Workers say AI makes them faster — but where does that saved time actually go?

This explores whether asking workers how much AI helps them gives an accurate picture of what AI actually changes at work. The short answer from the corpus: surveys capture something real, but it's usually the *feeling* of speed. They miss where that time goes afterward. The most striking pattern is that the same surveys that report big gains also reveal the hidden costs when they ask a second question. Zapier found 92% of enterprise users reporting productivity boosts, yet the average worker spends 4.5 hours a week cleaning up AI output How much time do workers really spend fixing AI mistakes?. Workday found that nearly 40% of reported time savings disappear into rework and verification, and only 14% of employees consistently come out ahead Where does AI's time savings actually go in practice?.

The gap shows up at every level, not just among individual workers. A survey of 750 executives found that perceived gains run ahead of measured ones, partly because operational improvements take time to show up in revenue Do AI productivity gains feel larger than they actually measure?. When researchers skip surveys and look at behavioral trace data (logs of what people actually do on their computers), the picture can flip. ActivTrak found that as AI adoption rose, people logged more hours in work apps, worked more weekends, and had less uninterrupted focus time than at any point in three years Does AI adoption actually reduce the work that employees do?. One explanation is that AI doesn't remove time from a task so much as move it, from doing the work to writing prompts and checking outputs Does AI really save time, or just change how we spend it?. A worker can honestly feel faster while spending the same total time on a task.

There's a deeper reason to doubt self-reports. People turn out to be poor judges of their own AI skill: one pooled analysis found almost no correlation (.055) between how competent people said they were with AI and how they actually performed Can self-ratings replace objective performance scores for AI competence?. Social pressure also bends the numbers in both directions. In Anthropic's interview study, most workers said AI saved them time, yet about 70% hid or played down their use because of stigma and worries about their jobs Why do workers hide productivity gains from AI use?. So a survey may overstate how much AI helps while understating how much people use it.

An individual's survey also can't see the costs that land on other people. Workers estimate that 15.4% of the work they *receive* is low-quality AI output, and each instance takes nearly two hours to deal with How much work that employees receive is actually unhelpful AI content?. That's often longer than it would have taken the sender to do the work properly. The sender records a time saving and the recipient absorbs the cost, and no single self-report captures both. BCG found a similar hidden cost: self-reported productivity rose with up to three AI tools, then fell with four or more, as the work of overseeing the tools turned into burnout and higher intent to quit Does using more AI tools always boost worker productivity?.

The most useful idea here may be that 'productivity' is the wrong unit to measure. Gains show up when people use AI on skills they already have, and they disappear when people use it to learn something new When does AI actually boost worker productivity?. Narayanan and Kapoor argue that AI speeds up only the middle 'execute' step of knowledge work, while deciding what to do and delivering the result stay the same or grow Does AI really compress all layers of knowledge work equally?. Studies of where workers actually hand off tasks to AI find it concentrated in information-heavy jobs, following what the technology can do rather than how often people chat with it Where have workers actually delegated tasks to AI?. A survey asking 'how much faster are you?' measures the one step that got faster and misses the steps that grew around it.