Doug Levin

Tokenmaxxing: When Compute Activity Masquerades as Productivity

The article warns that many firms are treating raw token counts as a proxy for AI productivity, a practice the author dubs “tokenmaxxing.” By tracking every token generated by large language models, companies create a vanity metric that rewards higher compute usage rather than tangible outcomes. This trend is especially prevalent among LLM vendors who incentivize token‑heavy workloads. The piece argues that such metrics mask inefficiency and inflate costs without delivering real business value.