Foreign media Fortune reported that the once-popular internal corporate practice of "tokenmaxxing" is cooling down. Tokenmaxxing refers to using the number of tokens consumed by employees or teams when invoking AI models as an approximate metric for innovation and productivity. However, as costs rise and ineffective requests increase, more companies are beginning to tighten this practice.
The article notes that companies such as Meta, Amazon, and OpenAI previously implemented formal or informal token ranking systems to encourage engineers to compete based on model usage volume. The issue is that once a metric becomes a target in itself, it often strays from its original purpose. The Financial Times previously reported that some Amazon employees had their AI agents perform tasks with no real practical value, solely to maintain usage metrics.
Cost pressures are beginning to emerge.
As generative AI is widely adopted within enterprises, model usage costs have risen rapidly. The article states that some companies have begun restricting employees' use of third-party AI agents, particularly tools that rely on high-end models. Meta has removed the employee-created token leaderboard; The Verge reports that Microsoft has canceled Claude Code subscriptions for employees in several key product teams.
Uber also disclosed that the company exhausted its entire annual token budget within the first four months of 2026, with part of the spending attributed to frequent use of Claude Code. Salesforce CEO Marc Benioff stated that the company paid approximately $300 million to Anthropic this year and expressed a desire for a more intelligent routing system in the future to allocate different requests to models with more suitable costs.
Businesses prioritize business outcomes.
The article argues that the primary reason companies are tightening their token metrics is not merely cost-cutting, but rather the gap between investment and returns. Uber’s Chief Operating Officer, Andrew Macdonald, recently stated that the company struggles to directly link improvements in employee productivity to the delivery of new user-facing features or overall business outcomes. Without clear business results, it becomes increasingly difficult to justify the ongoing costs of the model.
This is why simply tracking token consumption is increasingly seen as an ineffective management tool. It can reflect the scale of usage, but it cannot indicate whether those calls genuinely improved the product, processes, or revenue.
True returns come from process reengineering.
The article cites Azeem Azhar of Exponential View, who argues that the current misalignment between AI investment and productivity resembles the “productivity J-curve” commonly seen in the early stages of a new general-purpose technology. During the experimentation phase, companies often increase costs without immediate returns; efficiency gains only emerge sharply after business processes are redesigned.
Using the example of a factory undergoing electrical modernization, the article explains that companies initially only replace lighting or power sources, but significant productivity gains occur only after the factory layout and individual equipment are restructured around the new technology. Similarly, in AI, many companies are still at the stage of limited pilot projects or tool stacking, and have not yet moved into deeper process transformation.
Comments suggest that the decline in token usage competitions stems from the fact that they measure "how much was used" rather than "what was created." For businesses, the value of AI ultimately lies in product delivery, business models, and revenue performance—not in rankings of model calls.