Why American Tech Giants Don't Need 996
Why don’t American internet companies need 996, and yet have higher per capita output?
Many people simply attribute it to social culture “involution” and insufficient enforcement of the eight-hour workday, but I don’t think these are the main reasons. Many companies with overseas operations don’t implement 996 for their overseas teams, and they don’t even require clocking in, but their domestic teams still need 996. Why is that?
As a programmer who has some understanding of both domestic and American companies, I believe the main reasons are as follows:
- Higher customer unit price for American companies
- Lower time requirements for manual services from American customers
- Higher code quality of junior programmers in American companies
- Lower management costs for American companies
- Better use of tools and SaaS services by American companies
- Clearer goals and boundaries for American companies
- A few 007 heroes in American companies carrying the load
Higher Customer Unit Price for American Companies
A person with similar abilities, working the same amount of time, is likely to generate more revenue and profit in an American company than in a Chinese company. The reason lies in the customer unit price.
For example, an app targeting consumers (to C) overseas with a subscription fee of $9.9 per month is not considered expensive, but in China, it can only charge 20 RMB. For traditional internet applications, server and other operational costs are not expensive; R&D costs are the major expense. In this case, as long as the user base in China is large enough, the R&D costs can be spread out to make a profit. But for AI large model applications, the inference cost of large models is very high, about 10 to 100 times the server cost of traditional internet applications. If the subscription fee is too low, the more users there are, the more losses there will be.
To B (targeting enterprises) business is similar. For example, the business of generating images from text in bulk can sell for $1 per image overseas, but in China, it can only sell for 1 RMB. To B business is difficult to scale, and the customer unit price determines whether the business is profitable or not.
There are two reasons for the high customer unit price of American companies. The first is geopolitics; American companies control the global high-net-worth market, which is not easy for Chinese companies to enter. The second is technology-driven; American companies use stronger technological competitiveness to occupy some early markets or high-barrier markets.
I believe that the higher customer unit price is the most important reason why most American companies don’t need 996. With a higher customer unit price, they can hire higher-paid employees. The higher code quality, lower management costs, and better use of tools and SaaS services mentioned above are actually implications of stronger average employee capabilities.
Lower Time Requirements for Manual Services from American Customers
When I first visited the United States, I felt that the overall pace of American society is significantly slower than that of China. As a service provider, it’s certainly comfortable to slack off, but as a client, it’s not a pleasant experience.
We needed to deploy servers to a data center. What could be done in a week in China took two months in the United States. Data center staff only respond during working hours on weekdays, and if a response is needed outside of working hours, a high fee is charged.
Online services are the same. Websites for registering domain names often go down for maintenance, and even large SaaS services like CloudFlare have obvious bugs. Such products would not be acceptable in China. The delay in operations like starting and stopping virtual machines and changing configurations on the Google Cloud console is higher than that of domestic cloud providers. When I worked for Huawei Cloud, reducing these operation delays was a KPI for the team. I was surprised that a large cloud provider like Google Cloud had technical indicators that were not as high as Huawei Cloud.
Overseas, email communication is more common, while in China, instant messaging tools or phone communication are more common. This also reflects the difference in time requirements for manual services. In China, a WeChat message or phone call usually requires immediate resolution, regardless of whether it is during working hours. Email communication allows for longer delays, with the default being to handle it during working hours.
With the overall slower pace of society, customers are more tolerant of slow manual services, and as a service provider, there is naturally less need for 996.
Higher Code Quality of Junior Programmers in American Companies
Many people ignore the level of individuals when calculating indicators like per capita output and per capita salary. For example, the annual salary of a junior programmer in Beijing is only 200,000 RMB, while in the Bay Area, the starting salary for a programmer is basically $150,000. Therefore, they conclude that programmers in China who move to the United States will definitely see a significant improvement in their standard of living.
But they overlook one point: a junior programmer with a 200,000 RMB annual salary in China may not be able to find a job in a major company in the Bay Area. This is not just a language issue but more importantly, a capability issue.
In my experience, the entry-level threshold for junior programmers at several well-known companies in the Bay Area is roughly equivalent to that of a programmer with an annual salary of 500,000 RMB in Beijing. More than half of the programmers in major companies in China may not reach this salary level.
There is a saying that jumping from a major overseas company to a major domestic company can result in a promotion of about three levels. This is because the entry-level threshold for junior programmers is different. Going from 200,000 to 500,000 is roughly three levels. Therefore, those doing basic development work in major overseas companies must take on some management work when they move to major domestic companies.
The direct consequence of low-level junior programmers is poor code quality, and many domestic companies have only recently started to pay attention to code quality. For example, Huawei has been promoting software trustworthiness since 2019, requiring R&D personnel to pass coding ability exams similar to LeetCode and software engineering foundation exams. Code repositories must use best practices such as Pull Request, Peer Review, Committer responsibility system, code style checks, and CI/CD. All R&D personnel, regardless of their position, must submit a certain amount of code each year. Committer’s code quality opinions are important references for employee promotion and performance evaluation.
Many project managers and leaders in domestic companies do not understand technology. Junior programmers only need to implement functions quickly, regardless of whether the code is a mess. As a result, many companies face significant resistance when promoting code quality improvement plans. For example, some Huawei employees often complain that their code often fails to pass code style checks and unit test gates. After finally modifying the code to pass these gates, the Reviewer gives a few more suggestions, and the code is sent back. The time spent submitting code to the repository is longer than the development time. However, as employees improve their code style and deploy code style check tools locally, the time required to submit code to the repository gradually decreases.
The consequence of poor code quality is more bugs, requiring overtime to fix them, turning coding, a mental labor, into physical labor. For example, some teams need to accept development results before leaving work. When testing finds too many bugs, developers and testers have to work overtime to fix them, often staying late. Since passing the test means leaving work earlier, the code written during overtime is naturally patch upon patch, further deteriorating code quality. This is one reason why domestic companies often have 996.
Product quality is overseen by product managers and testers. Even if the code quality is poor, using 996 to pile up time can always fix it so that users don’t notice. But the developer ecosystem is directly affected by code quality. For example, many domestic large model API documents are incomplete, some sample codes don’t even have syntax highlighting, some sample codes don’t work (because the API was updated but the sample code wasn’t), some Python libraries don’t support asynchronous operations, and some appear to support asynchronous operations but are implemented synchronously internally. Some AI service SDKs have open-source code, but when you open it, you find that even the indentation is not uniform, the code is filled with various inexplicable variable names and comments, and a function can be hundreds of lines with five levels of indentation. A significant reason why many domestic products have a poor developer ecosystem is poor code quality.
Lower Management Costs for American Companies
Domestic companies have higher management costs, mainly reflected in low meeting efficiency, occupying a lot of development time, and some teams even find that they are in meetings all day and only have time to write code at night.
I believe there are three main reasons:
First, low communication efficiency and a lot of wrangling. Some people give long speeches, and some discussions go in circles without reaching a conclusion, only to bring up the same issues again later. Since discussions often exceed two hours, it’s hard for participants to stay focused. Many people learn to slack off during meetings, resulting in some people missing the discussion conclusions, further increasing communication costs.
One of my leaders at Huawei came from a foreign company and spoke very concisely, only explaining things once, often mixing in some English technical terms. Some Huawei colleagues would come to me after the meeting to ask what the leader meant. Over time, everyone paid close attention during meetings, and meeting efficiency improved.
Second, junior programmers with poor foundations need hand-holding, which is known as “nanny-style development” or “security guard-style development.” During meetings, if the technical solution is unreasonable, senior programmers naturally need to correct the issues, but junior programmers with poor foundations often don’t understand the first time and need repeated explanations. This naturally reduces meeting efficiency.
Third, domestic companies have more management levels, not because of the difference between “flat management” and “pyramid management,” but because there are more junior programmers, and programmers of the same level in domestic companies may be three levels higher than in American companies. The higher the management level, the more time is spent on management.
In contrast, most American companies have fewer meetings, lower management costs, and longer effective working hours for programmers.
Lower management costs are also an implication of higher average employee quality. High-level employees generally have strong self-motivation, good time management skills, and reasonable technical solution designs, leading to higher technical communication efficiency and significantly lower management costs.
Better Use of Tools and SaaS Services by American Companies
The full use of tools and SaaS is one of the important reasons why American companies remain efficient.
American companies are good at using existing mature tools and SaaS (Software as a Service) platforms to improve work efficiency and reduce repetitive labor. For example, engineering teams widely use GitHub, Jira, Slack, Notion, and other tools for project management and team collaboration, greatly reducing internal communication and task tracking costs. In contrast, domestic companies often prefer self-developed or locally deployed solutions because self-development can “boost KPIs” and appear innovative, but in fact, it is inefficient and redundant.
Many tools that can improve programmer development efficiency are paid. Some domestic companies would rather hire an additional programmer with a monthly salary of tens of thousands than spend a few hundred yuan a month on development tools. For example, an editor like Cursor costs $20 a month, which many domestic companies find too expensive, even though a programmer’s hourly wage is more than $20.
SaaS services are even more of a key dividing line. I used the API of a domestic large model vendor, spending 5000 RMB a month, and they treated me as a major customer. I was surprised that spending only 5000 RMB made me a major customer. They said most domestic vendors buy GPUs to deploy open-source models themselves and don’t like using APIs.
American companies like to use third-party paid services, such as vector databases and large model APIs, and are willing to pay for better tools and services in exchange for higher productivity. In contrast, domestic companies, especially smaller ones, often choose local deployment or open-source projects to save costs. Although this seems cheaper, it is actually costly in the long run because maintenance and time costs are overlooked.