Tokenmaxxing — maximizing AI token usage as a productivity proxy — is examined through the lens of Goodhart's Law and misaligned incentives. Meta's internal token leaderboard was gamed by engineers running scripts that burned millions of tokens with zero productivity, mirroring Google's 'promomaxxing' culture where engineers manufactured complexity to earn promotions. The input→output→outcome chain is analyzed: more tokens don't guarantee better outcomes and can amplify noise, bugs, and coordination overhead. Tokenmaxxing does excel in rapid exploration scenarios where the goal is knowledge generation. The core argument is that incentive structures, not individual employees, are responsible when behavior diverges from company health.
Table of contents
When a measure becomes a target, it ceases to be a good measurePromomaxxingThe Input → Output → Outcome DiscrepancyWhere Tokenmaxxing ExcelsMy TakeawaysSort: