Picture the meeting where your company approved its AI budget. Someone put up a slide promising that employees would reclaim hours every week. Everyone nodded. Now look at your team’s calendar, your inbox, and the faces in Monday’s stand-up. Did those hours come back? For many managers, the answer is no—and the numbers suggest their experience is not unusual.

Upwork surveyed 2,500 workers across the United States, United Kingdom, Australia, and Canada between April 16 and May 5, 2024. Among the 1,250 C-suite leaders surveyed, 96 percent expected AI to boost productivity. Yet 77 percent of full-time employees said AI had added to their workload. Same technology, same companies, opposite experiences. That gap now has a name: the AI productivity paradox. If you manage people—or manage managers—it lands on your desk whether or not you asked for it.

This article examines what the research says about why AI-related workloads keep climbing, how the workday becomes more crowded, and what leaders can do without banning the tools their boards just paid for. Think of it as the conversation I would have with you over coffee if you said your team looks tired while the dashboards look great.

What the Berkeley Researchers Watched Happen

In February 2026, Aruna Ranganathan and Xingqi Maggie Ye of the UC Berkeley Haas School of Business published their findings in Harvard Business Review under a blunt headline: AI does not reduce work; it intensifies it. They studied employee behavior after AI tools arrived and found a recurring pattern. People worked faster, took on tasks that once belonged to someone else—or that no one had pursued before—and extended their hours without being asked.

No one forced these changes, and that detail matters. A designer who can generate ten layouts in the time it once took to sketch two does not stop at two. An analyst who can draft a summary in five minutes starts producing summaries no one requested. The tool makes more work possible, and people who care about their jobs fill the space. Ranganathan and Ye call the result workload creep and trace its cost: cognitive fatigue, burnout, weaker decisions, and, eventually, turnover.

The early numbers look impressive, which is precisely the trap. Output rises in the first few weeks, leadership sees the chart, and the chart becomes the new baseline. Then quality slips and people leave. By the time the resignations arrive, no one connects them to the earlier rollout. If you have watched a strong performer quietly disengage after being handed every difficult project, you already know how this story ends. AI simply accelerates the handoff.

How the Workday Gets More Crowded

ActivTrak’s 2026 State of the Workplace report offers the most detailed view of what happens inside the workday. The company analyzed 443 million work hours across 1,111 organizations. AI adoption among those organizations rose from 53 percent two years earlier to 80 percent in 2025, allowing the report to compare workplace patterns as adoption increased rather than relying on a survey of enthusiasts.

Productive hours rose 5 percent while the average workday shrank 2 percent. On paper, that looks like the promise delivered. But the time saved did not translate into a lighter workload; it shifted into a denser mix of communication and coordination. Email time increased 104 percent, chat and messaging time rose 145 percent, and time in business management tools climbed 94 percent. Across the categories ActivTrak measured, increases ranged from 27 percent to 346 percent.

Meanwhile, the quality of attention declined. Daily focused time fell by 23 minutes. Focus efficiency dropped to 60 percent, a three-year low. Multitasking climbed 12 percent, collaboration time jumped 34 percent, and the share of employees at risk of disengagement rose 23 percent. ActivTrak captured the distinction in one line worth taping to a monitor: maximum output and sustainable performance are not the same thing.

For managers, the pattern is clear. AI makes individual tasks faster, but the additional drafts, messages, and analyses create more review and coordination work. Your people are producing more while concentrating less. Ask them to describe their day, and you will hear about fragments rather than finished thoughts.

The Tax Nobody Budgeted For

Now add a cost that never appears on the productivity dashboard. In September 2025, BetterUp Labs and the Stanford Social Media Lab surveyed 1,004 full-time U.S. desk workers about a habit they called workslop: AI-generated content that looks polished but moves the work nowhere. Think of a slide deck with confident headings and no real analysis, a summary that misses the one point that mattered, or a code-review comment that sounds smart but is wrong.

Forty percent of respondents had received workslop in the previous month. Workers estimated that 15.4 percent of the material reaching their inboxes fit the description. Each incident took an average of 1 hour and 51 minutes to fix or decode. BetterUp valued that burden at $186 per employee per month—more than $9 million annually for a company with 10,000 workers.

Here is the finding that should command managers’ attention. Fifty-four percent of managers reported receiving workslop, compared with 38.5 percent of individual contributors. You sit where everyone’s output arrives for review. When your team sends half-checked AI drafts upward, you become the quality filter—and no one has cleared space on your calendar for that work. More than 53 percent of respondents admitted forwarding AI-generated work they suspected was substandard. The problem, then, is not a handful of careless colleagues; it is the predictable result of rewarding speed without rewarding verification.

The damage extends beyond lost time. About half of those who received workslop viewed the sender as less creative, capable, and reliable. Forty-two percent considered the sender less trustworthy, and nearly a third said they were less likely to collaborate with that person again. When a team stops trusting its colleagues’ drafts, people begin redoing one another’s work—and the hours multiply.

Why the Corner Office and the Cubicle See Different Things

Korn Ferry’s Workforce 2026 report, based on responses from 16,000 employees in 11 countries, shows how the paradox varies across the organizational chart. Seventy-nine percent of CEOs said AI had improved efficiency at their companies. Only about half of individual contributors agreed. Fifty-two percent of all workers said AI had increased their workload, and 61 percent said they now carry the responsibilities of more than one role.

The U.S. figures are sharper. Fifty-three percent of American workers said AI tools increased the number of tasks expected of them. At the same time, 61 percent agreed that AI improved their personal efficiency, suggesting they are not anti-AI. They value the tools; they simply see efficiency being converted into more assignments rather than more breathing room. In the same survey, 43 percent of U.S. managers said they were too busy to deliver meaningful results, 44 percent said their company had laid off managers, and half felt exhausted by the pace of change.

Korn Ferry stated the central point plainly: efficiency gains and workload reductions are not the same thing. Executives measure the first; employees experience the second. When those two groups look at the same rollout and report different realities, the person in the middle must translate—and that person is you.

Upwork found the same fault line. Forty-seven percent of employees using AI said they did not know how to achieve the productivity gains their employers expected. Seventy-one percent of full-time employees reported burnout, and one in three said they were likely to quit within six months because of overwork. Kelly Monahan, managing director of the Upwork Research Institute, attributed the mismatch directly to organizations introducing new technology into old work models and failing to unlock the expected value.

The Squeeze on the Middle Layer

In June 2026, Julia Shin and Sandra Sucher published a Harvard Business Review article based on 18 interviews with partners, managers, and junior consultants at two major consulting firms. The small sample warrants caution, but the divide they observed mirrors the larger surveys. Partners used AI to broaden project scope and accelerate timelines with smaller teams. Junior consultants produced substantially more. Managers inherited new responsibilities: reviewing AI-generated content, catching errors, and teaching their teams to use the tools—all while meeting the same delivery targets with little guidance from above.

Place that finding beside the workslop data, and the pattern is consistent. Senior leaders capture the upside, junior employees gain faster tools, and the people in between absorb the reviewing, coaching, and cleanup. When an organization treats AI adoption as a software installation rather than a redesign of how work gets done, the bill lands in the middle.

Senior leaders should recognize their role in this dynamic. You set the expectations that the middle layer must meet. If targets rise as soon as the tools arrive, no one in your organization receives a productivity dividend. They receive larger targets and a new set of risks to manage.

What to Do on Monday Morning

Start by measuring the right things. Track hours worked, focus time, and rework alongside output. If your dashboards show only output, they will keep declaring the rollout a success while your best people update their résumés. ActivTrak’s 60 percent focus-efficiency figure offers a useful benchmark.

Next, decide what will happen to the time AI saves before it disappears. Left unmanaged, that time fills with more tasks. Tell your team—and your boss—that part of the gain will support higher-quality work, part will support learning, and part will remain unallocated for recovery. If you cannot defend that allocation to your leadership, start with that conversation.

Set a verification standard for AI-assisted work. Ranganathan and Ye recommend building an AI practice: shared norms for using the tools, including deliberate pauses, sensible task sequencing, and stronger human oversight. In plain terms, the person who sends a draft owns it. If you would not put your name on it after reading it carefully, do not forward it. That rule attacks workslop at the source by shifting quality control from the recipient back to the sender.

Then protect the middle. If managers must now review AI output, coach employees on new tools, and maintain their previous targets, something must come off their plates. Reassess reporting requirements, meeting loads, and spans of control. When every employee can produce substantially more, a manager’s review burden can grow even if the team’s headcount does not.

Finally, ask people what they are doing now that they were not doing a year ago. Ask in a one-on-one conversation, not through a survey, and listen for the tasks no one assigned. Those tasks are evidence of workload creep, and each is either worth keeping deliberately or worth eliminating. Employees rarely volunteer them because they assume the extra effort is what you wanted.

Conclusion

Leading through AI adoption requires a simple discipline: do not confuse a faster team with a lighter workload. Research from UC Berkeley, ActivTrak, BetterUp and Stanford, Korn Ferry, and Upwork points in the same direction. AI accelerates tasks, expectations expand to consume the savings, attention fragments, and the people in the middle hold the system together until they no longer can.

None of this is an argument against AI. Your people say it makes them more efficient, and they are probably right. The question is what you do with that efficiency. Decide deliberately, or rising workloads will decide for you—and burnout will eventually appear in your attrition numbers.

You can lead differently from companies that later have to explain why their best people left. Choose one team this week. Measure its actual hours, focus time, and rework. Ask how employees’ days have changed since the tools arrived. Then decide together what the saved time is for. It may be one of the most valuable conversations you have about your AI plan.