Nobody announced it. There was no press release, no reorganisation memo, no all-hands slide. A contractor in Hangzhou simply stopped getting work, and the company she worked for quietly stopped hiring graduates.
Meanwhile her manager had started ranking the team by something with nothing to do with the work: how many tokens (词元) each person burned through the AI tools the company had ordered everyone to use.
Key takeaways
- Chinese firms are implementing small, dispersed AI-linked job cuts rather than large layoff rounds, with graduate hiring already reduced at one large internet company (Reuters analysis, 10 June 2026).
- In at least three cases Chinese courts have ruled against employers who dismissed workers in order to replace them with AI (Reuters; Legal Daily, 28 April 2026).
- Article 41 of the Labour Contract Law (劳动合同法) requires reporting to the labour administration before cutting 20 or more staff, or fewer than 20 but at least 10% of the workforce (Ministry of Human Resources and Social Security).
- In a Guangzhou case, a court ordered an employer to pay the double-difference in illegal-dismissal compensation of more than RMB 35,000 (≈ US$4,900 / HK$38,000) after it abolished a graphic designer's role and installed an AI drawing system (Guangzhou Political-Legal Network, July 2026).
- BOSS Zhipin reported that monthly new AI-related job postings grew 74.1% year on year in 2025, even as Citibank estimated 9.6% of Chinese jobs — roughly 70 million — face high risk of AI-driven displacement, rising to 13.6% for workers in their 20s.
Why the cuts are small on purpose
The Reuters reporting describes a pattern, not an event. Nine workers from technology, entertainment and advertising described employers that started letting individual roles go — often contractors — after ordering staff to use AI agents, while reducing graduate intake. An engineer at Alibaba's cloud division said AI-driven headcount reductions had begun in parts of the company but would likely unfold through gradual attrition rather than one mass round of layoffs. Alibaba did not respond to a request for comment.
Two constraints shape that behaviour, and both are specific to China. The first is arithmetic. Article 41 of the Labour Contract Law, published by the Ministry of Human Resources and Social Security, sets the threshold: where an employer needs to cut 20 or more people, or fewer than 20 but at least 10% of its total workforce, it must explain the plan to the union or the entire staff 30 days in advance, hear their views, and report the plan to the labour administration. Below that line, no such procedure is triggered.
The second is the labour courts. Reuters reported that Chinese courts have in at least three cases ruled against companies that dismissed workers specifically to replace them with AI. Legal Daily's survey of rulings in Beijing, Shanghai and Jiangsu found that courts generally treat technology upgrade and role adjustment as foreseeable management decisions, not the kind of "major change in objective circumstances" permitting unilateral termination.
What the Guangzhou ruling actually says
The clearest worked example is a design job in Guangzhou. A graphic designer, Wei (魏某), had worked at an intelligent-technology company for more than three years. The company introduced an AI drawing system and reorganised, abolishing the graphic designer position and creating a "full-service designer" (全案设计师) role in its place. It negotiated only over severance, paying economic compensation of more than RMB 33,000 plus a payment in lieu of notice of more than RMB 8,000, then issued a termination notice citing a major change in objective circumstances.
The Guangzhou Nansha District Court found for Wei, ordering the company to pay the double-difference in illegal-dismissal compensation of more than RMB 35,000. The Guangzhou Intermediate Court dismissed the company's appeal and upheld the ruling. Judge Chen Shiyuan (陈诗媛) framed the reasoning: whether AI substitution qualifies as a major change in objective circumstances cannot be decided in general terms, and the event must originate outside the employer's control. Here the company had itself proposed moving Wei into the full-service designer role, so its need for design skills had not disappeared. It had also offered only severance, without evidence of substantive discussion about transfer or training. That is a narrower and more checkable standard than "AI will replace jobs": a company wanting to cut must show it exhausted internal options first.
The metric nobody designed
The second thread is stranger than the layoffs. Some employers have begun measuring whether staff have adopted AI at all, and Reuters reported that one manager at a large Chinese technology company started ranking employees by token consumption in March, with the figure linked to performance reviews and promotion prospects.
Tokens are a unit of AI compute consumption. Counting them estimates how much of the tool people used. It does not measure output quality, and the Reuters source said as much. An engineer quoted there called the arrangement "relatively forced."The demand side of the same labour market is booming. BOSS Zhipin reported at its January 2026 partner conference that monthly new AI-related job postings on its platform grew 74.1% year on year in 2025, after 36.5% in 2024 and 8.5% in 2023. Non-technical AI demand grew fastest: sales, administrative and business roles up 682%, and legal, consulting and human resources roles up more than 200%.
Against that, Citibank estimated that AI could affect about 31% of Chinese jobs, with 9.6% — roughly 70 million positions — facing direct replacement risk, rising to 13.6% for workers in their 20s. The wider market is tight: China's education ministry put the 2026 college graduate cohort at 12.7 million, a record. According to Li An, chief scientist at BrainNet (脑机网), the token metric is the part to watch, because it moves management attention from what a role produces to whether a person has complied with a tool mandate — a shift measurable long before it is visible in any output.
Honest limitations
Most of the reporting rests on nine workers interviewed by Reuters, several anonymous and one describing a contractor arrangement invisible in any filing. Court decisions are public and specific, but they bind the parties in those cases rather than setting a general rule, and three adverse rulings is not a count of every decision reached. The displacement figures are one bank's estimates, not official statistics, and job-posting growth on one platform does not describe the whole labour market. Nothing here establishes that attrition-led trimming has produced net job losses at any named company, and the company behind the token-ranking example is not identified.
What readers can do now
- If your performance review now references AI usage, ask in writing how token consumption is weighted against output quality, and what output is measured.
- If you are restructuring roles, note the Guangzhou sequence: a newly created adjacent role, substantive discussion of transfer and training, and a report to the labour administration once Article 41's threshold is met.
- If you are a graduate, note that the 74.1% growth in AI postings sits in non-technical and composite roles, and that the hiring contraction at large internet firms is on the other side of the same market.