A new word has entered the enterprise AI vocabulary: tokenmaxxing. And what it reveals about how companies are measuring AI adoption should concern every business leader making AI investment decisions.

According to reporting by the Financial Times, cited extensively by Fortune, Amazon employees have been running the company’s internal AI tool — called MeshClaw — on trivial and unnecessary tasks simply to push their token consumption numbers higher. Why? Because Amazon built an internal leaderboard tracking token usage, set targets for more than 80 percent of developers to use AI each week, and made managers aware of individual usage statistics.

The result was predictable to anyone who’s studied how metrics shape behaviour: the leaderboard became the goal, not what the leaderboard was meant to measure.

What Is Tokenmaxxing?

Token consumption is the technical measure of how much AI infrastructure is used — every word processed by an AI model burns tokens. Token counts make sense as a cost metric. They are a disastrously poor proxy for business value.

Tokenmaxxing is what happens when you treat a cost metric as a success metric. Employees, under pressure to demonstrate AI adoption, run AI tools on tasks that don’t need them. They generate long reports that don’t get read. They have AI summarise documents they already understand. They ask AI to draft emails they would have written themselves in two minutes.

Token counts go up. Productivity doesn’t.

Amazon isn’t alone. Meta reportedly launched a similar internal ranking system — employees called it “Claudeonomics” — that ranked the company’s roughly 85,000 workers by token consumption. The leaderboard lasted only days after becoming public before Meta removed it. Amazon subsequently restricted visibility of team-wide usage statistics.

The Stakes Are Enormous

These internal metric games matter beyond employee morale. The AI infrastructure investment decisions at the world’s largest tech companies are partly justified by internal usage data. Combined 2026 capital expenditure from Amazon, Microsoft, Alphabet, and Meta is already pushing $700 billion. Some Wall Street projections exceed $1 trillion for 2027.

If a meaningful share of that reported usage is employees gaming leaderboards rather than genuine productivity gains, the investment thesis has a problem. D.A. Davidson analyst Gil Luria flagged the concern directly, noting that tokenmaxxing casts doubt on the validity of the demand signals underpinning these massive infrastructure commitments.

Meanwhile, the cost side of the equation is becoming clearer. Agentic AI workflows can consume up to 1,000 times more tokens than standard AI interactions. Uber’s CTO reported in April that the company burned through its entire 2026 AI coding tools budget in just four months.

What This Means for Business Leaders

The tokenmaxxing problem is ultimately a measurement problem — and it’s one that businesses of every size can fall into without realising it.

If you are evaluating your AI investment by tracking how often your team uses AI tools, you are measuring activity, not outcomes. The question is not “did my team use AI today?” It is “did AI produce a measurable result that justified its cost?”

Practical measures worth tracking instead:

• Time saved on specific repeatable tasks (with before/after comparison)
• Error rates in AI-assisted versus manually completed work
• Customer response times where AI handles first contact
• Revenue or cost outcomes linked to specific AI deployments

The companies winning with AI right now are not the ones with the highest token consumption. They are the ones who identified a specific problem, deployed AI precisely against that problem, and measured the actual outcome.

Building a team that can make those distinctions — between genuine AI value and AI theatre — requires a baseline of AI and data literacy that most organisations have not yet invested in. That is exactly the gap that structured education and practical deployment guidance exist to close.

What This Means for Business

The tokenmaxxing story is an early warning. As AI costs scale with usage and agentic systems multiply token consumption, organisations that measure adoption by activity rather than outcomes will find themselves spending more and achieving less. The discipline of outcome-focused AI deployment is not a luxury for early adopters — it is becoming the baseline requirement for any serious AI investment.