A 72-Hour Week Is an Honest Filter and a Bad Default

Some American startups now put a blunt question to candidates before the first interview: will you work from 9 a.m. to 9 p.m., six days a week? Wired reported in July^[1] that many U.S. startups are adopting the 72-hour schedule first popularized in mainland China as “996,” and the head of a staffing firm said several of his clients make a yes “a prerequisite for screening candidates.” One startup it described told applicants to expect more than 70 hours a week and served three meals a day at the office, Saturdays included. I would work it. People who love that grind should not be shamed for it. Stated up front, a 72-hour week is an honest filter: applicants see the price, and the employer learns who will pay it. As an industry default, it would be a bad one.

Chinese authorities moved against the schedule years ago. In August 2021, China’s labor ministry and Supreme People’s Court jointly published model cases on working hours and overtime pay^[2] for arbitrators and courts to consult. In the first case,^[3] a courier company fired a probationary worker who refused to work its 996 schedule. Arbitrators ordered compensation for unlawful dismissal, and the published analysis held that terms requiring overtime beyond the legal cap of 36 hours a month are void even when written into a contract. In the second, a tech company required a new hire to sign a contract annex volunteering for its striver program and giving up overtime pay. The analysis called the waiver obviously unfair and void; the case commentary added that prizing hard work is fine but cannot excuse an employer from its legal duties. American federal law makes a different bet. The Fair Labor Standards Act sets no limit on weekly hours for workers 16 or older^[4] and requires time and a half past 40 only for covered, nonexempt workers. That bet holds up only when the hours are posted before anyone signs.

Over 50 working weeks, a 72-hour schedule comes to 3,600 scheduled hours, 1,600 more than a 40-hour schedule. By the OECD’s count,^[5] U.S. workers averaged 1,796 hours actually worked in 2024, part-time and part-year workers included; the OECD warns that differences in national sources and methods make its figures unsuitable for comparing levels across countries. The schedule buys an employer a signal and, maybe, more output. The signal is who says yes. With the same software on every desk, I count that willingness as a sign of dedication, though a slight one: nobody has to work 996 to go from good to great. The output is less sure. Studying British munition workers in World War I, the Stanford economist John Pencavel^[6] found that output rose in proportion to hours below 49 a week, then more slowly, peaking at about 63: “Output at 70 hours differs little from output at 56 hours.” Shells are not software, and Pencavel notes that the threshold may differ for other workers. An employer that asks for 72 hours still has to measure what the extra 32 produce.

Annual hours worked per worker, 2024

The strongest objection is that a filter can harden into a standard. Law firms weigh billable hours in promotions. A 1996 study of two large firms^[8] by Renee Landers, James Rebitzer and Lowell Taylor found evidence for this conclusion: “Reliance upon work hours as an indicator leads to a rat-race equilibrium in which associates work too many hours.” Carried into tech, the same logic could spread 996 across the industry, screen out people with families, health problems or lives outside work, stall the four-day week and leave everyone else wondering whether they are falling behind. I grant the risk. A job ad makes the bargain visible; it cannot guarantee an applicant a different one. What I defend is the choice, not the standard. Not everyone can or wants to work 996, and that is fine. Great talent can come from anywhere, even from two very productive days a week, and an employer that treats hours as the test of talent will turn it away.

I would keep the choice legal and make its terms plain: the hours in the job ad, the pay beside them. When one founder offered his existing staff a 25% raise and double the equity to move to 996, just under 10% signed up,^[1] by his own account as Wired reported it. That is one offer at one company, but it is the kind of choice I mean. Four-day weeks and six-day weeks offered side by side let people choose more freely than either alone. For most people, 40 hours will stay the happy medium, and if new software multiplies what an hour produces, those 40 hours buy more than they used to. The burden of proof belongs to the employer asking for 72.