Count one thing before you change anything, change it, then count the same thing again a month later. That is the whole test. How a new tool feels is a poor guide to whether it is working, and there is good evidence that even experts misjudge it.
How far off can a feeling be?
Further than you would think. In a 2025 trial by METR, a research nonprofit, experienced open-source developers worked on real tasks in projects they knew well, some with AI tools allowed and some without. They expected the tools to speed them up by 24%. The tasks done with the tools took 19% longer. Afterwards, the developers still believed they had been about 20% faster (METR, July 2025).
Read it with its limits. It is one setting: senior people, code they knew intimately, early-2025 tools. It does not show that software slows everyone down. What it does show is the gap between feeling and fact, in people who measure things for a living.
Why does slower feel faster?
Because the effort changes shape. Starting from a draft feels lighter than starting from a blank page, even when checking and fixing the draft takes longer than writing it would have. Reviewing feels easy. It still costs minutes.
The same thing happens far from software. A reply drafted in seconds, then rewritten twice. Ten versions of a post, then half an hour choosing one. Motion is easy to mistake for progress.
Which number do you count?
One number the business already cares about, measured the same way before and after:
- replies sent within the hour
- quotes followed up to a yes or a no
- hours a week spent typing the same details twice
- errors that reached a customer
Count the whole job, including the checking and the fixing, and write the before number down before anything changes. Without it, nobody can prove the change worked, and nobody can prove it did not.
What goes wrong?
In the choice of number. Owners pick something easy to move (messages sent, tasks created) instead of something that matters (jobs booked, hours back). And they measure the new way for a week while it still has everyone's attention, and the old way from memory.
One step for this week
Pick your most repetitive weekly task and time it by hand, once, before anything about it changes. Write the number where you will find it in a month. If the task sits across several people, it shows up faster on the one-week list of work no customer sees.
We put a number on the work before you commit to anything. See how we work.


