Does AI really save you time at work? How to measure it

Most people who use AI feel faster. Feeling faster and being faster are different things, and the only way to know which one you have is to time it.

Why the feeling misleads you

In 2025 the research group METR ran a randomised trial with experienced open-source developers. Before the tasks, the developers expected AI tools to make them about 24% faster. Afterwards they believed they had been about 20% faster. Measured, they were about 19% slower.

That result is about one group doing one kind of work, so it does not say AI is slow. It says that how fast something feels is a poor measure of how fast it is. Waiting, prompting, reading and fixing all feel like progress while the clock keeps running.

The test, in five steps

Pick one task you do again and again, such as a weekly report, a client email or a summary. Then:

What to look at

Four columns are enough: time, quality, effort and rework. A saving of ten minutes that comes with an hour of cleaning up later is not a saving. A task that gets faster but worse is a different decision from one that gets faster and stays good.

If the number does not move, that is useful too. It tells you this task is not a good fit, and you can try another one instead of using AI on everything by default.

Make the next run faster

Most lost time comes from vague instructions. A clear brief with a goal, context, material, boundaries, the output you want and a way to check it usually removes a round or two of back and forth. The free Unpuzzled Brief worksheet lays those six fields out on one page.