Companies are beginning to track how employees use artificial intelligence tools, turning what was once an invisible productivity boost into a measurable — and increasingly scrutinised — workplace metric.
According to a report by The Wall Street Journal, firms including automation platform Zapier are now monitoring “token” usage — the units that measure how much computing power is consumed when employees interact with AI systems.
The shift reflects a broader change in how businesses evaluate AI: not just as a productivity tool, but as a cost centre that can be tracked, optimised, and, in some cases, controlled.
“We have this new kind of line item,” Brandon Sammut, Zapier’s chief AI transformation officer, told The Wall Street Journal, referring to the cost of AI-assisted work.
Every prompt, response, or automated task performed by AI consumes tokens, effectively allowing companies to quantify how much AI employees are using.
For text-based tasks, generating around 750 words uses roughly 1,000 tokens, with more complex workloads — such as coding or autonomous agents — consuming significantly more.
As companies scale adoption, some are beginning to compare usage across employees. At Zapier, unusually high token consumption can prompt closer scrutiny to determine whether it reflects inefficiency or high-impact output.
“We start to draw conclusions… whether that’s a golden pattern we want to multiply… or whether it’s an anti-pattern,” Sammut said in the report.
The push to measure AI usage comes as companies try to balance productivity gains with rising costs.
Brian Jabarian, a researcher at the University of Chicago’s Booth School of Business, told The Wall Street Journal that the relationship between AI usage and productivity is not straightforward.
“Everyone thought that you just use AI tokens, and you have an increase of productivity,” he said. “But the reality is more complicated.”
In some cases, high AI usage can translate into significant value. At AI startup Vercel, engineers used AI agents to generate a new codebase in a day's work that would typically take weeks — at a cost of around $10,000 in tokens.
“It’s a little bit like giving people a fire hose of fuel,” CEO Guillermo Rauch said in the report.
While many companies are still focused on encouraging employees to use AI tools, those further along are beginning to introduce controls.
At Kumo AI, token usage is now tracked at the individual level, with some engineers effectively using AI agents as always-on assistants. Other firms are beginning to explore limits on usage or restrictions on which models can be used for specific tasks.
Mark Hull, founder of Exceeds AI, told The Wall Street Journal that companies are likely to develop governance frameworks around AI usage as costs rise.
The growing use of token tracking signals a shift in how work is measured in an AI-driven workplace.
What began as a tool to enhance productivity is becoming something companies can monitor at a granular level — raising new questions about efficiency, cost management, and employee oversight.
“There’s going to be a lot of abuse,” Rauch said, referring to potential misuse of AI tools as access expands.
David Adubiina
