Enterprises grapple with AI costs and flawed token‑based productivity metrics

Large technology users such as Uber, Starbucks and other firms have begun curbing AI spending after token‑based pricing exposed runaway expenses. Uber's engineering teams exhausted their 2026 AI budget within four months, prompting the CTO to admit the tools were “too successful to afford at scale.” Starbucks tied a quarter of tech staff bonuses to the frequency of AI‑assistant use, a policy that encouraged higher token consumption rather than better outcomes. Similar pressures have emerged at companies like Accenture, where senior staff are urged to adopt AI tools or risk missing promotion opportunities, while management now seeks to limit usage.

Analysts argue that measuring employee productivity by AI token usage or time spent in AI tools misses the core value of the work. They note that more tokens do not guarantee higher quality; a developer could generate large volumes of low‑quality code that consumes many tokens, while sparing use of AI might solve critical problems more efficiently. The trend highlights a broader challenge for enterprises: developing meaningful metrics that capture AI‑assisted productivity without incentivising wasteful consumption.