It Got Easier. It Did Not Get Shorter.
Timed properly, assisted work took exactly as long as doing it yourself. It simply felt lighter. We have spent our whole lives using effort as a proxy for time, and the proxy has quietly come apart.
Time one ordinary task this week. Note the clock when you open the document, note it again when the document is finished, and do not consult your memory in between. Most people who try this are surprised, and the surprise runs in one direction. The work felt quick. The clock disagrees.
A paper released in May goes some way to explaining why. Sunny Yu, Myra Cheng and colleagues ran a preregistered experiment with 1,237 participants doing simple cognitive tasks, either alone or with an AI assistant. Before starting, people predicted how long each version would take. Afterwards they were timed, and asked how much mental effort it had cost them.
The predictions were confident and generous — roughly a minute of saving on a short task. The timings found no difference at all. What did move was effort: assisted work felt lighter, consistently and measurably.
Then the detail that makes the finding worth keeping. When participants imagined being helped by another person rather than a machine, the over-prediction vanished. The illusion is not about help. It is about this particular kind of help.
Effort was never the thing you were measuring. It was the gauge you read, because for the whole of human history it moved with the clock. It has come off the clock, and it is still the only gauge on the dashboard.
Why the proxy held for so long
Nobody carries a stopwatch. Duration gets estimated afterwards, from how much the thing took out of you, and until very recently that worked beautifully. Two hours of digging feels like two hours because it costs two hours' worth. Rewriting a paragraph six times feels long because it is genuinely tiring. The correlation was close enough to perfect that nobody had cause to notice it was a correlation at all.
The reason it survives with human helpers is probably that you retain a picture of what they are doing. You know roughly how long a colleague needs to draft a section, because you have drafted sections. You can imagine the labour and estimate from it. With a machine there is nothing to imagine — the process is invisible and appears instantaneous — so the only material available for the estimate is your own experience of the exchange, and your own experience is that it was easy.
Which is a fair description of how a great deal of work now feels from the inside. Not shorter. Smoother. The friction goes and the hours stay, and the day ends with a vague sense of having covered a lot of ground against a list that has not moved nearly as far as it should have.
The counterpoint deserves stating properly, because it is strong. Lower effort is not nothing. Fatigue is a real cost with real consequences, and a day that leaves something in reserve at six is a better day than one that does not, even at identical output. The tasks in the study were short and simple, too; on long, genuinely difficult work the time savings may well be real, and an experiment of this shape cannot say either way. All fair. But it is not the claim being made. The claim on every product page and in every internal memo is about hours, and hours are precisely what did not move.
Reading a different instrument
The correction is dull and it works. Use the clock. Note the time at the start of a block and at the end of it, on paper if that is easier, and keep it up for a fortnight. Not as a tracking regime — the last thing any of this needs is another dashboard — but as a recalibration. You are checking a broken instrument against a working one, which is what you do with any instrument you have stopped trusting.
Then keep something unassisted, and keep it deliberately. Not everything, and not out of virtue. One real piece of work a week that you begin and finish yourself, whole, because it is now the only honest source of information about what things cost. Lose that and you lose your sense of the price of anything, and someone who does not know the price of things makes steadily worse decisions about what to buy.
What remains is a question of what you are prepared to count as a good day. If the measure is how light it felt, the measure will keep improving and nothing else will. If the measure is what was actually finished by six, at least it can be checked.