The Cult Of Tokenmaxxing

While the Meta employee has since taken down the leaderboard, developers are under pressure to increase token consumption across the industry. In March, The New York Times reported that an engineer at OpenAI processed 210 billion tokens in a week, enough text to fill Wikipedia 33 times over. The article also found that one user at Anthropic ran up a $150,000 Claude Code bill in a single month.

Even Jensen Huang, founder and CEO of Nvidia, said on the All-In Podcast that he’d be “deeply alarmed” if a $500,000 engineer didn’t consume at least $250,000 worth of tokens. The reality is that there is an expectation emerging in the tech industry that developers and employees should be making use of as many tokens as possible.

“The recent trend of ‘tokenmaxxing’ reflects a desire to incentivise AI usage within organizations and presents the assumption that tokens are the base unit for AI usage - i.e., greater consumption indicates higher value of AI,” Jim Rowan, principal, U.S. head of AI at Deloitte Consulting LLP, told me via written commentary.

“While it’s a well-intentioned perspective, there are risks of turning tokens into a ‘vanity metric.’ For many organizations, this metric of consumption does not distinguish between AI usage and the actual value derived from AI,” Rowan said.

Tokenmaxxing In Action At Cleo

Although critics point out the risks of needless token consumption, many companies are going all-in on tokenmaxxing, particularly AI-first companies. One such company is Cleo, a financial assistant app founded in 2016, which is currently valued at $1 billion.

In April I had a video call with Barney Hussey-Yeo, founder and CEO of Cleo, who expressed his belief that tokenmaxxing was an essential productivity metric. “Everyone at Cleo is tokenmaxxing,” Hussey-Yeo said. “Everyone at Cleo is building software internally."

Hussey-Yeo notes that Cleo has changed rapidly thanks to advancements in 4.5 Opus and the Claude Code CLI. Today, employees are allowed to spend up to $1,000 per month on tokens if they are non-engineers (this increases to $2,000 a month if they’re an engineer).

During the call he said that Cleo has an all hands meeting coming up in which he will be encouraging employees to tokenmaxx. He also claims to have spent £27,000 (over $36,000) on tokens this month by having multiple agents running in the background during and outside work hours.

“You’re either AI native or you’re irrelevant.” Hussey-Yeo said. “Anyone that is not using Claude Code and like genuinely using it to improve their productivity, the way they work, how they interact with everyone else, they’re just not gonna make it.

“You kind of see the productivity levels," Hussey-Yeo added. "We’ve got 178 engineers, and you can quite easily see now the productivity levels are really rising with like a big cohort and then there’s a cohort of laggards and they’re the ones still like handcrafting code, so the output and productivity of native to laggard is gonna really, really accelerate."

High Token Consumption Vs. Tokenmaxxing

Companies like Meta and Cleo can be categorized as tokenmaxxers, but there are also a number of companies that encourage high token consumption but don’t fall into the same extreme usage. Starburst, a data lakehouse platform built with Trino, currently valued at $3.35 billion, doesn’t put any limits on token use, but it also doesn’t stress token consumption for its own sake.

“We have allowed our teams to have a complete free hand over token usage, and we haven’t put any guardrails on token maximization,” Jitender Aswani, vice president of engineering at Starburst, told me in a video interview. “We have an internal terminology; we call it ‘let a thousand flowers bloom,’ and if I prematurely put token maximization, I would be curbing the creativity or limiting the creativity or the creative solutions that are going to come from the RND team.”

Aswani says that Starburst doesn’t tokenmaxx but notes token consumption is a“good sign” in that it shows that employees are beginning to experiment with AI. For this reason, he wouldn’t put any limits on token utilization (though he does have some guardrails in place).

He also told me that the company has seen a significant increase in product velocity since December, with the time taken to move an idea to production seeing a 60% compression across the R&D team. One third of the company’s code is now generated by Claude, which he notes wasn’t the case as of three months ago.

It’s worth noting that Aswani doesn’t look at token consumption as a metric in of itself. Instead, he prefers to look at hard metrics like DORA metrics, developer velocity, usage of AI agents across the company, code quality and whether incidents in production can be resolved faster.

Are Tokens A Vanity Metric?

Starburst isn’t alone in approaching token consumption with caution. Stefan Camilleri, vice president of engineering at Typeform, also raised doubts about tokens as a hard metric that teams should measure.

“There’s a lot of discussion right now about ‘token maxxing,’ but focusing on token volume alone misses the point. What we care about internally is the value created per token: whether AI is helping engineers ship more reliable software, accelerate delivery, and solve harder problems," Camilleri told me via email.

“Using a lot of tokens is not necessarily a good signal. What we do is map expenditure to value. Where that mapping is strong, we dig in with those engineers, understand what they’re doing differently, and scale it across the org,” Camilleri said. It’s not just whether engineers are using AI, it’s whether they’re using it intelligently.

In an economic climate with greater board scrutiny, token usage without demonstrable business impact runs the risk of coming across as conspicuous consumption. So while encouraging token usage can remove barriers to innovation for many teams, it must also be tied to measurable metrics.

Updated April 14th 2026 with information about Cleo’s valuation.