Uber COO Andrew Macdonald said in a Rapid Response interview that the company's spending on AI tools is becoming "harder to justify internally." Two months earlier, CTO Praveen Neppalli Naga disclosed that the annual budget for Claude Code had been "burned through early." But the deeper issue: higher token consumption hasn't delivered proportional consumer feature output.
Budget Crisis: CTO Burned $1,200 in Tokens in Two Hours
Adoption of Claude Code among Uber's 5,000 engineers surged from 32% to 84% in months. Individual monthly spending ranges from $500 to $2,000. Naga himself consumed $1,200 in tokens during a single two-hour internal demo. CEO Dara Khosrowshahi stated on the latest earnings call that hiring is being slowed to offset AI investments—the bill for AI tools is now squeezing real headcount budgets.
Broken Causal Chain: More Tokens ≠ More Features
Macdonald said discussions with senior engineering leads revealed that higher token usage hasn't translated into a proportional increase in consumer-facing features. "That link doesn't exist yet," he said. "Maybe there's slightly more stuff being shipped, but drawing a line between those numbers and 'we shipped 25% more useful consumer features' is very hard." Token consumption measures usage, not value delivered. Salesforce has labeled such metrics "vanity metrics" and explicitly opposed using them to evaluate employee performance.
Macdonald also pointed out a blind spot: for engineers who don't pay out of pocket, AI tools "feel free," encouraging experimentation regardless of cost. But the company foots the bill. This cost misalignment is a structural driver of runaway token burn.
Industry Divide: Burn Hard vs. Measure Hard
Uber isn't alone. Google at I/O 2026 championed "tokenmaxxing," encouraging mass AI usage as a signal of engineering engagement. But a different path exists: Duolingo once included AI usage frequency in performance reviews, but employees questioned whether they should "use AI just for the sake of using AI." CEO Luis von Ahn admitted in a podcast that the policy was quietly withdrawn, saying "we ended up pushing something that in many cases doesn't apply." A healthcare company burned 1 trillion tokens in six months, generating $6 million in unplanned costs, with finance unable to trace the driver.
Macdonald didn't announce any cutbacks or plans to abandon AI tools. He simply articulated a problem many companies face but few senior leaders openly address. Measuring AI ROI remains a riddle with no standard answer—but the gap between "how much we used" and "what we got" is unmistakably wide.