Scott Wu, co-founder and CEO of AI coding startup Cognition, says companies have taken a wrong turn by using AI token consumption as a performance metric for programmers. On the Founders podcast, he argued that actual completed work and output should be the real measure.
The practice, dubbed "tokenmaxxing" in Silicon Valley, involves employees habitually spamming tools like Claude, Codex, or Cursor to pad their numbers on internal AI usage dashboards. Jacob Lauritzen, CTO of legal AI startup Legora, echoed the criticism on a podcast this month, calling it a "stupid" way to evaluate performance, as it encourages gamers to chase vanity metrics.
Andrew Feldman, CEO of chip developer Cerebras Systems, told the Bloomberg conference this month that providing unlimited token budgets was "very stupid" to begin with. He suggested companies prioritize cost efficiency, opting for low-cost open-source models rather than wasting premium closed-source models on everyday tasks.