Uber deployed Claude Code to engineers in December 2025. By April 2026, the company had consumed its entire annual AI budget - not because the tool failed, but because adoption took off faster than anyone planned.The numbers: 95% of Uber engineers now use AI tools monthly. 70% of committed code originates from AI. Monthly costs per engineer are running $500 to $2,000, depending on usage. The company's CTO said they're "back to the drawing board" on AI budgeting for next year.What's notable is what this implies for the industry. Most enterprises are still treating AI coding tools as a line item they can forecast like a SaaS seat license - fixed cost, predictable renewal. Uber's experience suggests the actual cost driver is adoption intensity, not seat count. A team that uses Claude Code heavily for multi-step agentic work generates orders of magnitude more API spend than one that uses Copilot for autocomplete.The companies that haven't hit this wall yet probably will. Uber's R&D spend is $3.4B annually, so even at the high end this is manageable for them. For a smaller engineering org, an unforecast 4x budget overrun on AI tooling could genuinely disrupt hiring or infrastructure plans.The interesting question isn't whether this is worth the cost - Uber clearly thinks it is or they'd restrict access. It's whether the productivity gains have been measured in a way that's comparable to the spend. Has your company tried to put actual numbers on the AI coding ROI, or is it mostly vibes and velocity estimates?
[link] [comments]