*The promise of workplace AI was always time: hours handed back to employees buried in routine work. A new study suggests the hours are real and that most companies are quietly losing them again. *
According to research from Workday, surveying 3,200 business leaders, 85% of employees now save between one and seven hours a week using AI, and nearly 40% of that saved time is immediately lost to rework.
Rework is the quiet villain of the AI-productivity story. The time an employee saves by having a model draft a document or summarise a dataset is real, but if the output then has to be checked, corrected and partly redone, much of the saving evaporates. Workday’s finding puts a number on an experience many knowledge workers will recognise: the AI is fast, and then you spend the time you saved making sure it was right.
The pattern shows up across the wider 2026 research, not just Workday’s. Stanford and BetterUp researchers have described “workslop,” AI-generated content that looks polished but lacks substance, estimating that 40% of US workers received some in a recent month and putting the cost at millions a year in lost productivity for a large organisation. The throughput goes up; the quality does not always follow; and somebody downstream pays for the gap.
The more structural finding is about who captures whatever value survives. PwC’s 2026 AI study concluded that nearly three-quarters of AI’s economic gains are accruing to about a fifth of companies, the ones treating AI as a lever for growth rather than a way to shave costs. That concentration matters.
It suggests the technology is not a rising tide lifting all firms equally, but an advantage compounding for organisations that have done the harder work of redesigning how they operate around it.
What separates the two groups is less the tools than the discipline. Buying AI licences and letting employees save scattered minutes produces exactly the dispersed, reworked, ultimately unbanked gains the Workday data describes. Converting those minutes into measurable value requires deciding what the freed-up time is for, and most companies, the study implies, have not made that decision.
The default, several of the 2026 studies note, is simply to demand more output from the same people, which redirects the saving into more work rather than better work, and in some findings into longer hours and higher burnout.
None of this says AI fails to save time. The consistent finding across the research is that it does, often substantially. The problem sits one layer up, in the organisational choices about what happens to the time once it is saved.
A tool that hands an employee an hour is only as useful as the company’s ability to do something deliberate with that hour, and the evidence so far is that most are letting it leak away through rework, quality problems and the reflex to simply pile on more.
For all the spending on enterprise AI, the studies point at an unglamorous conclusion. The bottleneck is no longer the technology’s capability but the management around it. The companies banking real gains are not necessarily the ones with the best models; they are the ones that decided, in advance, what the saved time was for. The rest are buying the hours and dropping them on the floor.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.