Uber finds efficient AI deployment strategy after budget issues

In the United States, Uber's Chief Technology Officer Praveen Neppalli Naga revealed that the company had to reassess its AI spending strategy after a significant budget was exhausted in a short period. This reassessment followed a push for employees to utilize AI tools, particularly Anthropic's Claude Code, which included creating leaderboards to encourage usage among software engineers. However, the trend of 'tokenmaxxing,' where companies incentivize AI use, led to many organizations, including Uber, realizing that the returns on investment were not justifying the rapid spending. As a result, Uber has shifted its approach to focus on efficiency rather than merely increasing usage. Naga noted that the company has successfully quadrupled the number of employees using frontier AI tools, which has contributed to a decrease in the cost per token. This was achieved through improvements in prompt caching, adjustments to default model settings, and evaluating new models for efficiency. Despite the increased adoption of AI, Naga emphasized that costs have not risen as expected, indicating a positive trend in AI cost management. However, he also warned of the potential risk of Jevons paradox, where increased efficiency could lead to higher overall spending on AI resources, despite lower unit costs. This phenomenon has been observed in the broader market, where spending on large language models has doubled even as token prices have significantly decreased. Uber's leadership continues to explore ways to unlock the innovation promised by AI, although the direct link between AI investments and tangible productivity gains remains elusive.