Tokenmaxxing: The New Empty Metric of AI Work

Temps de lecture : 5 min

The Rise of the Token Leaderboard

Let us be honest: when a company builds an internal leaderboard ranking employees by AI token consumption, something has gone sideways. Meta reportedly did exactly that—a system called “Claudeonomics” that awarded titles like “Token Legend” and “Cache Wizard” to its top 250 users. In one month, those employees burned through more than 60 trillion tokens. A Disney employee allegedly interacted with an AI assistant 460,000 times in nine days.

The internet quickly named this behavior tokenmaxxing. People learned to optimize their numbers, generating token usage for tasks that didn’t really need AI. Agents and digital delegates were set loose on processes that were, at best, marginally useful. The result? Meta’s leaderboard was taken down after it went public, and Amazon reportedly restrained its own internal tracking of AI activity.

The Problem with Metrics That Measure Activity, Not Value

Most people get this wrong: they assume that if you can measure something—token counts, API calls, hours of AI use—you’re tracking productivity. But what you’re actually tracking is activity. Activity is easy to game. Value is not.

I have very little patience for the kind of thinking that equates AI usage with employee performance. It’s the same flawed logic that once made lines of code a metric for developer output, or pages read a proxy for editorial quality. You end up with incentives that reward noise, not signal.

• Volume inflation: Simple tasks are padded with AI steps to boost counts.
• Quality neglect: Ironically, the more tokens you use, the more likely you’re using AI as a substitute for judgment.
• Cultural damage: When leaders reward token counts, they signal that appearance matters more than results.

Why the AI Usage Trend Is Actually a Trust Problem

If you strip away the noise, tokenmaxxing is not really about AI. It’s about trust—or the lack thereof. When managers don’t know how to evaluate judgment, they fall back on numbers. Numbers are comforting because they feel objective. But a machine-generated report is not the same as meaningful work.

One JPMorgan employee told the Financial Times that their AI team didn’t have a “good sense” of productivity. Another at KPMG described “AI churn,” where workers use AI to make ordinary tasks appear more complex. These stories are not isolated—they reflect a broader pattern across industries.

The Real Question for Decision-Makers

The real question is not, “How can we increase AI adoption?” The real question is, “How can we use AI to improve outcomes?” Token counts will not answer that. Neither will dashboard metrics or leaderboard titles.

From my view, businesses need to replace activity tracking with outcome measurement. Here is what that means in practice:

• Define the outcome first: What problem are you trying to solve? AI might help, or it might be a distraction.
• Use editorial judgment: Ask whether a task truly benefits from AI, or whether it’s just adding complexity.
• Evaluate the work, not the tool: Humans should be judged on the quality of their output, not the number of tokens they consumed.

A Sharper Standard for Measuring Work

As we move forward in 2026, organizations face a choice. They can double down on vanity metrics, creating cultures where employees game the system to appear productive. Or they can adopt a sharper standard, one that values outcome over activity, judgment over automation.

This is not complicated, but it is demanding. It requires leaders to engage with the nuances of their work, to set clear objectives, and to trust their people to make sound decisions with the tools at hand. When you do that, you’ll find that token counts become irrelevant—and that’s exactly when the real work begins.