‘Tokenmaxxing’ Is the New Silicon Valley AI Debate | Invesloan.com

Forget lines of code written, engineers have a new way to compete amongst each other.

Welcome to the era of “tokenmaxxing.”

Armed with shiny new AI coding tools, software developers across the tech industry have a wallet full of tokens to spend. Tokens, a measure of computing that determines how AI work is priced, have been floated as a form of compensation for engineers, even cropping up in job descriptions for AI fellowships at OpenAI and Anthropic.

But is token spending a good measure of developer productivity?

The question has lit up on social media this week as techies debate the concept of tokenmaxxing after The Information reported that some Meta engineers are racing to spend tokens to rank on an employee-made “Claudeonomics” dashboard that tracks usage and lets employees compete for titles like “Token Legend.” The company didn’t respond to a request for comment from Business Insider.

Some say it’s a helpful marker of employees embracing new tools; others say it could incentivize inefficient use of AI within companies — leading to performative gaming of the metric.

“Ranking engineers by token spend is like me ranking my marketing team by who spent the most money,” Linear COO Cristina Cordova wrote on X. “Don’t mistake a high burn rate for a high success rate.”

What is tokenmaxxing?

To understand tokenmaxxing, you first have to know what a token is. Large language models break words into numerical inputs, treating each token as roughly ¾ of a word. AI models charge based on the number of tokens used.

Tokenmaxxing, then, is the drive to spend as many tokens as possible. Meta and OpenAI are just some of the tech companies with token leaderboards, The New York Times previously reported.

While it’s difficult to measure how widespread tokenmaxxing has become, companies’ AI spending is clearly on the rise. The fintech company Ramp called it a “$1 trillion blind spot” on X, citing Gartner data showing that monthly AI spending among businesses has quadrupled over the last year.

It’s also a flex. Founders and future-forward engineers post their token spending on X to signal how all-in they are on AI. One xAI employee wrote that tech was turning every good idea “into theater.”

Y Combinator CEO Garry Tan appears to be a fan. Quoting a prior post that chided companies that are “stingy” with tokens, Tan wrote: “We’ve been tokenmaxxing longer than most people.”

Is tokenmaxxing a good incentive?

Some within the tech world argue tokenmaxxing is an effective metric; others call it reckless spending.

Khosla Ventures partner Jon Chu called token spending measurement an “absolutely stupid policy” on X.

“Plenty of my Meta friends told me folks have been building bots that just run in a loop burning tokens as fast as they can due to this policy,” he wrote.

Cursor employee Edwin Wee Arbus offered a nuanced take on the metric, calling it a “useful, fast proxy, but slightly flawed.” He compared it to body mass index, or BMI, which can provide health insights but does not capture muscle or bone mass.

While Nvidia CEO Jensen Huang hasn’t directly weighed in on “tokenmaxxing,” he has stressed the importance of engineers using a lot of tokens, saying that if a $500,000 engineer didn’t consume at least $250,000 worth of tokens, he would be “deeply alarmed.”

“The Pragmatic Engineer” newsletter author Gergely Orosz called the practice wasteful. “Devs game everything and anything seen as a target for more bonus or promos,” he wrote. “This was no different.”

BEP Research founder Ben Pouladian pulled a different takeaway from the trend, calling compute the bottleneck for innovation. “In the AI era, every employee becomes a compute consumer,” he wrote on X.

“Token spend is always an output not an input,” wrote Persona software engineer Arush Shankar, who previously worked at Square and Microsoft, according to his LinkedIn. “Worth looking at, but never in isolation. It’s a signal but not THE signal.”

Does your company track token spending or token use? Contact the reporter from a non-work email and device at [email protected], or on Signal at henrychand.30