Technology giants such as Microsoft and Uber are currently in the process of reassessing the costs associated with AI utilization. The root cause of this reassessment lies in the significant increase in token consumption, which has not been accompanied by the expected improvements in functionality. Consequently, these companies are reevaluating their strategies for investing in AI. A report from Goldman Sachs forecasts that by 2030, the deployment of agent AI will propel token consumption to surge by 24 times, reaching a staggering 120,000 trillion tokens per month. Notably, the token consumption of agent systems is projected to exceed that of a single AI chatbot by over 1,000 times. For instance, a three-person team at OpenAI once racked up token expenses exceeding $1.3 million within a single month. Similarly, Uber depleted its AI budget for 2026 within a mere few months. Despite the fact that over 80% of software engineers are employing agent AI and more than 60% of code is being generated by AI, management has expressed skepticism regarding the value of these investments. They argue that the heightened token usage has not resulted in a proportional increase in consumer-facing features, and that generating more code does not necessarily translate to an enhanced software experience. In light of these concerns, Microsoft has initiated cost-cutting measures by revoking access to Claude Code subscriptions, preparing to transition to its in-house Copilot CLI, and shifting GitHub's Copilot to a token-based billing model.