New ActivTrak data on 9,140 knowledge workers finds that where AI sits matters as much as whether people use it. Employees who reach for AI in a browser tab bounce between applications far more than anyone else, while those using AI built into their tools switch least of all.
Key Takeaways
Most AI rollouts are judged on adoption: how many licenses are active, how many prompts are sent, how many employees say they use the tools each week. New behavioral data suggests that is the wrong scoreboard. The same AI capability can make an employee’s day calmer or more fragmented depending on one design decision most companies never make on purpose: whether the AI lives inside the application where the work happens or in a separate browser tab.
The ActivTrak Productivity Lab released its analysis on October 29. It draws on digital activity from 9,140 knowledge workers across 467 enterprises between April 1 and September 25, 2026. Rather than asking people how AI changed their work, it measured what they actually did on their computers.
The sharpest finding concerns application switching. Employees whose AI is embedded in the tools they already use switch applications 0.6 times an hour. Employees using browser-based AI tools switch 1.7 to 4.2 times an hour. Non-AI users sit in between at 0.95. At the top of the range, browser-based AI users move between applications more than four times as often as colleagues who use no AI at all. ActivTrak calls the extra friction a “swivel-chair tax”: copy the problem out, paste it into the assistant, read the answer, carry it back, and repeat. Embedded AI users, by contrast, switch less than people with no AI at all.
Across all heavy AI users, defined as people who spend more than 45 minutes a day in AI tools across at least five sessions, switching runs at 1.1 times an hour, 15.2% more often than non-users. Their uninterrupted focus sessions average 37.8 minutes, against 48.2 minutes for light users, a 22% gap. Light users also spend more of their active time in the core applications of their jobs.
The switching sits alongside a second pattern. AI users’ workdays span an average of 8.4 to 9.1 hours, compared with 7.6 hours for non-users, a gap of up to 90 minutes. Heavy users add 18 minutes of after-hours work a day. They spend less time in formal meetings, 3% of their working time against 5% for non-users, so the extra hours are not coming from the calendar. “AI may help employees complete tasks faster, but our research shows it contributes to longer workdays,” said Gabriela Mauch, chief customer officer and head of the ActivTrak Productivity Lab.
The data shows an association, not proof that AI causes the longer days, and people who already work long hours may simply be the ones who adopt AI most. But it complicates the easy story. As CIO reported on November 3, the new findings sit uneasily beside ActivTrak’s earlier three-year study, which found productive time rising with AI while focus efficiency fell to a three-year low of 60% and task switching climbed 12%. Sanchit Vir Gogia, chief analyst at Greyhound Research, told the publication that “an improvement per paid hour can conceal effort left off the timesheet.”
Mauch offered a reason the gains may not be turning into shorter days. “We’re in an extremely productivity-driven economy, where if we get some, we want more,” she said, in comments reported by Gizmodo. When tasks finish faster, the time saved tends to be filled with more tasks rather than handed back.
ActivTrak broke the data down by role, and operations managers stand out. They spend 75.5% of their active time in core applications and switch 1.1 times an hour on average. During heavy coordination periods, switching rises to 4.1 times an hour, the same level as the most fragmented browser-based AI users. Software engineers, by comparison, spend 78.5% of their active time in core coding tools.
That matters because coordination is the job. An operations manager chasing a delayed shipment, reconciling a schedule, or preparing a status update for leadership is already moving between a ticketing system, a spreadsheet, chat, email, and a project tracker. Adding a standalone AI assistant to that loop adds one more window to cycle through. AI built into the systems of record removes a step instead of adding one.
The finding also reframes a familiar procurement debate. Companies have spent two years buying general-purpose AI licenses because they are fast to deploy and easy to measure. ActivTrak’s numbers suggest the cheaper question to ask before the next purchase is where people will use the tool, not just whether they will. “For leaders, the goal can’t simply be AI usage,” Mauch said.
The lesson from ActivTrak’s data is not that AI makes work worse. Embedded AI users switched applications less than anyone else in the study. The lesson is that AI placed beside the work, rather than inside it, turns every answer into a round trip. For operations leaders, the next gain may come less from adding AI than from moving it to where the work already is.
Guide
The swivel-chair tax is sharpest at the contact center desk, where every extra window adds seconds to a live call. This guide shows how cutting the applications agents toggle between improves handle time and first-contact resolution.
Download
Case Study
Operations managers’ switching jumps to 4.1 an hour during heavy coordination. This case study shows how one production operation cut turnaround in half by pulling workflows, asset tracking, and cross-team coordination into a single workspace.
Download
Whitepaper
Drafting a status update in a chatbot and then rebuilding it slide by slide is the round trip ActivTrak measured. This research quantifies how much time teams lose building and reformatting presentations, a drain AI inside the deck tool is designed to remove.
Download
Capacity finds 44% of consumers say AI has contradicted a company representative or its policies, and they are about as likely to believe the AI. One in five now escalate faster when AI is in the loop.
Cognizant finds 80% of employees want more AI than their jobs permit and up to 82% say their tools fall short. Two-thirds of office professionals already use AI their employer does not allow.
Gartner expects AI spending to jump 49.5% this year, largely through software companies already own. Workday’s research shows the cost when those tools stay disconnected: a workday a week spent gluing systems together.
