In April 2026, it emerged that a Meta employee had built an internal leaderboard, “Claudeonomics,” ranking the group’s 85,000 employees by the number of tokens they were burning through — with unlockable titles like “Token Legend” or “Cache Wizard.” The frontrunner alone churned through, in a single month, enough to bill more than $1.4 million; neither Mark Zuckerberg nor his CTO appeared in the top 250. Amazon and OpenAI had their own versions.

No, this isn’t a parody. It’s “tokenmaxxing,” the latest managerial fad: ranking employees by their token consumption, the way you’d pin up a sales leaderboard. The employee who “consumes” the most is, obviously, the most modern, the most committed, the most productive. You can already see the problem.

It is, without exaggeration, one of the most absurd management ideas I’ve ever come across — far ahead of the habit of judging a developer by the number of commits they pile up (or by how many times they’ve broken the CI), or of tracking mouse movements to make sure the employee isn’t off at the pool.

We’re confusing the fuel with the distance traveled#

A token is a cost, not a result. It’s gasoline, not mileage. Ranking someone by their token consumption is like congratulating the driver who empties their tank the fastest, without ever asking where they ended up — or even whether they started at all. While we’re at it, let’s also reward employees by the number of coffees downed, emails sent, or meetings that could have been an email: all measures of busyness, never of value.

The problem is structural, not anecdotal. A token measures an input, not an output. And productivity, by definition, plays out on the output side — what gets produced, solved, shipped, improved. Measuring the input and calling it productivity is confusing the electricity bill with the light.

Goodhart’s law is waiting around the corner#

“When a measure becomes a target, it ceases to be a good measure.” Put up a token leaderboard and you’ll get exactly what you measure: tokens. The cleverest will inflate their requests to climb the honor roll — three prompts where one would have done. The most cautious, meanwhile, will close the tool for fear of looking like spendthrifts, depriving themselves of a real gain.

Either way, the ranking no longer tells you anything about reality. It now measures only one thing: people’s talent for gaming the ranking. This isn’t a theoretical worry: at Amazon, employees started multiplying pointless requests to inflate their scores, to the point where the company eventually pulled the plug on its own and switched to a metric focused on the real usefulness of the work produced. Congratulations, you’ve invented a contest whose only required skill is winning the contest.

What if we at least measured the right ratio?#

Let’s grant that we absolutely want a number, because a dashboard without a number is unsettling. Then let’s look not at raw consumption, but at the ratio between what comes out and what goes in: who is the most productive per token consumed? There, at least, we’d reward efficiency — the well-thought-out request rather than the ten thousand tokens wasted going in circles.

Except that this ratio doesn’t fit in a gauge either. Comparing the “yield” of a developer, a lawyer, and a salesperson simply makes no sense: their tasks, their uses, their deliverables have nothing in common. And above all, how do you put a number on the noble numerator of the equation — judgment, creativity, the client relationship, the helping hand given to a panicked colleague at 6 p.m. on a Friday? Everything that truly makes an employee valuable conveniently slips past the counter.

The counter is not the work#

Measuring token consumption is mistaking the thermometer for the fever. At best, it’s useless. At worst, you install a form of surveillance that erodes trust and pushes everyone to optimize the wrong number — while the real work waits. The best proof? The backlash is already here: Meta’s leaderboard disappeared after the media uproar, Walmart capped internal AI usage, and executives — from Uber to several investors — now openly admit that these tokens spent translate into no gain at the company level. By mid-2026, “tokenmaxxing” no longer refers to a strategy to follow, but to the drift everyone is trying to curb. When a metric collapses as fast as it was erected, it wasn’t measuring much.

The question was never “who consumes the most?”, nor even “who consumes the best?”. It remains, exactly as it was before AI: who solves the right problems, and solves them well? No token counter will ever answer that one. And the day a manager thinks otherwise, maybe it isn’t the employee who ought to be ranked.

Sources#