Enterprises spent two years selling employees on artificial intelligence as a way to do less low-value work. New research on what workers actually experience day to day says the opposite is happening in a large share of jobs: AI is adding a second job on top of the first, just one nobody budgeted for.
The shift: AI is generating its own workload
A survey of 1,000 U.S. full-time professionals in finance, healthcare and the public sector, released this month by enterprise browser company HERE, found that 48% say AI has created roughly as much extra work as it has saved. Nearly a third of respondents spend at least half their workday re-entering information between AI tools and the systems of record they still have to maintain by hand. Sixty-four percent say AI has made it harder, not easier, to access the applications and content they need to do their jobs, and half report a rise in alerts and notifications since AI tools arrived.
“Employees are under intense pressure to be more productive, so it’s not surprising many are turning to consumer AI tools outside their approved environments,” said Mazy Dar, CEO of HERE. That pressure is showing up as a shadow-IT problem: 72% of respondents admit to bypassing their organization’s AI restrictions, 36% use personal devices to keep AI activity off company systems entirely, and 34% run personal AI accounts on company hardware. Only 27% say they anonymize sensitive data before feeding it into an AI tool, even when policy requires it.
Why this is not last year’s “AI adoption” story
The first wave of workplace AI coverage was about whether employees would use the tools. This wave is about what using them actually costs in attention and coordination. Forty percent of respondents say they now handle a higher volume of tasks specifically because they are overseeing or correcting AI output, and 45% say managing AI takes up more of their day than their core responsibilities do. That is a different failure mode than low adoption: it is high adoption colliding with tooling that was never redesigned around the extra verification, re-entry and context-switching AI use actually demands. HERE’s researchers call the resulting overhead a “toggle tax,” the accumulated cost of constantly switching between the AI layer and the legacy systems it was supposed to replace, not integrate with.
The evidence is not isolated to one survey
This lines up with a broader pattern the HR-tech press has been documenting all year rather than a one-off finding. Federal Reserve research published earlier this month found the productivity payoff companies promise on earnings calls is still mostly a promise, not something showing up cleanly in output data. And a separate survey found workers trust AI output less than they actually rely on it day to day, which is its own kind of hidden tax: verifying work you do not fully trust takes time the tool was supposed to save. Put together, the picture is less “AI doesn’t work” than “AI works differently than the productivity case assumed,” and the gap between those two claims is exactly where most enterprise rollouts are currently getting judged.
Who feels it worst, and who does not
The toggle tax is not evenly distributed. Sixty-three percent of executives in the HERE survey identify as AI “super users” against 14% of individual contributors, and 79% of executives say they feel very confident using AI for complex tasks versus 39% of individual contributors. That confidence gap matters because it is executives, not the individual contributors absorbing the re-entry and verification load, who are approving the next AI budget line. A rollout that looks like a clean win from the C-suite dashboard can still be adding hours to a frontline employee’s week, and 69% of workers in customer-facing roles say AI has already changed how those interactions play out, not always for the better.
The generational split tells a related story. Forty-one percent of Gen Z respondents identify as AI super users versus 6% of Baby Boomers, and 42% of Gen Z report a significant jump in productivity expectations tied to AI, against 6% of Baby Boomers. That gap is often read as a training issue: get older workers comfortable with the tools and the toggle tax narrows on its own. HERE’s data suggests it is at least as much a systems issue, since even confident, high-usage employees still report re-entry and verification overhead. Comfort with a tool does not remove the extra step of reconciling what it produced with the system the company actually runs on.
What it means for the HR leader
This is a workforce-planning problem before it is a technology problem. If 30% of a workforce is spending half its day on manual re-entry between AI and legacy systems, that is lost capacity that will not show up in an engagement survey until burnout does, and HERE’s data ties negative AI experiences directly to higher reported mental fatigue. HR teams that own AI rollout metrics should be tracking toggle time and workaround rates the same way they track adoption rates, because adoption without integration is not the win it looks like on a dashboard. The 72% workaround rate is also a governance signal: employees are not ignoring AI policy out of carelessness, they are routing around tools that were never rebuilt to fit how the work actually flows.
What to do about it
Three moves separate rollouts that reduce work from rollouts that relocate it. First, measure toggle time and re-entry volume before declaring a rollout successful, not just seat counts or query volume. Second, treat a high workaround rate as a design failure in the approved tools, not a discipline failure in employees, and fix the gap the workaround is filling. Third, weight AI budget decisions toward integration work, connecting the AI layer to the systems people already use, over adding another standalone tool that creates one more thing to check. The organizations narrowing the toggle tax are the ones treating it as a workflow-design problem, not a training problem.
Source: HERE via GlobeNewswire
