Workers who use artificial-intelligence tools are logging longer days than colleagues who avoid them, according to new observational data from the workplace-productivity software vendor ActivTrak. The company analyzed digital work activity from 9,140 knowledge workers across 467 enterprises between April 1 and September 25, 2026, and found that AI users averaged between 8.42 and 9.08 hours of digital work per day, compared with 7.58 hours for non-users — a gap that reaches up to roughly 90 minutes.
The finding, reported by Yahoo Tech, cuts against one of the most common promises attached to workplace AI: that the technology would hand employees back their time.
A 9,140-Worker Dataset Covering Five Months
ActivTrak’s sample drew on digital activity rather than self-reported hours, covering 467 enterprises over a roughly five-month window this spring and summer. The headline comparison is straightforward: between 8.42 and 9.08 hours of tracked digital work per day among people using AI tools, against 7.58 hours among those who do not.
The company drew a further line inside the AI-using group. It defined heavy AI users as employees engaging with AI tools for more than 45 minutes across at least five sessions daily. That group, ActivTrak said, showed the most pronounced patterns in the data.
Heavy AI Users Switched Apps More and Focused Less
According to ActivTrak’s analysis, heavy AI users:
- logged 18 extra after-hours minutes;
- switched applications 15.2% more often than non-users;
- averaged only 37.8 minutes of uninterrupted focus, versus 48.2 minutes for light users;
- spent 3% of their workday in formal meetings, compared with 5% for non-users.
That last number matters for interpretation. Because heavy users were in fewer formal meetings, not more, ActivTrak said simple meeting-load growth is a less likely explanation for their longer hours. The company cautioned, however, that other factors such as job role, seniority and company culture could also account for the difference.
ActivTrak Says the Link Is an Association, Not a Cause
ActivTrak was explicit that its numbers show an association between AI use and longer workdays rather than a causal relationship. It laid out two competing explanations that the data cannot separate: employees who already work longer hours may simply be more likely to adopt AI tools, or AI may enable workers to absorb more tasks and responsibilities than they otherwise would.
The company also flagged the limits of its own dataset. The data are observational and drawn from ActivTrak’s customer base, so the results may not generalize to hourly workers, unionized environments, or organizations using different monitoring systems.
Eight Months Inside a 200-Person Technology Company
A separate line of research points in a similar direction, with a far more granular view. UC Berkeley Haas researchers Xingqi Maggie Ye and Aruna Ranganathan spent eight months embedded at a technology company of roughly 200 employees, completing more than 40 semi-structured interviews as part of an ethnographic study of AI adoption in practice.
Their findings indicate that generative AI did not free up employee time. Instead, workers moved faster, took on broader responsibilities, and stretched work into hours previously treated as downtime — often without being directed to do so.
The Berkeley team identified three specific intensification patterns at the company:
- workers expanded the scope of their roles;
- they used AI during former break periods;
- they maintained multiple simultaneous AI-assisted workstreams.
Ye and Ranganathan also raised a sustainability concern: short-term productivity gains may be difficult to sustain if cognitive demands and performance expectations keep rising in parallel.
Sanders Pushes a 32-Hour Workweek as the Policy Fight Shifts
The policy argument around these findings centers less on whether AI produces a benefit than on who collects it. Sen. Bernie Sanders has proposed a 32-hour workweek at no loss of pay or benefits, arguing that the financial benefits of AI and automation should reach working families rather than flow primarily to employers and corporate executives.
Read against that proposal, the ActivTrak and Berkeley results suggest that expanded output and longer workdays are, for now, the more visible outcomes than reduced hours. Taken together, the two studies point to a gap between task-level efficiency and worker-level workload: finishing individual tasks faster means little if the task list grows faster still.
That is the unresolved question in both datasets — whether AI-driven efficiency ends up translating into more output demanded of the same workforce, or into fewer hours required from it. Neither the ActivTrak activity logs nor the Berkeley interviews settle it, but both describe workplaces where the first outcome is currently the one showing up in the numbers.
A Workforce Picture Alongside AI’s Physical Costs
The findings arrive as scrutiny of the AI buildout widens beyond the office. Investors have been reassessing valuations tied to the technology, a shift tracked in reporting on how tech stocks face AI scrutiny after a reported $1 trillion drop, while the infrastructure supporting these tools has drawn its own examination, including analysis of how AI data centers could undo 15 years of US climate gains.
For the 9,140 knowledge workers in ActivTrak’s sample, the measurable effect was narrower and more immediate: AI users were still at their keyboards after their colleagues had logged off.
