Faster work, same deadline

The premise

If the day a result comes out is set by deadlines, sign-offs and whoever is next in line, then making one person’s work faster won’t bring that day forward.

That’s a condition, not a conclusion. This piece doesn’t say AI is useless, or that productivity isn’t rising.

Side by side at the next light

I was on a road with a 70 limit, stuck behind a car doing 40. I overtook it, annoyed.

At the next traffic light, the car I had just overtaken pulled up right beside me.

Work felt the same. I worked hard and tried to make things easier and quicker. Since AI arrived, my work really has become easier and faster. Yet the monthly report still came out on the date it was due. If anything it slipped later. It didn’t come out earlier.

The car stops at the next light. The report comes out on the set date.

So why doesn’t one person’s faster work bring the result forward?

When motors turned the old shafts

A century ago, factories ran into a similar puzzle.

In a 1990 paper, the economic historian Paul David traced how American factories took up electricity. In 1899, electric motors provided less than 5% of the mechanical power driving factory machinery. It took roughly another 20 years for that share to reach half. Electricity only began to show up in manufacturing productivity in the early 1920s, four decades after central power stations first opened for business.

Why did it take so long? Part of the answer lies in what the new motors did at first.

Factories back then had long drive shafts running along the ceiling, with belts dropping down to turn each machine. The shafts were turned by a steam engine or a similar engine. At first, electric motors simply took over turning those shafts. In David’s account, the old belts, shafts and even the engines that had powered them stayed in place. There were gains: less power was needed, and machine speeds were easier to control. But the layout of the factory stayed the same.

The change came when each machine got its own motor. With the overhead shafts gone, buildings could be lighter. Instead of tall multi-storey mills, built that way to keep long shafts short, firms put up single-storey plants with machines laid out in the order materials moved through them. Machines became easier to move, and production lines easier to rearrange. David finds that about half of the jump in US manufacturing productivity growth in the 1920s, compared with the decade before, can be accounted for statistically by the spread of electric motors.

On David’s account, electric motors had been in factories since the 1890s. Their effect didn’t appear in the productivity figures until factories were rebuilt around them.

This account has its critics. A study of factories near early hydroelectric plants found that where electricity did arrive, productivity gains were rapid, which the authors say runs against the established story. But the same study notes that those gains came with new investment in equipment and changes in how firms were organised. Why electrification took so long to pay off is still debated, and this piece doesn’t try to settle it. It keeps only the ground both sides share: a faster motor on its own wasn’t enough; what surrounded it had to change too.

Where did my 30 hours go?

Back to today. Here’s a rough sum I did.

Say I do more than ten tasks a day and save more than ten minutes on each. That’s two hours a day, or about 30 hours a month. On paper, I’ve saved 30 hours. So where did they go?

Ten minutes on its own is a scrap: you can’t do much with it. To use that time for something else, it has to be turned into a different kind of time.

What does the research say?

In Denmark, two economists surveyed workers in 11 occupations, about 25,000 each time over two rounds, and matched their answers with official records. Workers who used AI chatbots said the tools saved them, on average, 2.8% of their working hours. That’s far less than the gains of more than 15% often found in experiments with the same occupations. And the official records showed no change in earnings or hours worked: effects larger than 1% were statistically ruled out.

Asked what they would do with the time they saved, most said they would spend it on other tasks, including new work that AI itself had created.

The paper has since been revised. The March 2026 version comes under a new title, Still Waters, Rapid Currents, and rules out effects larger than 2% rather than 1%. The figures above are from the July 2025 version.

A second study points the same way. In an experiment across 66 firms and 7,137 office workers, some staff were picked at random and given an AI tool. Those who used it spent two fewer hours a week on email, and did less work outside normal hours. But beyond those individual time savings, the researchers found no sign that giving the tool to individuals changed how much work they did, or what kind. Three of the study’s four authors work at Microsoft Research, part of the company that makes the tool.

Both studies end up in the same place. Time was saved, and it didn’t disappear; it went into other work. But the amount and kind of work, earnings and hours all stayed the same.

My two hours a day is about a quarter of a working day, far more than the 2.8% in the survey. The survey average includes days and tasks where people don’t use AI at all, so the two numbers don’t measure the same thing. I simply set them side by side. Still, within what both studies looked at, the time saved didn’t show up as more work done or higher pay.

The traffic light isn’t about your speed

Overtaking the car ahead was about my speed. Stopping at the next light was about the light.

Reports work the same way. My part of the job can get faster, but the day the report comes out is set by the reporting cycle, the order of sign-offs and the next person’s schedule. The ten minutes I save wait in front of all that.

It was the same with factory motors. While they turned the old shafts, the flow through the factory stayed the same. It changed only once the layout, the buildings and the order of work were redesigned around them.

In Belgium, 34.5% of firms with ten or more employees used AI in 2025, against 20.0% across the EU. The tools have already arrived. But a tool arriving is not the same as changing the path the work takes.

What this piece does not claim

• That AI fails to raise productivity
• That companies or managers deliberately slow work down
• That reorganising will quickly produce results
• A verdict on why electrification took so long to pay off

What we checked, and what we couldn’t

Checked

• In 1899, electric motors provided less than 5% of US factory mechanical drive; it took about 20 more years to reach half; the productivity effect appeared in the early 1920s, four decades after central power stations first opened — David (1990)
• Early electric motors turned the existing drive shafts, and the belts, shafts and engines stayed. Once each machine had its own motor, lighter buildings, single-storey plants and rearranged layouts became possible — David (1990)
• About half of the acceleration in manufacturing productivity growth in 1919–29 is accounted for statistically by the growth of electric motor capacity — David (1990)
• Near early hydroelectric plants, productivity gains from electrification were rapid, and came with investment in equipment and organisational change — Fiszbein et al. (NBER Working Paper 28076, not peer-reviewed)
• Denmark: users reported average time savings of 2.8% of work hours; no effect on earnings or hours (effects above 1% ruled out); most said they would use saved time for other tasks — Humlum and Vestergaard (2025)
• Experiment with 7,137 workers at 66 firms: two fewer hours a week on email; giving the tool to individuals didn’t change the quantity or mix of tasks — Dillon et al. (2025)
• Firms with ten or more employees using AI in 2025: Belgium 34.5%, EU 20.0% — Eurostat

Couldn’t check

• My 30 hours is a rough personal estimate
• What workers actually did with saved time (the Danish question asked what they expected to do)
• Why electrification took so long to pay off
• Any study measuring whether small scraps of time add up
• Time saved by Belgian workers
• Whether reporting and sign-off structures are the same in every workplace