Summary of the Key Points in Plain Language
This article hits on the sense of community that almost all professionals have experienced in the past two years: used to, it would take three afternoons to draft a report, but now AI can produce it in just 10 minutes. You think you can finally leave work on time, only to have your boss casually say, “This version is good, but produce two more for comparison.”
Eighty percent of professionals worldwide believe that AI has increased their work efficiency, but the actual data shows that the average daily working hours have decreased by less than 10 minutes, while weekend overtime has increased by 40%. In China, 73% of employees use AI multiple times a week, which is 1.5 times the global average. However, the average weekly working hours remain above 48 hours, effectively meaning an extra day of work has been added. The time saved by AI hasn’t increased employees’ rest time; instead, it has secretly raised the bar for work quality, leading to more tasks such as revising ineffective AI-generated content and attending unexpected meetings. As a result, everyone finds themselves more exhausted despite the supposed increased efficiency. The problem isn’t with AI itself but with the fact that most companies don’t know how to allocate the additional capacity it provides, simply using the tool to assign more work.
---
Detailed Explanation of Each Point
1. The Counterintuitive Fact: The 10 Minutes Saved by AI Are Taken Away by Weekend Overtime
Many thought that with the widespread use of AI, working hours would gradually decrease, similar to how assembly lines reduced labor. However, the data shows that the average daily working hours have only decreased by 9 minutes and 36 seconds—equivalent to the time it takes to drink two more cups of coffee in the morning. The increase in weekend overtime has more than made up for this reduction. Moreover, although the official statistics show a slight decrease in nominal working hours, the work intensity has increased significantly: communication time has increased by 145%, and the ability to focus on tasks without distractions has dropped to a three-year low. It’s like building an eight-lane highway; although cars can travel faster, everyone is constantly changing lanes, using navigation, and avoiding traffic jams, resulting in more fatigue than in the pre-highway era. The boundary between workdays and weekends is blurring, and no one notifies you of the overtime, so you end up working extra hours without realizing it.
2. The Hidden Competition: The Bar for “Good Work” Has Been Secretly Raised
The sneaky aspect of AI is that it changes the standard for what counts as “completed work.” In the past, using a film camera, producing ten photos and selecting one good one was considered acceptable. With digital cameras, not taking 100 photos and picking the best one would be considered a lack of skills. Now, with AI, the cost of producing any content is almost zero. What used to be a well-organized document is now expected to include interactive visuals and multiple versions. No one has officially changed the rules, but everyone assumes that producing more versions is just part of the job. This is the “Jevons Paradox” discovered over a century ago: while the steam engine increased coal efficiency, coal consumption increased because previously unaffordable tasks became possible. With AI, the time required to complete a report has dropped from three hours to 30 minutes, but the frequency of boss requests has increased from once a week to three times a day. The faster your efficiency improves, the more new tasks arise. Those who use AI daily feel their productivity has decreased by four times compared to those who don’t—this isn’t because you’re working slower, but because the standards have risen faster than your output.
3. The 2 Hours You Save with AI End Up Being Used by Your Colleagues
A new type of “work waste” has emerged: content generated by AI that looks well-organized but is actually useless. Surveys show that 15% of the work received by professionals is of this quality, and each piece of such content takes nearly two hours to process. You either have to rewrite it entirely, ask the sender to revise it, or use it reluctantly, which results in additional work. For example, if you save two hours writing a report with AI, your colleague has to spend two hours fixing its flaws. The total workload hasn’t decreased; it’s just shifted from the writing stage to acceptance, error correction, and additional tasks. Everyone receives last-minute meeting notifications, responds to emails at midnight, and participates in cross-timezone meetings, all due to this type of ineffective work. These tasks are not counted in official statistics but still require real effort.
4. No One Dares to Lower the Raised Standards
The most problematic aspect is that no one wants to lower the standards set by AI. Ninety-six percent of executives believe AI will double employee efficiency, but 77% of employees feel that it has increased their workload. In reality, 95% of companies have not seen actual profit gains from AI; instead, the workload has increased. This is because once the “faster conveyor belt” of AI is in motion, no one remembers its original speed. If a company were to lower the standards (e.g., requiring fewer versions or no interactive visuals), it would be seen as a defeat in the market. No management team wants to take that responsibility, so standards continue to rise, and the extra work falls on frontline employees. Product managers have to code with AI, data analysts have to handle additional tasks, and job descriptions remain unchanged, turning “you can use AI” into “you must use AI.”
5. The Solution Lies in Understanding What “Completing Work” Really Means
Most AI training programs are ineffective. Instead of focusing on how much work AI can do, companies should first define what counts as “completed work.” For example, decide how many versions of a report are needed, under what circumstances no further revisions are required, and what deliverables need additional visuals. Set these standards before starting work, not after multiple versions are produced. Even cutting out low-value tasks should be part of performance evaluations, rather than simply measuring the amount of work done. For individuals, it’s simple to clarify the standards before taking on a task: ask what the final version is for (e.g., a report for the boss or just a draft for internal discussion) and define the boundaries clearly. This can save a lot of time.
In the AI era, the most valuable skill is not working faster but clearly defining what “completed work” means.