Token Pricing Is Starting to Dictate When Programmers Clock In

A couple of days ago, I saw a post on V2EX titled "We've started shift attendance."

According to the post, because Zhipu and DeepSeek charge higher token prices during peak hours, a startup began adjusting attendance: employees take staggered breaks, and weekday lunch breaks have been pushed to after 2 p.m.

AI replacing programmers is still a distant prospect, but it has already started dictating when programmers work.

The comment section was lively too.

Some suggested switching to night shifts outright, and others fast-forwarded straight to "three-shift rotations":

Large models cannot stop.

Someone even said programmers are about to sit at the same table as assembly-line workers.

At first glance, it feels a bit absurd.

But on second thought, this logic is not entirely impossible.

When a factory buys an expensive piece of production equipment, aside from necessary maintenance, it basically never shuts it down casually. Because if the machine stops for an hour, that's an hour's worth of output lost.

This is also why many factories run three shifts: workers can rotate and rest, but the machines must run at full capacity around the clock.

Now, AI computing power is starting to resemble this kind of production equipment.

Many programmers no longer subscribe to just one model; they buy various Coding Plans simultaneously. How many calls they can make per day and how fast the model responds directly affect the project's delivery speed.

At this stage, tokens are no longer just a software expense—they are beginning to carry the flavor of "production capacity."

This is actually nothing new.

Back when early computers were a scarce resource, programmers had to queue up to apply for machine time, even waiting specifically until nighttime to run their programs. Later, personal computers became widespread, computing power became cheap, and people no longer had to revolve around the machine.

Now, AI seems to be bringing that back.

Someone else in the comments said that AI short-drama companies have already started putting all staff on night shifts. I don't know if that's true, but it reminded me of something a friend in a group chat said a few days ago:

The short-drama industry has already transformed from a creative team into an engineering team.

My understanding of engineering is breaking down the process, using AI to transform the workflow, and improving the overall end-to-end process efficiency.

Now it seems that the engineering of AI might also bring back the whole set of production capacity, cost, and even shift scheduling.