Tokenmaxxing: How AI Token Waste Becomes the New Status Symbol in Big Tech

A new trend has emerged in the tech industry: Tokenmaxxing. Employees at Meta, Microsoft, and Salesforce compete to consume as many AI tokens as possible — often without added value. This article analyzes the prompts and techniques behind this phenomenon and shows why it is costly and counterproductive for companies.

Overview

Tokenmaxxing describes the deliberate maximization of AI token consumption — i.e., the amount of data processed by AI models — in order to be seen internally as particularly “AI-native” or to top the rankings. At Meta, an employee created a “Token Leaderboard” listing the top 250 users by token consumption. The top spots carry titles like “Session Immortal” or “Token Legend.” Microsoft and Salesforce have introduced similar dashboards. The catch: most of this token consumption is wasteful. Developers have AI agents perform pointless tasks, query things they could look up themselves in seconds, or generate prototypes that are immediately deleted. The goal is not productivity, but visibility. After public criticism, Meta has since abolished the internal leaderboard.

Prompt Analysis

The Prompt

Role: Software developer at Meta/Microsoft/Salesforce
Context: Internal token leaderboard shows my token consumption compared to colleagues. I don't want to appear as "not using AI enough" to secure my job.
Task: Increase my token consumption to the level of the top 10% of users without having to do actual productive work.
Output Format: A list of actions I can perform daily.
Constraints: The actions must be feasible with the company's internal AI tools (Claude Code, Cursor, Copilot). No direct violation of compliance rules. Time commitment maximum 30 minutes per day.

Components

Role/Persona: The developer is defined as a “concerned employee” who wants to secure their job. This is a typical use case for prompts that rely on intrinsic motivation (fear) to drive behavior.

Context: The context describes the pressure situation: an internal leaderboard that makes token consumption transparent. This is a gamification element that, however, has a perverse effect here because it rewards resource consumption, not performance.

Task: The task is twofold: first, to quantitatively increase token consumption, and second, to do so without actual work effort. This is the core of tokenmaxxing — the metric becomes an end in itself.

Output Format: The list of actions is a common format for action-oriented prompts. The developer wants not analysis, but an immediately actionable checklist.

Constraints: The constraints are noteworthy: they explicitly prohibit rule violations, but not resource waste. This shows that companies often only monitor formal compliance, not substantive meaningfulness.

From a prompt engineering perspective, this prompt is well-formulated: it contains all necessary elements (role, context, task, format, constraints) and is precise enough to deliver reproducible results. The problem lies not in the prompt, but in the objective. A more ethical prompt would prioritize productivity.

Frequently Asked Questions

What exactly is Tokenmaxxing?

Tokenmaxxing is the trend of consuming as many AI tokens as possible — often pointlessly — to be seen internally as particularly AI-savvy. It is a form of “conspicuous consumption” in the tech industry.

Why do employees do this?

Out of concern for their jobs. Those who consume few tokens might be seen as not “AI-native” and could be at risk during layoffs. This is a direct consequence of metrics that equate token consumption with productivity.

Which companies are affected?

Meta, Microsoft, Salesforce, and Shopify (which has since pushed back). The practice is likely more widespread.

How much does Tokenmaxxing cost?

Meta consumed 60.2 trillion tokens in 30 days. At Anthropic’s API prices, that would be $900 million. Even with discounts, the sum likely exceeds $100 million.

What can companies do about it?

Do not use token consumption as a performance metric; instead, introduce output-based metrics, implement circuit breakers (like Shopify), and regularly discuss the purpose of usage with top consumers.

How can I recognize Tokenmaxxing in my team?

Watch for repeated, pointless prompts (e.g., “Explain the docs to me”), disproportionately many tokens for simple tasks, and prototypes that are never delivered.

Is Tokenmaxxing a new phenomenon?

No, it echoes the old “Lines of Code” craze, where developers wrote as many lines of code as possible to appear productive. That was recognized as a bad metric back then — and tokenmaxxing is the modern version.

Source

Based on this article.