While many teams track token usage as a primary AI ROI metric, Anthropic views it differently. High token usage doesn't correlate with high value, as it's easy to waste tokens. Instead, a token count of zero is the most important signal, as it clearly indicates someone is not using the AI at all.

When companies measure AI adoption by counting tokens used, it creates a perverse incentive. Employees and their teams create agents to perform pointless tasks simply to boost their metrics, leading to fake productivity and problematic artifacts.

In the current 'capability exploration' phase, companies incentivize developers to use as many AI tokens as possible. This serves as a visible, albeit inefficient, signal of AI adoption to management, prioritizing quantity over quality.

A trend called "tokenmaxxing" is emerging in Silicon Valley, where companies like Meta use leaderboards to track employee AI token usage. This reflects a corporate bet that higher token consumption correlates with increased productivity, turning AI usage into a new, albeit gameable, performance metric for engineers.

Gamifying AI token consumption via internal leaderboards, as seen at Meta, creates perverse incentives. Employees may burn tokens to climb the ranks rather than to solve real business problems. This "tokenmaxxing" promotes conspicuous consumption of compute, a vanity metric that masks true productivity and ROI.

According to Mike Cannon-Brookes, advanced enterprises are not tracking AI success by counting tokens. Instead, they are asking harder questions about overall output, such as engineering productivity and quality. They understand that high token usage doesn't always correlate with high productivity, shifting focus from raw usage to tangible business outcomes.

The initial approach to AI adoption was often "token maxing"—using as many tokens as possible under the assumption that more usage equals more value. A more sophisticated and sustainable strategy is "output maxing," which focuses on achieving the desired result while actively minimizing token consumption and cost.

Simple leaderboards tracking token usage lead to 'token maxing'—engineers burning tokens to look productive. A better approach is to use hack days and demos to reward and showcase high-impact output, which implicitly encourages effective AI use.

Encouraging high AI token usage ('token maxing') becomes actively harmful when an employee lacks fundamental skills. They use expensive tools to produce poor work faster, amplifying their negative impact instead of driving positive outcomes. This is a significant hidden risk in broad AI adoption.