Big Tech Gamified AI Adoption. Now Nobody Can Afford the Score.

Meta built a leaderboard. It ranked every one of its 85,000 employees by how many AI tokens they consumed. The top slot earned the title "Token Legend." The person who held it burned through 281 billion tokens in a single 30-day window. Zuckerberg did not crack the top 250.

They called the dashboard "Claudeonomics." In its first month, total usage across the company exceeded 60 trillion tokens. Meta took the leaderboard down after The Information reported on it and the numbers started circulating on social media.

By then the damage to the budget was done.

Gergely Orosz of the Pragmatic Engineer newsletter gave the phenomenon a name: tokenmaxxing. Great for AI vendors, he wrote. Bad for everyone else.

How It Works

The "-maxxing" suffix is borrowed from Gen Z. Looksmaxxing. Sleepmaxxing. The concept: optimize a single variable as aggressively as possible. In this case the variable is AI token consumption, and the optimization is being driven by employers who decided that usage volume equals AI readiness.

The logic seemed reasonable at the start. Companies wanted engineers to adopt AI coding tools. Engineers are competitive. Put up a leaderboard. Celebrate the top users. Watch adoption follow.

It did. What also followed was a spending pattern that no finance team had modeled.

Engineers started running multiple agents in parallel. Writing longer prompts. Structuring entire days so several AI workflows ran simultaneously. The goal stopped being "use AI to ship better code" and became "climb the board." One Meta engineer told the Pragmatic Engineer that some production incidents were caused by what looked like careless AI code generation, as if the developer was more focused on volume than quality.

Meta is not alone. Microsoft has had an internal token leaderboard since January. Amazon is running a similar system. Disney launched an AI Adoption Dashboard tracking token usage across Cursor and Claude, functioning as a leaderboard whether they call it one or not. Top power users at Disney make hundreds of requests daily. Tens of millions of tokens.

Uber Burned Its Entire Year in Four Months

No company illustrates the spending problem more concretely.

Uber rolled out Claude Code to its engineering team in late 2025. Adoption was encouraged. Internal leaderboards ranked developers by usage. By February, Claude Code usage had nearly doubled. By April, 84% of Uber developers were classified as agentic-coding users. And by April, Uber had burned through its entire 2026 AI budget. The full year. Gone in four months.

CTO Praveen Neppalli Naga told The Information he was "back to the drawing board, because the budget I thought I would need is blown away already."

Monthly API costs per engineer ranged from $500 to $2,000. Across the company, 95% of engineers used AI tools monthly. 70% of committed code was AI-generated. 1,800 code changes shipped per week with no direct human input. The tool proved too successful to afford at scale.

ServiceNow hit the same wall. CIO Kellie Romack described the rapid cost increases as a difficult management challenge. Second major public company to admit it blew through its full-year Anthropic budget in the first few months of 2026.

What the Leaderboard Was Really Doing

After Meta took down Claudeonomics, one long-tenured engineer offered a theory about why the leaderboard existed in the first place. The engineer told the Pragmatic Engineer they suspected driving AI usage was the actual goal, even if nobody said it out loud. More usage produces more real-world traces. More traces produce better training data for Meta's next-generation coding models.

The leaderboard was not measuring productivity. It was farming training data from the company's own workforce. Gamification was the harvesting mechanism.

Meta CTO Andrew Bosworth has not confirmed that theory. But he has publicly endorsed the spending. He told Forbes that his best engineer was spending the equivalent of his salary in tokens but was "5x to 10x more productive" as a result. "It's like, this is easy money," Bosworth said. "Keep doing it."

Maybe that is true for the top performer. The question is whether it is true for the other 84,999 people on the leaderboard who were all chasing the same metric while nobody measured whether the output justified the spend.

The Governance Vacuum

Tokenmaxxing inverts the normal relationship between cost and accountability. Every other budget in a company is tied to outcomes. Marketing spend gets measured against pipeline. Engineering headcount gets measured against shipping velocity. Cloud spend gets measured against uptime and throughput.

Token spend gets measured against nothing. There is no standard enterprise metric connecting AI token consumption to business outcomes. Companies are tracking inputs, how much AI is being used, without tracking outputs, what the AI usage actually produced. The leaderboard rewards consumption. Nobody is measuring whether the consumption created value.

The WRITER/Workplace Intelligence survey found that only 29% of executives report meaningful ROI from generative AI. But 97% say AI has been deployed. The gap between deployment and return is exactly where tokenmaxxing lives. Massive adoption. Massive spend. Minimal measurement.

An analyst at Wedbush told Fortune that Amazon's approach to gamifying token usage "doesn't sound very healthy."

That is one way to put it. Another: the enterprise world just recreated the engagement metrics trap from social media, except instead of optimizing for time-on-site they are optimizing for compute burn.

Where the Money Went

Here is what finance teams are now reconstructing.

An engineer running a single agent against a moderate codebase is a $500-a-month problem. The same engineer running multiple parallel agents, long-context refactors, and test-generation pipelines is a $2,000-a-month problem. At scale, a 5,000-person engineering org running at Uber's adoption rates can generate token bills that rival the team's total salary cost.

KPMG's Q1 2026 AI Quarterly Pulse found that U.S. organizations project average AI spending of $207 million over the next 12 months. Nearly double the figure from the same period last year. Goldman Sachs surveys show large companies already overrunning their AI budgets by orders of magnitude. And as Swami Chandrasekaran, head of AI and data labs at KPMG North America, told Marketplace: "Even, like, a quarter, two quarters ago, nobody bothered about LLM consumption costs."

They bother now.

The Correction Already Started

The tokenmaxxing era may already be peaking. Meta took its leaderboard down. Anthropic is splitting its subscription into interactive and automated usage pools effective June 15. GitHub is moving all Copilot plans to usage-based billing on June 1. The flat-fee model that made uncapped token consumption possible is being dismantled in real time.

But the cultural problem will outlast the pricing correction. Companies spent six months telling employees that AI usage is a performance metric. Engineers internalized it. They optimized for it. They built habits, workflows, and professional identities around being high-consumption AI users. Telling them to slow down now, after celebrating them for speeding up, creates a credibility problem that no pricing change fixes.

What Enterprise Leaders Should Do Instead

Stop measuring AI adoption by consumption. Start measuring it by outcomes.

Define what AI success looks like before rolling out the dashboard. Not after the budget is gone. Build cost governance into AI deployment from day one. Set per-team and per-user spend caps. Require that AI tool usage be tied to specific projects with measurable deliverables. Treat token budgets the way you treat cloud budgets: with alerts, approvals, and accountability.

And stop gamifying adoption. A leaderboard tells your workforce that usage is the goal. If usage is not the goal, do not build a system that rewards it.

The companies that survive the tokenmaxxing hangover will be the ones that figured out the difference between using AI and using AI well. Everyone else just bought the most expensive scoreboard in corporate history.