How Metric-Obsessed Corporate Culture Created the Token Maxxing Trend

Token maxxing (sometimes written as tokenmaxxing) is a workplace trend where employees deliberately maximize the use of AI tools, which they measure by burning as many AI processing units (tokens) as possible.

In the world of Large Language Models (LLMs) like ChatGPT, one token is roughly equivalent to four characters, or three-fourths of an English word. 1

Until recently, using millions of tokens required hours of typing manual prompts. However, the advent of “agentic” AI tools (like Anthropic’s Claude Code) has changed things. These systems can work unsupervised for hours, with a single agent spawning dozens of subagents to handle different parts of a task and generating thousands of tokens at each step.

Because tokens are quantifiable, they have become a convenient metric for management to measure how “AI-pilled” their workforce has become. Companies like Meta and Shopify have even started factoring AI usage into performance reviews. 2 Generous “token budgets” have effectively become a new tech perk.

Supporters say heavy token use is more than a signal of productivity. It incentivizes people to work harder, says Abhi Jothilingam, growth marketing lead at the customer support startup Sendbird, where employees are ranked from “Beginner” to “AI God.”

“People are motivated by competition,” Jothilingam told The Wall Street Journal.

Some Silicon Valley leaders say it’s a matter of survival

While critics see token maxxing as a waste of money, proponents argue it’s a calculated investment in AI literacy. For many leaders, the intense competition in the tech space makes high AI adoption a necessity.

The belief is that companies with more AI-savvy staff will outperform those with a human-first approach, and that you cannot build an AI-native workforce by being frugal.

“It is existential for us,” said May Habib, CEO of the generative AI platform Writer. “We are in the most competitive space that has ever existed and will ever exist.” 3

Habib argues that while the metric is gameable and some employees waste tokens on personal projects, high AI usage is the only path to innovation. She views the costs—which, in her company, reached $50,000+ for 10 billion tokens consumed by a single employee—as necessary to force a psychological mindset shift.

Sendbird’s Jothilingam said that the leaderboard has led to him testing out more ideas for AI apps and building them faster.

Ultimately, for companies that believe that their success hinges on AI adoption, the idea is that encouraging employees to experiment with it through token maxxing may be the most effective way to teach the limits of that technology.

With many companies now tracking AI adoption through these metrics, workers have found a few inventive ways to inflate their usage, with some even using employer-provided tokens to work on their personal projects and side hustles.

At companies like Sendbird and the AI financial assistant Cleo, engineers deploy multiple agents to continue working while the human is asleep or away. Cleo CEO Barney Hussey-Yeo noted he personally racked up over $36,000 in monthly token costs by keeping multiple agents running in the background at all hours. 4

Some employees have also reportedly built bots designed to run in infinite loops and query the AI for tasks that serve no business purpose. Jon Chu of Khosla Ventures said on X that Meta engineers were building bots that “just run in a loop burning tokens as fast as they can.”

Others opt for padding their requests with unnecessary context, as AI models charge more when they are fed larger amounts of data.

Tokenmaxxers take advantage of this by uploading entire codebases, lengthy PDFs, or redundant datasets to the AI’s context window (where you upload text, files, or instructions) for simple questions that only require a sentence of context.

How do employees feel about token maxxing?

The implementation of AI leaderboards has led to some tech workers viewing the practice as a competitive sport. For others, it’s about survival.

In that sense, the trend is being fueled by what San Francisco venture capitalist Nikunj Kothari describes as “token anxiety.” In a world where white-collar workers fear obsolescence, heavy AI use can seem like a strategic move to prove to bosses that you are keeping up with the times.

“Dinner conversations used to start with ‘What are you building?’” Kothari wrote. “That’s over. Now it’s ‘How many agents do you have running?’ People drop the number the way they used to drop their follower count.”

A potential backlash

On the other hand, not all employees are fans of AI in the workplace at all. Some are even going as far as to get religious exemptions from using AI at work.

“It doesn’t seem sustainable,” one OpenAI employee anonymously told The New York Times in their aforementioned report.

Employees on the anonymous professional platform Blind point out that token burning focuses on the wrong metrics.

“It’s crazy, it’s a graded metric for us, so not only do I find myself using AI for everything but I also intentionally use more expensive models for the simplest tasks,” said one user.

Another user wrote, “Motivating employees to innovate by counting tokens they spend is the same as motivating students to study by counting books they took from the library.”

Despite these sentiments and the costs associated with such heavy token use, with the AI boom being so widespread, many companies (especially in the tech space) still see it as a greater risk not to push their employees to use AI.

The token maxxing trend has divided Silicon Valley.

On one side, leaders like Nvidia CEO Jensen Huang view high token consumption as a proxy for modern tech fluency. Huang said on the All-In podcast that he would be “deeply alarmed” if a $500,000 engineer did not consume at least $250,000 worth of tokens per year.

Notably, Nvidia stands to gain from the current AI boom, as the company’s hardware and software form the foundational backbone used to run modern AI models.

But others in the space, like Habib and Jothilingam, clearly agree that token maxxing is a no-brainer in the current tech sphere.

On the other side, critics argue that this is a classic vanity metric.

“Ranking engineers by token spend is like me ranking my marketing team by who spent the most money,” said Christina Cordova, chief operating officer at a software company called Linear. “We may not have hit our KPIs, but Joe spent $200k on a branded blimp that only flies over his own house, so he’s getting promoted to VP!”

The AI adoption failures of Amazon and Uber

Token-based tracking has resulted in massively inflated budgets. A recent Gartner study forecast that AI agent software spending will reach $207 billion in 2026, up from $86.4 billion in 2025.

Uber is currently feeling the sting of that boom. According to Fortune, Uber burned through its entire 2026 AI coding tools budget in just 4 months after launching an internal leaderboard to incentivize adoption.

Uber’s COO Andrew Macdonald recently said that while autonomous agents are writing roughly 10% of the company’s code, the actual consumer value isn’t matching the server bills. “That link is not there yet,” Macdonald said in an interview. “So, if you’re not actually able to draw a direct line to how [many] useful features and functionality you’re shipping to your users, that trade becomes harder to justify because [AI] is not free.”

Amazon experienced a similar issue. The company launched an internal leaderboard to encourage staff to experiment with AI, but as reported by The Financial Times, employees managed to game the system by deploying automated agents to perform AI busywork with limited business value.

Amazon was forced to shut the leaderboard down after costs rose without any corresponding benefits to their true priorities.

According to one article by the Society for Human Resource Management (SHRM), that conclusion was almost inevitable.

“Employees tend to perform to the metrics by which they are evaluated,” journalist Roy Maurer wrote for SHRM. “The pattern mirrors other workplace measurement failures, where activity metrics eventually become disconnected from organizational goals.”

For critics, token maxxing is a textbook example of Goodhart’s Law: “When a measure becomes a target, it ceases to be a good measure.”

Experts suggest that leadership must distinguish between adoption (burning tokens) and performance (business outcomes). As written in one Forbes article, token maxxing “is similar to the environment in the 1990s or early 2000s, when software engineers were judged by how many lines of code they wrote.” 5

The most obvious solution is to measure outcomes instead of the process used to achieve them. For example, rather than simply congratulating the developer with the highest AI consumption, Sendbird’s CEO John Kim told Forbes that leaders should look at how much of that AI-generated code makes it into production.

Only time will tell if today’s tokenmaxxers are actually building the skills to become better employees. If it turns out to simply be productivity theater, companies are paying millions of dollars to fund it for no real benefit.