Proponents of AI say it is making white-collar jobs more efficient. The eating habits of employees are I’ll tell you another story.
When companies like Atlassian, Block, and Snap announced mass layoffs earlier this year, they cited AI as a common factor. They said the technology allowed them to adjust the composition of their workforce.
But data from Sharebite, a corporate meal benefits and delivery platform founded in 2015, suggests that AI may be leading some workers to log longer hours rather than shorter hours. The company said the number of customer orders placed on Saturdays in the first quarter more than doubled compared to the same period in 2025. Orders placed after 6pm on weekdays and weekends increased by 57% in that time, and overall user growth increased by 36%.
This imbalance suggests a disproportionate increase in activities outside of traditional working hours. This is a modern echo of the Pentagon Pizza Theory, the idea that a spike in late-night food orders indicates government employees are working overtime ahead of a major event.
The findings come as many companies are increasing pressure on employees to adopt AI, including by tying its use to performance reviews that influence pay raises and promotions.
Sharebite CEO Dilip Rao, a former Wall Street investment banker, told Business Insider that he believes there is a connection between the surge in after-hours orders among his clients (hundreds of companies of various sizes, primarily in the technology and financial sectors) and the AI boom.
“Based on data across our enterprise customer base over the past 12 to 18 months, people are not working less,” he said. “If anything, the activity is late in the day and extends into the weekend.”
Rao’s paper falls in line with a growing body of research showing that AI is increasing the workload of professionals. For example, a study published in February in Harvard Business Review found that employees who use AI take on a wider range of tasks and spend more time than those who don’t.
A separate report published this year by researchers at the University of California, Berkeley, reached a similar conclusion. Based on an eight-month study of employees at small U.S. technology companies, researchers observed that employees were multitasking more, taking on a wider range of responsibilities, and working longer hours.
Similarly, a 2025 report published by the National Bureau of Economic Research links increased exposure to AI to longer working hours and decreased leisure time. The authors said this is primarily because AI complements human labor, rather than replacing it.
more work, more food
According to the Pentagon Pizza Theory, often credited to pizza shop owners in the early ’90s, federal employees who suddenly order a large number of pies must be working overtime to prepare something important, or so it seems.
Sharebite’s food delivery data could provide similar insights into the impact of AI on workers, Rao said, linking the spike in evening orders to companies’ adoption of the technology.
“This seems to be more of a change in the way work is done than a reduction in jobs themselves,” Rao said, adding that the data highlights how essential human talent remains as companies adopt AI.
Workers may be working long hours for reasons other than or in addition to AI. In recent years, employers have increasingly prioritized tangible results over loyalty and performance expectations. These changes, along with cuts to benefits and, in some cases, core benefits, reflect a labor market in which workers are less able to push back on overtime demands than before.
There are several reasons why AI can help workers extend their workdays. One is that this technology can cause hallucinations and introduce errors in its output. After using it to copy, write code, or create images, users must review the results to ensure accuracy, and correcting mistakes requires additional time.
Neil Thompson, an innovation scholar at the Massachusetts Institute of Technology’s Computer Science and AI Lab, said AI may be causing workers to spend more time on the job because they haven’t yet learned how to integrate it into their workflows.
“There is usually a transition period where organizational processes need to change,” he said. “It will be less efficient at first.”