The Rise of the Purple Collar Worker

On June 5, 2026, the Financial Times reported that the National Security Agency was using the most capable cyber model Anthropic has ever built, and that half a dozen of the company’s engineers had been embedded inside the agency to make it work. The most sophisticated version of the argument against human work holds that it is not the jobs that disappear but their value, that machine capability will so thoroughly compress the price of cognitive labor that the human becomes uneconomical even where the human remains possible. Here was the most advanced artificial intelligence deployment in American history, and it ran on six human beings doing a job the American labor system cannot name.

Almost no one asked what those six people do all day. They translate intent into instruction, judge what the machine produces, and adapt it when the mission shifts. The model supplies capability. The humans supply direction, and there is no occupational code for this work in the United States, no credential, no training pipeline. The gap sits at the center of national security.

The job has a name. The name is in Mandarin.

紫领. Purple collar. The color sits between the blue collar of the production line and the white collar of the office, because the worker sits between them, and over the past two years the term has moved from management literature into the formal vocabulary of Chinese labor policy. Its fullest treatment comes from researchers at Renmin University, in fieldwork conducted with Lenovo and TCL, who define the role through two four-character phrases: 一专多能, one specialty and many capabilities, and 手脑并用, hands and brain together. That second phrase is the whole thesis compressed into four characters. The competency model they built names six qualities, and not one is technical: judgment, integration, communication, the willingness to keep learning. Coding does not appear. Machine learning does not appear. What the model describes is the portion of the work a machine cannot do without a person standing beside it, and the six engineers at Fort Meade would recognize every factor on the list.

Around that description, a state apparatus has been assembling for seven years. Since 2019, China’s labor ministry has formally recognized more than ninety new occupations, including artificial intelligence trainer, a role with a national standard, three certification levels, and an officially estimated shortage of 1.2 million people. One of those trainers, Hu Pingping, practiced as a gastroenterologist before she began teaching medical knowledge to healthcare models that now support rural doctors. She did not stop being a physician. Her knowledge stayed, and the form of the work changed, and a credential was waiting for her when it did.

The scale has no American analogue. One telecommunications company, China Telecom, reported training and certifying 167,000 employees to work alongside artificial intelligence by late 2025, each on a personalized learning path with machine coaching built into the platform. The state oil company runs a system that simulates colleagues, supervisors, and clients, so that employees rehearse collaboration with a machine partner as the very method of learning it. Beginning this year, AI literacy becomes part of national teacher certification. Last spring I translated a set of vocational documents from a single Beijing district that no English outlet has covered, describing a school given its own corporate-sponsored AI engineering college and reporting that ninety-three percent of its teachers had completed retraining. They read like construction logs. The transformation is far enough along that its paperwork has become boring.

A fair reader should discount some of this, and I say that as someone whose work involves discounting Chinese institutional claims for a living. Certification counts can overstate real skill. Yet the occupational codes are published documents, verifiable independent of any throughput claim, and so are the national standards, the competency model, and the demand projection. That projection is the number that should have crossed the Pacific by now: China forecasts demand for more than thirty-one million purple collar workers by 2035, nearly a quarter of its manufacturing workforce. Discount the training statistics by half and the gap is still hundreds to one. You only project demand for something you intend to supply.

Something deeper than policy separates the two approaches, and it is worth naming directly. The question at stake is whether intelligence stays embodied. The purple collar worker is not a transitional figure to be automated away once the technology matures. In the Chinese construction, the worker is the mature form, the point where machine capability meets human hands, human context, and human accountability and is made to produce something neither achieves alone. The bet could be wrong. China’s own researchers open their founding purple collar report by citing a projection that 200 to 300 million Chinese workers may be replaced by artificial intelligence by 2049, and they build the training system anyway. They are not naive about displacement. They have decided that the answer to displacement is to train the worker who survives it.

The Western posture offers a different answer. Elon Musk has promised a humanoid in every factory. Sam Altman has projected an economy where a single person with an AI co-pilot runs a billion-dollar company alone. The vision these leaders share is an end to scarcity that arrives precisely because human labor has been engineered out of the loop. The most cited American study of the question found that roughly one in five workers could see at least half their tasks reshaped by these tools, a figure that resolves to about thirty-one million people. The same magnitude as China’s number, read in the opposite spirit. China saw thirty-one million workers to build. America saw thirty-one million workers to replace. It would be wrong to say China rejects automation; it installs more industrial robots than the rest of the world combined. The distinction is that China is building the robots and naming the humans who will direct them, treating machine and worker as a single system, while the Western reflex is to build the robots instead of the humans. These are not two forecasts of the same event. They are two decisions about what the event is for. Meanwhile the United States Office of Personnel Management, in a document framed around winning the global race for AI dominance, committed to hiring 250 AI fellows across the entire federal government. China Telecom certified 167,000 at one company. The ratio is 668 to one.

The clearest evidence sits inside the doom itself. The same week as the Fort Meade story, Anthropic published a report warning that artificial intelligence may soon improve itself with no human in the loop, and grounded the warning in a chart showing code produced per engineer rising to eight times its former rate. Read the axis. Per engineer. The figure offered as proof that the human is obsolete carries the human in its denominator, because output with no person in it is a chart no one has published. What that chart actually shows is human work, multiplied, by people doing exactly what the purple collar worker does.

This essay is the third in a series I have published in this venue over the past month. The first held that the self is relational, that we have always come to know ourselves through others, and that a machine can join that process without dissolving our authorship of it. The second held that human knowledge is produced the same way, through collaboration, and that our institutions have built detection regimes and plagiarism classifiers that mistake the visible trace of that collaboration for fraud. Both essays were building toward a person. The purple collar worker is who they were building toward. Everything they argued in the abstract now has a body, a wage, an occupational code, and a national training system on one side of the Pacific. On the other side, Gene Walinski, who is seventy years old and works at Home Depot, uses an AI assistant on his phone to do his job better and keeps the phone in his pocket. He is Hu Pingping’s American twin, separated from her by nothing except the institutions around him. The culture that should have named his work is too busy proclaiming its obsolescence to notice he is already doing it.

The components of an answer are knowable, because another country just demonstrated the full sequence, from name to projection to standard to school, and every component has an American institution already equipped to carry it. It is not a question of capability. It is a question of what each civilization believes a human being is for.

—  —  —

Russ Wilcox is the founder and CEO of ArtifexAI and the publisher of The Pacific Divide, where he writes on artificial intelligence, institutions, and the contest over cognitive sovereignty. He is a contributing analyst at the Jamestown Foundation’s China Brief, has published in The Diplomat, and has spoken at the World Economic Forum in Davos across three consecutive years. He reads the Chinese, Western, and classical sources on these questions in their own traditions and is at work on a book about the self in the age of machines that would author it.