Meta employees burned 60 trillion tokens in a month on a dare. That's over $100 million. And the creator of OpenClaw is spending $20,000 a day on OpenAI tokens. I dug into what tokenmaxxing actually is.
Someone inside Zuckerberg's company built a leaderboard called "Claudeonomics" — ranking 85,000 employees by token spend. The top 250 earned titles like Session Immortal and Token Legend. Gergely Orosz from Pragmatic Engineer talked to Meta engineers and found that people were running agents that burned tokens for nothing, and some SEV incidents turned out to be caused by carelessly generated code. If you convert 60 trillion tokens at Anthropic API prices, you get around $900M. Even with discounts — over $100M.
To put that in perspective, let's look at one person. Last week, OpenClaw creator and OpenAI engineer Peter Steinberger posted his monthly bill — $1.3 million, 603 billion tokens, roughly a hundred Codex agents managed by a three-person team, all paid for by OpenAI. Meaning Meta in that same month burned through about a hundred Steinbergers.
A day after The Information published the story, the leaderboard was taken down. A long-time Meta engineer suspects the whole thing was originally a way to collect traces for training the next model — expensive, but Meta can afford it.
The phenomenon in Silicon Valley has already been dubbed tokenmaxxing, and Meta isn't alone. At Microsoft, a similar dashboard has been running since January. An engineer Orosz spoke with admitted to deliberately asking AI questions whose answers are already in the docs, prototyping features he'll never build, and firing up an agent even when doing it by hand would be faster. Not for productivity. To avoid looking "not AI-native enough" compared to colleagues.
Salesforce took it even further. There's a Mac widget that shows your spend every 15 minutes. Minimum target: $100/month on Claude Code, $70 on Cursor. There's a web tool where you can see any colleague's spend. The maximum limit ($250/$170) is lifted with one click, and in some teams it's been removed entirely. Developers calibrate their spend to stay just above average and generate projects they'll never ship to production.
The only place this actually worked out was Shopify. They renamed the leaderboard to "usage dashboard" and added an automatic circuit breaker that cuts API access if one engineer's spend suddenly spikes — which usually means an agent went haywire 😀 Plus, the head of engineering personally reviews what the top spenders are actually burning tokens on. When there's a real conversation with a manager behind the number, the urge to torch tokens for a title disappears.
There's an old parallel to all of this. Poets used to be paid by the line, and there's a story that Mayakovsky wrote in his signature staircase style precisely for that reason — more lines, more money. He denied it and said it was about rhythm. But even if the story's true, at least Mayakovsky was producing something worth reading.
How many tokens are you burning? Per day/week/month? Next post I'll share my own AI spending.