Uber Ends Tokenmaxxing Strategy to Prioritize AI Efficiency

Uber Technologies Inc. is ending its tokenmaxxing era, a strategy implemented in early 2026 that incentivized employees to maximize the use of frontier AI tools like Anthropic's Claude Code. The company previously used leaderboards to rank software engineers by their AI usage, a practice that led Uber to exhaust its 2026 AI budget within the first few months of the year.

Chief Technology Officer Praveen Neppalli Naga announced a shift toward treating AI efficiency as an engineering problem. To stabilize spending, Uber implemented improved prompt caching, optimized default model settings, and integrated open-weight models. CFO Balaji Krishnamurthy reported that these measures decreased the cost per token, allowing the company to maintain stable spending even as the number of employees using AI tools quadrupled since January.

CEO Dara Khosrowshahi stated the company is pursuing thousands of small ecosystem improvements rather than single large breakthroughs. Current applications include rider destination suggestions and a grocery Cart Assistant that has doubled average cart sizes. Internally, the company laid off approximately 10% of its customer service staff last month to capture efficiency gains.

Despite these operational shifts, leadership has noted challenges in measuring ROI. COO Andrew Macdonald observed in May that the company has yet to establish a clear link between AI investment and a measurable increase in the production of useful consumer features. Additionally, Khosrowshahi noted that robotaxis require more extensive real-world testing and government approvals than digital AI chatbots.