The US automation provider Zapier has introduced an internal dashboard that records the token consumption of each employee. Tokens are the unit in which AI language models process and account for text. The more an employee works with artificial intelligence, the more tokens he or she uses – and then has to explain why. If it turns out that, in the opinion of management, the tokens were used without good reason, this could have disciplinary consequences, explains Brandon Sammut, Chief AI Transformation Officer at Zapier.
Zapier is not unique when it comes to AI monitoring. However, other companies have different priorities: The consulting firm Accenture now ties the promotion of managers directly to whether they even use the company’s own AI tools.
AI use in companies: quality over quantity
Token tracking or tracking log-ins is being used in more and more companies. Generating 750 words costs around 1,000 tokens – significantly more for code, agent tasks or complex analyses. First of all, there is nothing bad about tracking: Anyone who measures the AI use of their workforce also keeps an eye on the costs. This is legitimate, necessary, and will soon become mandatory anyway with the EU AI Act, according to which companies must control their AI systems. But anyone who believes that such a consumption dashboard can also control the quality of use is confusing the instrument with the goal.
Because the token tracking does not see what is happening behind it. A recruiter, for example, who takes every job advertisement draft from the chat window without checking it, generates the same statistics as a colleague who, with the same token usage, uses the AI as a sparring partner and critically revises its suggestions. Both look the same in the dashboard.
Is the AI zombie apocalypse imminent?
Worse still: Anyone who knows that only consumption is being measured has no reason to think more than necessary and to ask critical questions of the AI or to rebuild the agent because the results were not satisfactory. Token tracking inadvertently sets an incentive in the wrong direction – away from judgment and towards output. The consulting firm Hogan Assessments calls people who work with AI “AI zombies.”
The term is deliberately lurid. The phenomenon behind it is not: According to Hogan Assessments, 75 percent of knowledge workers already use AI tools, but 60 percent of employers worldwide also see critical thinking as a key skills gap. The more AI is used without targeted training in judgment, the larger this gap will become.
AI transformation: Training is essential
There are definitely ways to sensibly manage the use of AI in your own company and train employees in how to use the new technology – without having to track employees’ AI consumption with a sharp pen.
The example of Amadeus Fire shows how this could work. The personnel service provider not only sent its employees to AI training, but also consistently started with management: CEO Robert von Wülfing, COO Monika Wiederhold and numerous other managers went through the “AI Pioneers” program at the Technical University of Munich together and made this visible internally.
Wiederhold is also involved in the “Alliance of Opportunities”, a corporate network that sees AI qualification as a national task. “It is not the best AI that decides, but the best qualifications of the employees,” she says in an interview with Human Resources, which will be published on March 25, 2026.
Token budgets can be managed. Judgment cannot be bought, it has to be trained, and across all hierarchies. Anyone who understands this will control the AI transformation in the company. Those who only count watch as the risks of AI become reality.
Sven Frost is responsible for HR tech, which includes the areas of digitalization, HR software, time and access, SAP and outsourcing. He also writes about recruiting and employer branding. He continues to be responsible for the editorial planning of various special human resources publications.
