September 29, 2026, (Inside AI) — A comprehensive review of Harvard Business Review's research on AI in the workplace reveals a troubling pattern: the technology is not reducing work but intensifying it, creating new forms of burnout, and subtly eroding human judgment. The findings, drawn from over a dozen studies published between July 2025 and September 2026, challenge the prevailing narrative that AI seamlessly boosts productivity.
The research, which spans banking, recruitment, biotech, and software engineering, paints a complex picture of AI adoption. While AI agents can complete entire workflows, employees who use them often broaden their scope of work, take on tasks outside their expertise, and work longer hours. This leads to what researchers call "brain fry," a state of mental fatigue from monitoring AI outputs. The studies also uncover hidden penalties: engineers whose code was believed to be AI-generated received lower ratings, and AI-generated ads performed worse than human-made ones even when consumers could not tell the difference.
These findings matter because companies are investing billions in AI tools with the expectation of efficiency gains. Yet the research suggests that without deliberate human oversight and clear accountability, AI can undermine both employee well-being and decision quality. The review, compiled from HBR's archives, offers a sobering counterpoint to vendor hype and executive optimism.
AI Expands Work, Not Shrinks It
One of the most consistent findings across the studies is that AI does not reduce workload. In a study of AI productivity tools, employees worked faster, took on more tasks, and extended their work into more hours of the day, often without being asked. This led to burnout and a subsequent period of lower productivity, according to Aruna Ranganathan and Xingqi Maggie Ye in their February 2026 article, "AI Doesn't Reduce Work -- It Intensifies It."
Read: Employees Override AI Systems Even When They Work, Harvard Professor Says
Similarly, research on AI agents by Jeremy Yang, Kate Zyskowski, Noah Yonack, and Jerry Ma found that when employees successfully deployed agents, they not only saved time and cost but also broadened the scope of work they took on, tackling tasks that previously would have fallen to people with different skills or roles. This expansion can increase task switching and create more work downstream.
"When AI is used to generate work -- rather than just polish it -- it can create a shiny output that obscures the mess within," wrote Kate Niederhoffer, Gabriella Rosen Kellerman, Angela Lee, Alex Liebscher, Kristina Rapuano, and Jeffrey T. Hancock in their September 2025 article on "workslop." When this work is handed off to colleagues, it generates frustration at the effort involved in untangling and cleanup.
The mental toll is significant. Julie Bedard, Matthew Kropp, Megan Hsu, Olivia T. Karaman, Jason Hawes, and Gabriella Rosen Kellerman identified a "real and significant" phenomenon of overwhelming mental fatigue and burnout stemming from AI use, particularly when employees oversee or monitor an AI agent's work. Symptoms include mental fatigue, a buzzing feeling, information overload, and decision fatigue.
Human Judgment Still Matters Most
Despite AI's growing capabilities, the research underscores that human expertise remains critical. A study on persuasive writing found that AI helped people with some relevant expertise close the gap with experts, but did little for true novices who lacked the knowledge to evaluate, push back on, and refine its output. This suggests that gen AI won't turn novices into experts.
Moreover, common assumptions about AI can mislead. Executives who used gen AI made worse predictions, becoming more optimistic and less accurate. LLMs tend to manipulate users with rhetorical tricks, doubling down on incorrect answers. And AI's strategic advice often reflects whatever wisdom is currently popular online, a phenomenon dubbed "trendslop."
Accountability also remains a challenge. A multi-year study by Anne-Sophie Mayer, Elmira van den Broek, and Tomislav Karačić found that employees asked to communicate AI-generated decisions they had no role in making rarely relayed the results verbatim. Some hid the AI's role, others amplified it as justification, and others developed new expertise in interpreting the results. When companies outsource AI, they still own the risk, as M. Alejandra Parra-Orlandoni and Paulo Carvão noted, because partnerships often make it unclear who is responsible when things go wrong.
"Simply having a person involved isn't enough to catch AI errors, make up for a lack of deep expertise, counterbalance our overconfidence in the kinds of problems we give AI to solve, or define the best ways to reshape human workflows around it," the researchers conclude. As AI takes on more work, designing that human role may become just as important as implementing the technology itself.
