.
A new culture of “tokenmaxxing” is sweeping through Silicon Valley, as tech workers compete to consume massive amounts of artificial intelligence data to prove their productivity.
The trend involves engineers using “agentic” AI tools to process billions of tokens — the basic units of data used by AI models — often running automated systems 24 hours a day. While some executives encourage the heavy usage, others warn that the practice has become a form of “productivity theater” that prioritizes raw data consumption over actual work quality.
But to grasp the sheer scale of what is happening inside these companies, you first have to look at the underlying currency of this whole trend — the token. A token is essentially the atomic unit of AI data processing. While ratio can vary, a good rule of thumb is that one token is roughly three quarters of a standard word.
Love reading and writing? Submit a post to our blog.
So a 75-word paragraph is about 100 tokens. And every single time you prompt an AI, every time it writes a line of code or even just processes logic behind the scenes, it consumes tokens, and those tokens carry a hard financial cost.
Internal Leaderboards and High Costs
At companies like Meta and OpenAI, employees have competed on internal leaderboards that track how many tokens each worker burns through. Meta’s “Claudeonomics” dashboard recently showed that 85,000 employees used a combined 60 trillion tokens over a 30-day period. The highest individual user at Meta averaged 281 billion tokens in a month, an amount equivalent to hundreds of millions of words.
Tech leaders have largely egged on this behavior. Meta Chief Technology Officer Andrew Bosworth reportedly told engineers there was “no limit” to token spending if it boosted output, while Nvidia CEO Jensen Huang said he would be “deeply alarmed” if highly paid engineers were not spending hundreds of thousands of dollars on AI tokens annually.
The financial costs are significant. At Anthropic, a single user of the “Claude Code” system reportedly racked up a bill exceeding $150,000 in one month.
The Rise of Agentic Tools
The explosion in token usage is driven by new “agentic” coding tools, such as OpenClaw and Codex, which can work unsupervised for hours to edit large code bases or write entire programs from a single prompt.
Unlike manual AI use, where a person might use 10,000 tokens to write an essay, these autonomous agents can spawn sub-agents to handle different tasks simultaneously. Ege Erdil, co-founder of the startup Mechanize, estimated his own usage at between 1 billion and 10 billion tokens a week, noting that “it doesn’t really take that much” when agents run continuously.
Pushback and New Metrics
Despite the enthusiasm, some major tech firms are pushing back against token consumption as a “vanity metric”. Meta recently shut down its internal token leaderboard after reports of its massive usage surfaced.
Salesforce has introduced a new measurement called Agentic Work Units (AWUs) to focus on business results rather than data burn. The metric is designed to track completed tasks, such as how quickly an AI agent resolves a customer service issue.
“I could tokenmax by running endless loops… but if customers didn’t actually get that much work out of it, then what’s the point?” said Madhav Thattai, an executive vice president at Salesforce. Other industry leaders, including the CEOs of HubSpot and Appian, have echoed this sentiment, arguing that “outcome maxxing” is more valuable than token consumption.
As companies like Meta and Shopify begin to factor AI usage into formal performance reviews, many engineers say they feel a “token anxiety” to keep up. Industry experts suggest that for many workers, using AI at an accelerated pace has become a necessary career strategy to avoid being seen as obsolete.
I’m a freelance writer with 6 years of experience in SEO blogging and article publishing. While you’re here, get the latest updates by subscribing to my newsletter.
