By Katherine Bindley

Artificial-intelligence automation platform maker Zapier has a new kind of dashboard to keep track of its workers' AI use.

The new hot metric? How many tokens they are burning.

AI output, which has turbocharged productivity and promises to change the nature of work, might seem like it materializes out of thin air. But it's really the work of data centers churning through prompts and interpreting them in an elaborate and expensive, if unseen, process.

Every time a person prompts a bot or has an agent write code, the computing resources are measured in tokens. For text-based AI, it's fairly simple: Generating 750 words takes about 1,000 tokens. It gets more complicated when you're writing code, creating video and audio or enlisting agents to perform elaborate, dayslong tasks, but the idea remains the same: The more work you do, the more tokens are used.

"We have this new kind of line item," says Brandon Sammut, Zapier's chief AI transformation officer. The assistance AI provides -- whether it's handling a support ticket or closing a deal -- has a cost and companies need to bake that into their thinking.

While token pricing has gone down, token costs can be higher for some newer, more sought-after models -- and companies' use is generally going up. Companies sometimes opt for pay-as-you-go plans, while others might buy enterprise plans that include a certain amount of use per worker.

Most businesses are still just trying to get their employees to even use AI. But those that are further along are already tracking token use and starting to tally the costs. They're scouting whose AI strategies should be amplified after generating a great return, and what wastefulness should be squashed.

If Zapier leaders see someone's token use is five times higher than his or her peers, they get curious about what's going on: The person could be massively inefficient or a superstar, depending on what became of those tokens.

"We start to draw conclusions," says Sammut, "whether that's a golden pattern we want to multiply across their peers or whether it's an anti-pattern that we want to coach our way out of."

Brian Jabarian, a researcher with the University of Chicago's Booth School of Business who studies how new technologies reshape workplaces, says companies need to start measuring token use.

"Everyone thought that you just use AI tokens, and you have an increase of productivity, and we call it a day," he says. "But the reality is more complicated."

Suppose a company saves money up front by using AI as a recruiter. If the AI doesn't do a good job, the company either has to pay a human to sort it out, or spend more AI tokens solving the problem. When a company gives AI tools to 500,000 employees, he adds, "these token problems become first-order."

Some workers who know they should be using AI might think blasting through tokens will earn them a badge of honor. It depends what they have to show for it.

A senior engineer at the AI cloud platform startup Vercel deployed a team of AI agents to analyze a research paper and build a new critical-infrastructure service based on it. The bots generated a valuable code base in a day -- something that would have taken humans weeks, if not months.

The bill for that work? Around $10,000.

"It's a little bit like giving people a fire hose of fuel," says Guillermo Rauch, Vercel's CEO, who gives his employees an unlimited token budget.

At Kumo AI, a startup with around 60 employees, the company started monitoring token use on a per-engineer basis at the beginning of the year. Hema Raghavan, the company's co-founder, says her good engineers use AI agents as though they have an army of junior helpers. Some engineers like to ski on the weekends but their agents continue performing tasks.

Raghavan says token spends are part of a broader research-and-development cost structure that can't be evaluated in isolation. Sometimes high costs in one part of the company can mean savings elsewhere.

"We found examples where the agents actually helped us write more optimal codes, so it's actually taken down our cloud costs," she says.

Mark Hull founded Exceeds AI, which makes software that measures the impacts of AI across engineering organizations. Most of his customers are currently focused on getting all their employees to use AI. The furthest along measure their overall return on AI investments, but not at a granular level.

While Hull doesn't think individual employees will be watching a token meter, they will need to be thoughtful about the most efficient and effective ways to use AI. He predicts companies will develop governance around token use, such as placing limits on which models can be used for given tasks. AI can even help here, automating such choices.

Hull recently used Anthropic's Claude Code to develop three different workflow tools, totaling around 300,000 lines of code. The cost in tokens? Around $2,000. He decided everyone in his 15-person company should use the coding platform going forward.

"Within 48 hours our costs spiked," he says, adding that he has since put limits in place.

Rauch, the Vercel CEO, says that so far his highest token spenders are also his top performers. For now, he's comfortable letting them run. He estimates that $10,000 spend for a day's work probably saved him millions.

Still, it won't be unheard of in the near future for firms large and small to consider pulling an employee aside when they see surges in token use.

"They could be using them on side projects. They could be using it on their own, startups, side hustles, whatever," he says. "There's going to be a lot of abuse."

Write to Katherine Bindley at katie.bindley@wsj.com

(END) Dow Jones Newswires

March 17, 2026 22:00 ET (02:00 GMT)

Copyright (c) 2026 Dow Jones & Company, Inc.