on where AI can help and where humans can win. “The tools are very easy to purchase and implement,” she said. “It’s really hard to actually redesign an organization because when you do that, you get caught up in internal empires and power struggles.”
But it’s still worth the effort, she says.
