April 17, 2026 Uber says its spending on AI coding tools has already exceeded internal forecasts as adoption accelerates across its engineering teams. The company’s CTO said the scale of usage has outpaced planning, forcing a reset on how AI budgets are calculated.
Speaking to The Information, Uber CTO Praveen Neppalli Naga said: “I’m back to the drawing board, because the budget I thought I would need is blown away already.”
Internally, software development at Uber is shifting toward what Naga described as “agentic software engineering,” where AI systems generate code with minimal human input. The change is already measurable: about 1,800 code changes each week are written entirely by Uber’s internal AI agent, and nearly 70 per cent of committed code is now AI-generated.
Adoption across the workforce is equally broad. Around 95 per cent of Uber engineers use AI tools monthly, while the company’s internal coding agent has scaled from contributing less than 1 per cent of code changes to roughly 8 per cent in just a few months. The speed of that ramp-up is central to the cost pressure now emerging.
A significant driver of those costs is how AI tools are priced. Anthropic, whose Claude Code product is used by Uber, has shifted enterprise pricing toward a hybrid model combining per-seat fees with required usage commitments. Companies must estimate token consumption in advance and pay for that capacity, regardless of actual usage.
This model changes how engineering budgets are managed. Instead of tracking licences, teams must forecast compute usage — a moving target when adoption is still growing rapidly. The removal of earlier enterprise discounts has further increased cost sensitivity for large deployments.
The surge in usage has also introduced a new internal dynamic sometimes referred to as “tokenmaxxing,” where organisations track and compare how much AI compute engineers consume. Reports suggest some companies are even ranking employees based on token usage, turning it into a performance signal.
That shift raises a different concern regarding how higher usage does not necessarily correlate with better outcomes. As AI becomes embedded in development workflows, companies are now balancing productivity gains against a cost structure that scales with every prompt, query, and generated line of code.
