Uber's CTO Declares the End of the "Tokenmaxxing" Era, Shifting Focus to AI Efficiency
Uber is concluding its intensive period of "tokenmaxxing," a strategy that encouraged widespread adoption of artificial intelligence within the company. This shift, announced by Uber's Chief Technology Officer, Praveen Neppalli Naga, marks a new focus on optimizing AI utilization and managing associated costs more effectively. The company observed a quadrupling in the number of employees engaging with advanced AI tools, coinciding with a notable reduction in the expenses per AI token.
The "tokenmaxxing" phenomenon emerged earlier in 2026, where businesses pushed for maximum AI integration into employee workflows, sometimes linking AI usage to performance evaluations. However, Naga's recent statements indicate that while initial adoption was high, the focus is now transitioning towards intelligent and cost-effective application of these technologies. This change is partly attributed to Uber's success in lowering AI-related expenditures through advancements in prompt caching, the selection of more suitable default AI models, providing engineers with clearer insights into their AI consumption, and exploring open-source models.
The financial implications of this strategic pivot were also highlighted by Uber's finance chief, Balaji Krishnamurthy, during the company's second-quarter earnings call. He emphasized that despite being in the early stages, the company is already witnessing cost-efficient improvements in developer productivity, including a doubling in code output for engineers. This suggests that the initial investments in AI are beginning to yield tangible benefits through optimized implementation.
Uber's journey through the "tokenmaxxing" era gained attention earlier in the year when Naga disclosed that the company had exceeded its 2026 budget for Anthropic's Claude Code. He also shared in a LinkedIn post from March that their internal coding agent was responsible for approximately 1,800 code changes weekly. However, concerns regarding the return on investment for escalating AI costs were raised in May by Uber COO Andrew Macdonald, who noted that proportional productivity gains were not always evident.
This challenge is not unique to Uber, as the broader technology sector grapples with the complexities of achieving favorable returns on significant AI investments. Companies like Coinbase are exploring solutions such as model switching, where intricate tasks are routed to advanced models while simpler, repetitive operations are handled by more economical ones. Furthermore, the increasing need for AI cost-efficiency has spurred the growth of new businesses specializing in helping companies manage their AI expenditures through consultancy services and the development of cost-effective inference infrastructure.
In essence, Uber's move away from simply maximizing AI token usage towards a more judicious and efficient deployment strategy mirrors a growing maturity in the enterprise AI landscape. The initial phase of widespread adoption is giving way to a more refined approach, where the emphasis is on smart resource allocation and demonstrable value generation from artificial intelligence technologies. This strategic evolution is poised to redefine how businesses integrate and leverage AI for sustainable growth and innovation.