Startup Makes Programmers Work Shifts to Save on AI Tokens
A startup with about a dozen employees has started putting its programmers on shift work — not because servers need 24/7 monitoring, but to save on AI tokens. An employee described the arrangement on V2EX, a long-running Chinese tech forum, saying the company relies heavily on coding services from DeepSeek and Zhipu. Since those models charge more, or burn through quotas faster, during peak hours, the company simply scheduled everyone around the cheap windows.
Under the new attendance policy, each employee has to take one day off during the workweek and one day on the weekend — they can pick the days, but the plan must be set in advance. Weekday lunch breaks have also been pushed to after 2 p.m. The original poster later revealed that the company has only around a dozen people, yet subscribes to several coding plans at once: MiniMax, GLM, DeepSeek, and Volcano Engine.
DeepSeek's latest API pricing sets weekday peak rates at twice the off-peak price, and starting August 23, weekends are now discounted all day. Zhipu's coding plan also calculates quota by the clock, charging just 50% of the base point cost for model calls during off-peak hours.
Commenters were quick to note the irony: humans are now rearranging their working hours around AI's peak-pricing schedule. One user chimed in that their own boss has heard about the approach and is considering copying it.