Companies Learn to Track AI Use --- Bills for computing can balloon for those that don't spend tokens wisely

“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 (3 pages) 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 are 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 as high as his or her peers', they get curious about what is 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 chief executive, who gives his employees an unlimited token budget.

 

Rauch says that so far his highest token spenders are also his top performers. For now, he is comfortable letting them run. He estimates that $10,000 spend for a day's work probably saved him millions.” [1]

 

1. Companies Learn to Track AI Use --- Bills for computing can balloon for those that don't spend tokens wisely. Bindley, Katherine.  Wall Street Journal, Eastern edition; New York, N.Y.. 19 Mar 2026: B4.