AI saved you an hour. Work took it back.

Employment Hero surveyed 3,290 business leaders and 5,454 employees across Australia, Canada, New Zealand and the UK this spring. Nearly three in four employees said AI had made them more productive, although only one in four called the improvement significant.

The revealing number is what happened next. Among employees who saved time, 27% moved it into higher-level strategic or creative work. Another 26% simply did more tasks. Fourteen percent spent it managing the AI. Only 21% said AI let them work fewer hours. For 12%, it saved no time at all.

There is review work hiding in the result too. Sixty-three percent of employees said AI created more work checking outputs. Employers saw a brighter picture: 65% said AI was accelerating the business. Both can be true. The company gets more output while the worker gets a new pile of drafts to inspect.

This is a survey, not a stopwatch study, and Employment Hero has an interest in a positive employment story. Its report is unusually candid about one limit: it cannot tell whether AI caused stronger companies to grow or whether stronger companies were simply more likely to adopt AI. Treat the figures as a useful view of worker experience, not a universal law.

Statistics Canada found that workplace use of generative AI rose from 17% of workers in September 2024 to 30% by July 2025. The agency also found a steep education gap: 37% of workers with a bachelor's degree or higher had used generative AI at work, compared with 7% of workers with a high school diploma or less.

The International Labour Organization sees the same unevenness from another angle. Its July brief estimates that nearly 80 million workers in ASEAN hold jobs with more than minimal exposure to generative AI, but says widespread job disruption is not yet visible. The ILO warns that the effect may show up first in task design, work intensity, monitoring, pay and access to entry-level experience—not only in headline layoffs.

That matters because 'learn AI' is starting to function like free-floating career advice. The Employment Hero survey says half of workers believe their employer is doing little or nothing to help them build AI skills, while 41% think the responsibility is theirs. The bargain gets strange fast: learn the tool on your own time, use it to increase output, then spend part of the gain checking its work.

A useful AI tool can absolutely remove drudgery. The mistake is assuming that a faster task automatically produces a better workday. Most organizations know how to claim freed capacity. Far fewer know how to leave it free.

Imagine a weekly client report that used to take three hours and now takes one. There are at least four possible outcomes. The employee leaves two hours earlier. The employee does deeper work with the same deadline. The business adds two more reports. Or the employee spends an hour correcting the draft and another hour proving the numbers are safe to send. All four may appear in a dashboard as AI adoption.

The sharp question is not 'How many people used the tool?' It is 'What stopped happening?' Which recurring task disappeared? Which meeting got shorter? Which evening check was removed? If the answer is nothing, the tool may still be useful, but it did not return time. It raised the production ceiling.

Priya Rao would put the missing labor back into the measurement. Start with one repeated task and count the old minutes, AI run time, checking, corrections, reruns and any cleanup pushed to a teammate. Then track where the net time went. 'More output' and 'fewer working hours' are different outcomes; a company should stop presenting one as the other.

Mina Torres is less interested in the spreadsheet than in the end of the day. Her test is whether someone can close the laptop without carrying a second shift in their head. If the assistant drafts five things but leaves five uncertain things to remember over dinner, the speed is real and the relief is not.

These views pull in different directions on purpose. Measurement can expose the missing minutes. It cannot decide what the organization is allowed to do with them. That part is policy, management and—where workers have leverage—negotiation.

Pick one boring task. Do not start with a department-wide transformation or a company AI mandate. Choose the report, inbox pass, schedule cleanup, quote draft or research brief that reliably consumes part of the week.

For five runs, record the full before and after: active minutes, waiting time, review, corrections, repeat explanations, help from other people and whether the result was actually used. Also record the consequence. Did another task replace it? Did a deadline move? Did the person stop earlier? Did someone else inherit the checking?

At the end of the week, make one deletion decision. If the AI-assisted version works, remove an old step, check-in, duplicate document or meeting. If nothing can be removed, say that plainly. You bought more capacity. You did not buy time back.

There is nothing wrong with a business choosing more output. There is something wrong with selling that choice to workers as a shorter day when the day never got shorter.