If an employee completes in three hours a task that used to take eight, can their employer claim the five hours saved? This question, raised by the rise of AI in organizations, opens a conflict over the value of progress. The hourly framework is insufficient to answer it.
The case is presented as schematic. An employee discovers they can prepare a file in three hours that used to take them a full day. After testing tools, correcting their errors, and building a reliable method, they present it to their manager, who congratulates them before assigning them two additional files. The company achieves a productivity gain, but for the person who originated it, the innovation translates into an immediate increase in their workload.
This scenario raises an often-avoided question: who should benefit from an accelerated task? The employer, who pays the salary and bears the costs of the activity? The employee, who sometimes invested heavily to find a better method? Or both?
Shared progress, but not always recognized
In France, actual working time depends in particular on the employee’s availability to their employer. However, this framework answers a contractual question, not that of the distribution of the value created. Therefore, assigning more tasks is not necessarily a good decision.
The employer often finances licenses, data, training, and infrastructure, and assumes the errors. Recovering time can improve service or preserve a fragile activity. But some employees design, on their own, processes that can be passed on to their colleagues, and the company then benefits from an innovation it had not imagined. The employee does not become the owner of five hours of their day, but they have produced a lasting improvement in the organization. A strictly hourly reasoning makes this contribution invisible: the person who invented a better method can be evaluated the same as the one who merely applies it.
The text takes the example of two employees who automated the same task. The first shares their process and sees their objectives increase, while the second keeps their method to themselves and retains more flexibility. The system then implicitly rewards silence. A concealed automation, which depends on its creator, also escapes controls, and the company learns less.
Real gains, but difficult to measure
A task completed faster does not automatically free up an equivalent amount of time in the day. Results must be verified, decisions made, and clients sometimes called back. The real gain therefore depends on the complete process, not just the stopwatch applied to one step.
A European Commission survey, conducted in February and March 2026 in eighteen member states, reports that employed people who say they save time thanks to AI estimate this saving at 7.4 hours per month on average. These are user declarations, not an objective record of freed-up time. In a study published in 2025 in the Quarterly Journal of Economics, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond observed an average increase of 15% in the number of problems solved per hour among 5,172 customer support agents using AI assistance. This result sets no rule on working hours or on sharing the benefits. Being more productive at one stage does not mechanically mean working less time.
