Enterprises Shift from Token Counts to Outcome-Based AI Metrics

Major technology companies are moving away from using token consumption as a primary metric for measuring artificial intelligence adoption and value. Reports indicate that Amazon and Meta employees engaged in "tokenmaxxing," artificially inflating usage scores by delegating unnecessary tasks to AI agents to meet internal leaderboard targets. This behavior highlights the flaws of volume-based metrics, which often reflect inefficiency rather than business value and create unpredictable costs for finance teams. In response, Salesforce has introduced the Agentic Work Unit (AWU), a pricing and measurement model based on discrete tasks completed by AI agents, such as resolving customer inquiries or executing workflows. This shift aims to provide greater cost predictability for enterprises and align AI usage with tangible outcomes. As agentic AI is forecasted to account for 30% of enterprise software revenue by 2035, the industry is prioritizing metrics that measure actual work done over raw data processing volume. This transition addresses the growing disconnect between technical activity and financial visibility, offering a more honest assessment of AI productivity while mitigating the perverse incentives created by traditional token-based tracking systems.