Who gets the time saved through AI?

The most solid answer comes from Denmark, where 2 economists linked a survey of 25,000 employees in 11 occupations exposed to AI to payroll and hours records. According to their answers, users there gain on average 3% of their working time. They also say what they do with it.

The time saved therefore stays within work. As our analysis on productivity points out, the authors measure no effect of AI on wages or on recorded hours, and rule out any average effect greater than 2%. They put forward, as a hypothesis, that employees and companies absorb the change by redistributing tasks, without pay or hours bearing any trace of it.

AI also creates work. In the same survey, 8% of users whose employer does nothing to encourage AI say they have taken on entirely new tasks because of AI, and 17% where the employer actively pushes its use. 42% of these new tasks involve content production, and 35% the review of what AI produces, ethics and compliance. The authors note that those who report saving the most time are also those who take on the most new tasks.

A field study goes further. For 8 months, 2 researchers from the University of California, Berkeley observed an American technology company of about 200 employees. They describe tasks that expand, a faster pace and lengthening working days. It is a single case, published in the Harvard Business Review and not in a scientific journal, and it shows at close range how the freed-up time fills up.

The trial conducted in 66 companies, where an assistant was given by random draw to some of 7,137 employees, makes it possible to track time in applications. Employees with the tool spent 1.4 hour less per week on email on average, and 2 hours for those who actually used the tool. They worked a quarter of an hour less per week outside their usual hours, 9% less. As for the rest, the authors write that a large part of the time saved seems to have gone to activities that they cannot measure.

The quarter of an hour less outside hours, out of 1.4 hour saved on email, is the only part of the gain that the trial shows going back to employees. The authors do not know what the rest was used for: they see neither more emails handled, nor more meetings, nor more documents.

One could answer us that this time already has a recipient. An employee is paid for their hours, and an hour saved and spent on other tasks benefits the organisation and its clients: files handled sooner, better-reviewed work. Wages would follow later, as they followed earlier technologies, with the delay described in our analysis on productivity. The trial in 66 companies points in this direction: employees with the tool did not answer more emails, attend more meetings or write more documents, and the authors do not see any new work.

We accept part of the objection. Reallocating time to other tasks can be a real gain, and nothing in these studies says that it is lost. We do not accept it as a rule, for 2 reasons. This gain is not yet visible either in what companies produce or in pay, so nobody can say today who benefits from it. And letting the answer emerge on its own amounts to choosing, without saying so, to produce more, with a use of AI that nobody counts.