Three in 10 workers now say artificial intelligence has added to their workload rather than reduced it, even as employers ratchet up productivity targets. At the same time, a large randomized experiment found that AI tools can cut email reading time by roughly half an hour per week for active users. The gap between those two findings captures a growing tension. AI may save minutes on specific tasks while expanding the total volume of work employees are expected to handle.
The Efficiency Promise vs. the Workload Reality
The pitch for workplace AI has always centered on doing more with less. Tools like Microsoft 365 Copilot promise to summarize threads, draft replies, and triage inboxes faster than any human can. A randomized experiment involving over 6,000 workers at 56 firms, conducted by Microsoft Research authors and published on arXiv, found that Copilot users reduced their weekly time reading email by about half an hour under treatment-on-the-treated estimates, with smaller intent-to-treat reductions across the broader sample. A companion field study, also published on arXiv, reported sizeable reductions in email-related time among generative AI adopters, measured in hours per week and percentage terms.
Those findings sound like clear wins. But the same companion study flagged something less encouraging: changes beyond core hours, including work that spilled outside regular schedules. Saving 30 minutes on email matters far less if that freed-up time simply gets absorbed by new assignments or extended evening sessions. And that is exactly what broader workforce surveys are starting to show.
Employers Raise the Bar After AI Adoption
A ResumeTemplates.com survey found that 3 in 10 respondents say AI has increased their workload as employers raise productivity expectations. The mechanism is straightforward: once managers see that a task can be completed faster with AI assistance, they expect employees to take on additional tasks in the time that was freed up. The efficiency gain does not translate into shorter days or lighter loads. It translates into higher output targets.
This dynamic explains why individual time savings and aggregate workload growth can coexist without contradiction. A worker who spends 30 fewer minutes reading email may now be expected to produce an extra report, respond to more clients, or attend additional meetings. The tool did exactly what it promised at the task level. The organization, however, treated the savings as surplus capacity rather than as relief.
Public-relations channels have amplified this storyline. Corporate communications distributed through platforms like PR Newswire emphasize efficiency and innovation, while workers describe a more mixed reality in internal surveys and exit interviews. Media resources such as PRN Media then help translate those upbeat corporate messages into news coverage, even as independent reporting surfaces concerns about burnout and intensifying workloads.
ActivTrak Data Points to Intensified Work
Workforce analytics firm ActivTrak examined AI users’ digital activity 180 days before and after they began using such tools on the job, according to reporting in the Wall Street Journal. The analysis found that AI intensified work patterns rather than easing them, with users packing more tasks and application switches into each day. That 180-day window is significant because it captures not just the honeymoon phase of a new tool but the period when organizations begin adjusting expectations around it.
The ActivTrak findings sit in direct tension with the Microsoft-backed arXiv studies. One set of evidence shows email time dropping for Copilot users; the other shows overall work intensity climbing for AI adopters more broadly. Both can be true simultaneously. Cutting time on one task does not prevent the total workload from expanding when the organizational response is to pile on new demands. The conflict between these datasets is not a contradiction so much as a story told at two different scales: the task level and the job level.
The Productivity Paradox Returns
Erik Brynjolfsson, a Stanford economist affiliated with MIT Sloan, has framed this disconnect in blunt terms. “This is one of the great puzzles of our era: amazing technologies, but so far, slow productivity growth,” Brynjolfsson has said, describing what he sometimes calls a paradox. The pattern echoes earlier waves of technology adoption, from personal computers in the 1980s to smartphones in the 2010s, where measurable productivity gains lagged years behind deployment.
But the current AI cycle has a feature those earlier waves lacked, speed of organizational response. Companies are not waiting years to adjust expectations. The ResumeTemplates.com survey data shows employers already raising the bar while tools are still being rolled out. That compressed timeline means workers feel the squeeze almost immediately. They get the tool and the higher quota in the same quarter.
Economists often argue that it takes time for organizations to redesign workflows, retrain staff, and rethink business models around new technologies. In the interim, employees can end up in a worst-of-both-worlds scenario. They must master unfamiliar tools while still hitting legacy targets, only to see those targets ratcheted up once they start to realize any efficiency gains.
Why Email Time Tells Only Part of the Story
The headline claim that AI is transforming email deserves careful unpacking. The primary research from the arXiv experiments documents reductions in email reading time for Copilot users, not increases. What appears to be growing is the total volume of communication and task output that organizations expect, which can make the overall time spent on email-adjacent work feel larger even as the per-message processing time shrinks.
Consider a practical example. A marketing manager who once spent two hours reading and responding to 60 emails per day might now process each message in half the time with AI help. But if the organization responds by routing 120 messages to that manager, the net time spent on email has not changed or may have grown, even though the tool performed exactly as advertised. The efficiency gain was real. The workload reduction was not.
This distinction matters for how companies measure return on investment for AI tools. Vendor dashboards that highlight minutes saved on email or documents can be misleading if they are not paired with metrics on total hours worked, after-hours activity, or burnout indicators. A narrow focus on task-level efficiency risks obscuring a broader shift toward more intense, more fragmented days.
Designing AI Adoption Around People, Not Just Output
The emerging evidence suggests that AI’s impact on work is not technologically predetermined; it is shaped by managerial choices. Organizations that treat time savings as an opportunity to reduce stress, allow deeper focus, or shorten the workday will see a different outcome than those that immediately convert every minute saved into additional output.
Several practical steps follow from that insight. First, companies can explicitly define how reclaimed time should be used—whether for strategic planning, learning, or genuine rest—rather than leaving it as an unclaimed pool that managers can quietly fill. Second, leaders can track not only productivity metrics but also indicators of overload, such as after-hours logins or rapid increases in task volume following AI deployment. Finally, workers themselves can be involved in designing AI-enabled workflows, ensuring that tools support sustainable pace rather than relentless acceleration.
The research from Microsoft, ActivTrak, and workforce surveys all point in the same direction: AI is powerful at shaving minutes off individual tasks, but without intentional guardrails, those minutes are quickly reinvested into more work. The real test for this generation of AI will not be how quickly it can summarize an inbox, but whether organizations can resist the reflex to turn every efficiency gain into a new baseline for how much employees are expected to do.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
