Uber exhausted its entire 2026 artificial intelligence budget in just four months, with monthly API costs per engineer ranging from $500 to $2,000. The overrun forced the company to implement spending caps and raised internal questions about whether AI tools are delivering promised productivity gains.

Uber rolled out Claude Code access to its engineering team in December 2025. Adoption accelerated rapidly through the first quarter of 2026 as developers discovered the tool’s multi-step coding capabilities and agentic reasoning features. Usage doubled by February as word of mouth spread internally.

The explosive adoption created an unanticipated cost problem. What seemed like a modest enterprise license became a massive operational expense when engineers across the company began using Claude Code for daily development tasks, code review, debugging, and architectural planning.

In response, Uber imposed monthly spending limits of $1,500 per AI coding tool per employee. The caps apply specifically to agentic coding software such as Cursor or Anthropic’s Claude Code. This ceiling-based approach prevents runaway bills but also frustrates engineers who discovered genuine productivity benefits.

More provocatively, Uber’s leadership acknowledged internal skepticism about the connection between rising token consumption and tangible business outcomes. In a May 2026 interview, executives stated: “It’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features.'”

This frank assessment is notable. Many tech companies have adopted AI tools enthusiastically while struggling to quantify productivity gains. Uber’s public acknowledgment of this gap adds credibility to broader industry doubts about AI-driven productivity.

The company did not abandon AI tools. Instead, it implemented governance: tracking usage, measuring code quality and deployment velocity, and correlating tool consumption with product outcomes. Uber’s path forward involves disciplined evaluation, spending controls, and realistic expectations about extracting value from powerful new technology.