Beijing is writing down the workers it spent two decades training and calling it modernization.
Here is what the Associated Press filed on August 24, with the named policy instruments attached.
Fei Zhaojun worked as a computer programmer in Beijing. His supervisor asked, two weeks before the dismissals, whether artificial intelligence could soon replace humans in coding jobs. Two weeks later Fei was laid off along with about 160 colleagues from the same workplace (Chan Ho-him, “Chinese workers are adapting as AI job takeover worries grow,” Associated Press, August 24, 2026).
The dismissal pattern extends beyond the single workplace. The AP wire reports AI displacement extending from computer programming to script writing to physical tasks, with each segment of the labor market exhibiting the same pattern. State policy sets the rate. Firms comply. Workers exit.
Three policy instruments are doing the acceleration the wire describes, and naming them turns the analytical frame.
First, the Made in China 2025 plan released by the State Council in May 2015 set a ten-year roadmap prioritizing ten advanced-manufacturing sectors, with robotics, information technology, and artificial intelligence explicitly named. The roadmap carried subsidies, procurement preferences, and provincial-industrial-park siting decisions. The industries the plan named are the industries the AP wire documents as the first displacement sites. The fit is exact.
Second, the New Generation Artificial Intelligence Development Plan released by the State Council in July 2017 set the operational target: China the world’s premier AI innovation center by 2030. The plan named as the priority technology the same technology the prior plan had identified as the sector to subsidize. The two instruments compose the policy stack the AP wire is reporting on.
Third, the NDRC (National Development and Reform Commission) and MIIT (Ministry of Industry and Information Technology) implementation documents channel provincial and municipal subsidies to firms deploying AI applications and robotics, including local-government incentives documented in subsequent Five-Year Plan annexes. The wire’s finding that “government policies encourage individuals and businesses to deploy AI applications and robotics across all aspects of life” is the public-facing description of the deployment-subsidy system those operating-level documents specify. The policy instruments are named. The behavior is documented.
Three things are worth establishing before the talking points arrive.
First, displacement velocity is a policy choice, not a market outcome, and the productivity convention built to score market outcomes is not adequate to score it. The National Bureau of Statistics methodology for measuring industrial-output-per-worker — the productivity metric the State Council and the NBS itself treats as the headline indicator — counts the gain at the firm and does not count the loss at the household. The convention is unchanged from the period when the same workers were being added to the payroll by state-directed human-capital investment: school subsidies, university enrollment targets, and the engineering pipeline documented across successive Five-Year Plans since the early 2000s. The workers in Beijing’s programming sector were trained at public expense to staff the knowledge economy the same State Council is now accelerating past. The productivity convention books the gain. The convention does not book the human-capital write-down at the household. The convention is what the firm wants the convention to be.
Second, the macroeconomic risk the wire identifies is the standard industrial-policy externality applied to a labor stock the state built. Some economists quoted in the AP wire say the disruption could undermine the strength of China’s economy, the world’s second largest. The result is the textbook industrial-policy externality — meaning the cost the displacing operation pushes off its own books and onto workers, households, and the demand side: gains retained at the firm and the policy apparatus, costs pushed onto the displaced workers and the households that supported their training, and a demand-side feedback channel running from suppressed household income through weakening consumption to eroding domestic demand. The 2017 AI Plan’s stated priority on building an indigenous AI ecosystem depended in part on the demand-side premise that domestic consumption would continue to grow. The displacement the wire documents erodes the premise the plan was built on. The feedback runs the wrong direction.
Third, the unemployment-insurance backstop the operation requires is not documented in the wire. Urban-employee unemployment insurance in China is administered through the Ministry of Human Resources and Social Security (MOHRSS) under the 2010 Social Insurance Law and the unemployment-insurance regulations issued by the State Council. The replacement-rate schedule and the duration caps are publicly specified. The AP wire does not document whether Fei Zhaojun or the 160 colleagues dismissed with him are receiving any unemployment-insurance payment, any retraining subsidy, or any income-support bridge during the displacement period the policy itself accelerated. The labor-market adjustment velocity the policy stack specifies is faster than the social-insurance system is built to absorb. The Ministry is named. The mechanism is named. Whether the mechanism is operating at scale in this displacement is the question the wire leaves open. The question is the finding.
Two observations the AP wire makes worth preserving in the record.
The wire documents that the dismissals followed the supervisor’s question about AI substitution by exactly two weeks. The two-week interval was the firm’s adjustment time to a state-set policy rate. The supervisor’s question was the firm’s internal rehearsal of a decision the policy stack had already prepared.
The wire notes that workers like Fei are figuring out how to adapt. Adaptation is the press-release framing for what is, in the fiscal-mechanics framing, a transfer of trained human capital from the payroll to the unemployment line, executed by a policy choice, documented in the wire, and not accompanied by the income-support bridge the social-insurance system is supposed to provide. The cost is on the worker. The benefit is on the firm and the policy apparatus.
Two recommendations in the affirmative tradition this column applies to industrial-policy choices that externalize cost onto the workers whose training the policy built on.
First, the National Bureau of Statistics productivity data and the MOHRSS unemployment-insurance data need to be published in the same release cycle, on a comparable unit, so that the household income loss the displacement produces is scored against the firm’s productivity gain at the same authority and on the same publication date. Productivity gains at the firm scored against human-capital write-downs at the household are not a methodological refinement. They are what honest accounting of a state-directed displacement operation looks like.
Second, the displacement velocity the policy instruments specify is faster than the social-insurance system is built to absorb. Industrial-policy accelerants that target a country’s own human-capital investment without an income-support bridge for the displaced produce the textbook labor-extraction result. The pattern is documented. The remedy is documented. The choice to omit the remedy is a policy choice.
The AP wire is dated August 24, 2026. The methodology does not require Beijing to acknowledge the cost. The accounting convention does.