Chinese Workers Now Stage AI Productivity Theatre for Their Bosses

In January, Nvidia chief executive Jensen Huang set off what Chinese tech workers now call the Tokenmaxxing wave. His remark, widely circulated online, was that he would be deeply uneasy if a $500,000-a-year engineer burned less than $250,000 worth of tokens in a year. Across China’s big internet companies, the number of tokens consumed by AI tools quickly became a proxy for effort. Use the models heavily and you looked committed; use them lightly and you looked like you were not doing the job.

By June the bill had arrived. Companies that had encouraged near-unlimited use of large language models began to scrutinise the return on that spending. The shift did not calm anyone. Instead, according to a widely shared account of the year inside Chinese tech firms, it sharpened the anxiety: workers now had to demonstrate that the AI tools were paying off, even when the tools were doing little of the work.

The result is a new kind of office performance. Employees keep sessions open, generate long documents, and run prompts whose main purpose is to leave a visible trail of activity. The point is not the output but the record of effort. A worker who quietly does a task without AI risks looking slower than a colleague who wraps the same task in a thick layer of model-generated drafts. In a year when headcount decisions are being made on productivity metrics, that is a real risk.

This is the awkward middle of the AI adoption story. The tools are genuinely useful for drafting, summarising and coding, but the gains are uneven and hard to attribute to any single person. Managers want a measurable number. Token counts, prompt histories and usage dashboards supply one. What they measure is activity, not value, and employees have learned to optimise for the measure rather than the goal.

The pattern echoes earlier enterprise software rollouts, when login frequency and seat counts became the yardstick for tools nobody used well. AI adds a twist: because the models are cheap to call and easy to run in parallel, the performance is almost free to produce. A worker can manufacture a convincing impression of AI-driven efficiency in an afternoon, which makes the metric even less trustworthy.

For Chinese companies, the stakes go beyond office politics. Firms are under pressure to justify heavy spending on computing power and model subscriptions, and many are reorganising teams around the assumption that AI will lift output. If the productivity gains are partly theatrical, those plans rest on sand. Some organisations are responding by shifting to outcome-based reviews, judging finished work rather than tool usage, though the transition is uneven.

There is also a human cost. Workers describe fatigue from maintaining a performance that never ends, and reluctance to admit that a tool did not help. The fear of falling behind, the FOMO that drove the token boom, has not faded; it has simply changed shape. Now the anxiety is about proving returns rather than proving enthusiasm.

The next test comes as companies set budgets for the year ahead. Watch whether Chinese employers retire token and usage dashboards in favour of shipped products, resolved tickets and revenue, and whether employees stop treating AI activity as a form of job security. Until then, the most visible sign of AI adoption in many offices may be the appearance of it.